As the global energy landscape undergoes profound transformation, wind power has emerged as a cornerstone of renewable energy development. The international community has reached a broad consensus on the urgent need to reduce carbon emissions and transition toward low-carbon economies. According to the International Energy Agency, renewable power generation is projected to increase tenfold by 2050, with wind energy playing an increasingly critical role in this expansion. In 2024, the total installed wind power capacity in China reached 520.68 GW, representing a remarkable growth trajectory from 130.75 GW in 2015. This exponential growth reflects not only technological advancements but also strong policy support for clean energy transitions.
| Year | Total Installed Capacity (GW) | Annual Growth Rate (%) |
|---|---|---|
| 2015 | 130.75 | – |
| 2017 | 163.67 | 25.2 |
| 2019 | 209.15 | 27.8 |
| 2021 | 328.48 | 57.1 |
| 2023 | 441.34 | 34.4 |
| 2024 | 520.68 | 18.0 |
However, the inherent intermittency and stochastic nature of wind speed present significant technical challenges for grid integration. Wind power fluctuations can severely impact voltage stability, frequency regulation, and overall power system reliability. The random variations in wind output, particularly during rapid wind speed changes or gust events, can cause substantial deviations from scheduled power delivery, potentially triggering protection mechanisms and even leading to system instability. While curtailing wind farm output provides a short-term solution, this approach neither supports large-scale renewable energy integration nor optimizes wind resource utilization. Therefore, developing effective strategies to smooth wind power fluctuations has become a critical research priority for ensuring grid security and stability. Among the various technological solutions, the energy storage system has demonstrated exceptional potential due to its ability to dynamically regulate power output, providing both rapid response capabilities and flexible energy management.
This thesis proposes a comprehensive hybrid energy storage system (HESS) configuration scheme that integrates energy-type and power-type storage technologies, coupled with optimized control strategies to enhance both wind power stability and storage system economics. The energy storage system serves as the core enabling technology, providing the necessary flexibility to accommodate renewable energy variability while maintaining grid reliability.
Wind Power System Modeling and Characteristics
The fundamental principle of wind power generation involves the conversion of kinetic energy from wind into electrical energy through a series of sophisticated mechanical and electrical processes. The wind turbine captures wind energy through its blades, which drives the rotor and transmits mechanical power through the drivetrain to the generator. The power output characteristics of a wind turbine can be mathematically expressed through the aerodynamic power equation:
$$P_{wind} = \frac{1}{2} \rho \pi r^2 v^3 C_P(\beta, \lambda)$$
where $\rho$ represents the air density, $r$ denotes the rotor radius, $v$ is the wind speed, and $C_P$ is the power coefficient that depends on the pitch angle $\beta$ and tip speed ratio $\lambda$. The power coefficient, which represents the efficiency of wind energy capture, reaches its theoretical maximum at the Betz limit of 0.593, though practical turbines typically achieve 0.35-0.45. The tip speed ratio is defined as:
$$\lambda = \frac{\omega r}{v}$$
where $\omega$ is the angular velocity of the rotor. Modern wind turbines employ sophisticated maximum power point tracking algorithms to maintain the optimal tip speed ratio, thereby maximizing energy capture under varying wind conditions. The output power characteristics exhibit distinct operational regions: a cut-in region where generation begins, a maximum power tracking region, a rated power region where pitch control limits output, and a cut-out region where the turbine shuts down for safety.
The stochastic nature of wind speed results in highly variable power output that can be characterized by its statistical properties. The short-term fluctuations within minutes and the diurnal variations across hours present different challenges for grid integration. Within the context of this research, the 1-minute and 10-minute fluctuation rates serve as key metrics for evaluating wind power smoothing requirements.
Energy Storage System Technologies and HESS Topology
The selection and configuration of the energy storage system components represent critical decisions in hybrid system design. This study examines four distinct storage technologies, each offering unique performance characteristics that complement one another in hybrid configurations. Lithium-ion batteries provide high energy density and excellent round-trip efficiency, making them ideal for energy-intensive applications. The electrochemical principles governing lithium-ion battery operation involve the reversible intercalation and deintercalation of lithium ions between anode and cathode materials during charge and discharge cycles.
The equivalent circuit model for the lithium-ion battery can be expressed as:
$$\begin{cases} U_o = U_{oc} – U_1 – U_2 – IR \\ I_1 = C_1 \frac{dU_1}{dt} + \frac{U_1}{R_1} \\ I_2 = C_2 \frac{dU_2}{dt} + \frac{U_2}{R_2} \end{cases}$$
where $U_{oc}$ represents the open-circuit voltage, $R$ is the internal resistance, and the RC networks model the dynamic voltage behavior during transient responses. Vanadium redox flow batteries offer exceptional scalability and long cycle life, making them suitable for large-scale stationary storage applications where footprint constraints are less critical. The electrochemical reactions in the vanadium flow battery involve different oxidation states of vanadium ions:
$$\begin{cases} VO^{2+} + H_2O \rightarrow VO_2^+ + 2H^+ + e^- \\ V^{3+} + e^- \rightarrow V^{2+} \end{cases}$$
Supercapacitors deliver ultra-fast response and exceptional power density through electrostatic charge storage at the electrode-electrolyte interface. The energy storage mechanism involves the formation of an electric double layer without chemical reactions, enabling millions of charge-discharge cycles. Flywheel energy storage converts electrical energy to kinetic energy through a rotating mass, providing high-power capability with excellent cycle life.
| Parameter | Lithium-ion Battery | Flow Battery | Supercapacitor | Flywheel |
|---|---|---|---|---|
| Power density (W/kg) | 250-2000 | 50-150 | 5000-15000 | 1000-5000 |
| Energy density (Wh/kg) | 100-250 | 20-60 | 5-15 | 5-50 |
| Response time | 10-100 ms | 100-500 ms | 0.1-5 ms | 1-10 ms |
| Cycle life | 3000-10000 | 10000+ | 500000+ | 1000000+ |
| Efficiency (%) | 85-95 | 75-85 | 90-98 | 85-95 |
| Self-discharge | 2-5%/month | 0.5-1%/day | 5-20%/day | 100%/day |
The HESS topology significantly influences operational flexibility and system performance. In this research, I selected the fully active configuration where each storage technology connects to the DC bus through its dedicated bidirectional DC/DC converter. This architecture enables independent control of charge/discharge power for both energy-type and power-type storage components, maximizing capacity utilization and operational efficiency. The power balance relationship in the wind-storage system can be expressed as:
$$P_{grid}(t) = P_w(t) – P_{hess}(t)$$
$$P_{hess}(t) = P_{en}(t) + P_{pn}(t)$$
where $P_{grid}$ represents the grid-connected power, $P_w$ is the wind farm output, and $P_{en}$ and $P_{pn}$ denote the power contributions from energy-type and power-type storage components respectively.
Wind Power Fluctuation Smoothing Strategy
Wind power fluctuations must be controlled within specified limits to ensure grid stability. The Chinese national standard imposes maximum allowable power change rates based on wind farm capacity. For the 50 MW wind farm studied in this thesis, the 1-minute fluctuation limit is 5 MW and the 10-minute limit is 16.67 MW. The fluctuation rate calculation is defined as:
$$\lambda = \frac{P_{max} – P_{min}}{P_{installed}} \times 100\%$$
To achieve the required smoothing performance, I developed a weighted filtering algorithm combining the recursive moving average and exponential smoothing methods. The recursive moving average operates by calculating the arithmetic mean of data points within a sliding window:
$$P_{sli}(t) = \frac{P_w(t-L/2+1) + P_w(t-L/2+2) + \cdots + P_w(t+L/2)}{L}$$
Exponential smoothing assigns exponentially decreasing weights to historical observations, emphasizing recent data while retaining information from past measurements:
$$S_n = \alpha Y_n + (1-\alpha) S_{n-1}$$
The weighted filtering method dynamically adjusts the balance between these two approaches based on the instantaneous volatility of wind power:
$$P_{grid}(t) = c \cdot S_n(t) + (1-c) \cdot P_{sli}(t)$$
where the adaptive weight $c$ is determined through a sigmoid function of the standard deviation:
$$c = \frac{1}{1 + e^{-a(\sigma(t) – b)}}$$
This adaptive mechanism ensures rapid response during high-volatility periods while maintaining smooth output during relatively stable conditions. The standard deviation of wind power output is calculated as:
$$\sigma(t) = \sqrt{\frac{1}{n} \sum_{t=1}^{n} (P_w(t) – \bar{P}_w)^2}$$
| Method | 1-min max (MW) | 1-min rate (%) | 10-min max (MW) | 10-min rate (%) |
|---|---|---|---|---|
| Raw data | 15.525 | 31.05 | 19.050 | 38.10 |
| Moving average | 4.963 | 9.93 | 15.375 | 30.75 |
| Weighted filtering | 4.844 | 9.69 | 15.400 | 30.80 |
The simulation results demonstrate that both methods satisfy the grid connection requirements. However, the weighted filtering approach exhibits significantly reduced time lag compared to the conventional moving average method, enabling more responsive power regulation. The reduced latency proves particularly valuable during rapid wind speed changes, where timely response prevents excessive deviation from scheduled grid power.
PSO-VMD-Based Power Distribution Method
After obtaining the hybrid energy storage system reference power, the next critical challenge involves allocating this power between different storage technologies based on their frequency response characteristics. Variational Mode Decomposition (VMD) provides a sophisticated signal processing framework for decomposing the power signal into distinct frequency components. The VMD algorithm constructs and solves a constrained variational problem to decompose the original signal into K mode components with specific center frequencies and bandwidths:
$$\min_{\{u_k\},\{\omega_k\}} \left\{ \sum_{k=1}^{K} \left\| \partial_t \left[ \left(\delta(t) + \frac{j}{\pi t}\right) \times u_k(t) \right] e^{-j\omega_k t} \right\|_2^2 \right\}$$
subject to the constraint that the sum of all modes equals the original signal:
$$\sum_{k=1}^{K} u_k(t) = f(t)$$
To solve this constrained optimization problem, the augmented Lagrangian function is constructed:
$$\mathcal{L}(\{u_k\},\{\omega_k\},\lambda) = \alpha \sum_k \left\| \partial_t \left[ \left(\delta(t) + \frac{j}{\pi t}\right) u_k(t) \right] e^{-j\omega_k t} \right\|_2^2 + \left\| f(t) – \sum_k u_k(t) \right\|_2^2 + \left\langle \lambda(t), f(t) – \sum_k u_k(t) \right\rangle$$
where $\alpha$ is the penalty factor and $\lambda(t)$ represents the Lagrange multiplier. The Alternate Direction Method of Multipliers (ADMM) iteratively updates the modes, center frequencies, and Lagrange multipliers until convergence. The frequency-domain update for each mode is:
$$\hat{u}_k^{n+1}(\omega) = \frac{\hat{f}(\omega) – \sum_{i \neq k} \hat{u}_i(\omega) + \frac{\hat{\lambda}(\omega)}{2}}{1 + 2\alpha(\omega – \omega_k)^2}$$
The center frequency update equation is expressed as:
$$\omega_k^{n+1} = \frac{\int_0^{\infty} \omega |\hat{u}_k(\omega)|^2 d\omega}{\int_0^{\infty} |\hat{u}_k(\omega)|^2 d\omega}$$
Despite VMD’s advantages over Empirical Mode Decomposition in mitigating mode mixing, the algorithm’s performance critically depends on the proper selection of parameters: the decomposition layer number K and the penalty factor $\alpha$. In this research, I employ Particle Swarm Optimization to determine these optimal parameters automatically. The PSO algorithm initializes a population of particles, each representing a parameter combination, and iteratively updates their positions and velocities based on both individual and collective best solutions:
$$v_{i,n+1} = \omega v_{i,n} + c_1 \eta_1 (p_{i,n} – x_{i,n}) + c_2 \eta_2 (g_n – x_{i,n})$$
$$x_{i,n+1} = x_{i,n} + v_{i,n+1}$$
The fitness function for parameter optimization uses envelope entropy, which measures signal complexity and periodicity:
$$E_i = -\sum_{j=1}^{N} p_{i,j} \lg(p_{i,j})$$
where $p_{i,j}$ represents the normalized envelope amplitude of the $i$-th IMF component. Lower envelope entropy indicates more pronounced periodic characteristics in the decomposed signal. The optimized parameters for this case study were determined as K = 10 and $\alpha$ = 1930.
Following the VMD decomposition, the Hilbert transform is applied to analyze the frequency characteristics of each mode. The Hilbert transform generates the analytic signal:
$$Z_i(t) = IMF_i(t) + jH[IMF_i(t)] = A_i(t) e^{j\phi_i(t)}$$
where $H[IMF_i(t)]$ denotes the Hilbert transform and $A_i(t)$ represents the instantaneous amplitude. The instantaneous frequency is then derived as:
$$\omega_i(t) = \frac{1}{2\pi} \frac{d\phi_i(t)}{dt}$$
The comparison between EMD and PSO-VMD decompositions reveals significant advantages of the optimized approach. The EMD method exhibits substantial mode mixing phenomena, where adjacent IMFs contain overlapping frequency components, making it difficult to establish a clear frequency boundary for power allocation. The PSO-VMD method, by contrast, produces well-separated modes with minimal frequency overlap, facilitating precise allocation between energy-type and power-type storage devices. Based on the Hilbert transform analysis, I identified IMF7 and higher-frequency components as the boundary between storage types, leading to the following power reconstruction:
$$P_{high}(t) = \sum_{i=1}^{m} IMF_i(t)$$
$$P_{low}(t) = \sum_{i=m+1}^{K} IMF_i(t)$$
The simulation results clearly demonstrate that the high-frequency components, characterized by rapid power changes and relatively low energy content, are allocated to the power-type storage (supercapacitor or flywheel), while the low-frequency components dominated by sustained power trends are assigned to the energy-type storage. This complementary allocation pattern optimally utilizes the inherent characteristics of each storage technology: power-type storage excels at absorbing high-frequency fluctuations due to its rapid response and high cycle life; energy-type storage efficiently manages sustained energy shifts due to its high energy density.
Fuzzy Logic SOC Regulation Strategy
Storage system operation must maintain state-of-charge within safe bounds to prevent overcharge/discharge damage and extend equipment lifetime. The power-type storage devices, despite their excellent power-handling capability, possess relatively limited energy capacity. During sustained high-frequency fluctuation periods, their SOC can rapidly approach limits, necessitating dynamic adjustment of power commands. I developed a fuzzy control strategy to perform real-time SOC monitoring and power correction, ensuring reliable and stable energy storage system operation.
The fuzzy controller operates with two inputs: the current SOC of the power-type storage and its rate of change between consecutive sampling intervals. The output variable represents the power correction coefficient $K_s(t)$. The control rules incorporate operational logic to prevent SOC limit violations:
| SOC(t) | C_SOC | C_SOC | C_SOC | C_SOC | C_SOC |
|---|---|---|---|---|---|
| NL | NS | ZO | PS | PL | |
| VS | PL | VL | ML | MS | S |
| S | PL | PL | VL | ML | VS |
| M | ZO | S | L | VL | VL |
| L | VS | ML | L | VL | VL |
| VL | VS | S | MS | ML | VL |
The fuzzy inference process employs the Mamdani method with triangular and trapezoidal membership functions for input/output variables. The corrected power commands for each storage device are computed as:
$$P’_{en}(t) = P_{en}(t) + (1 – K_s(t)) \cdot P_{high}(t)$$
$$P’_{pn}(t) = K_s(t) \cdot P_{high}(t)$$
Simulation results demonstrate that the fuzzy control strategy maintains both energy-type and power-type storage SOC values within their designated safety ranges throughout the operational period. Without the fuzzy regulation, the power-type storage SOC would breach its upper limit during sustained charging periods or reach critically low levels during extended discharging. The fuzzy controller effectively prevents SOC limit violations while preserving overall smoothing performance, thereby reducing stress on storage components and extending their operational lifetime.
Capacity Optimization Configuration Model
The economic viability of the hybrid energy storage system critically depends on the proper sizing of both power and energy ratings. Undersized configurations fail to achieve required smoothing performance; oversized systems incur excessive investment costs. I developed a bi-level optimization model to determine the optimal capacities that minimize the total lifecycle cost while satisfying technical constraints.
Rated Power and Capacity Calculation
The actual charge/discharge power commands must account for conversion efficiency losses. The adjusted power for the energy-type storage is expressed as:
$$P_{en}(t) = \begin{cases} P_L(t)/\eta_{bd} & P_L(t) < 0 \\ P_L(t) \times \eta_{bc} & P_L(t) > 0 \end{cases}$$
Similarly, for the power-type storage:
$$P_{pn}(t) = \begin{cases} P_H(t)/\eta_{sd} & P_H(t) < 0 \\ P_H(t) \times \eta_{sc} & P_H(t) > 0 \end{cases}$$
The rated power is determined by the maximum absolute power demand:
$$P_{EN} = \max(|P_{en}(t)|)$$
$$P_{PN} = \max(|P_{pn}(t)|)$$
The rated capacity accounts for the cumulative energy exchange and the allowable SOC range:
$$E_{EN} = \frac{\max(E_{en}(t)) – \min(E_{en}(t))}{SOC_{en}^{max} – SOC_{en}^{min}}$$
Battery Life Quantitative Model
Battery degradation significantly affects the economic performance of the energy storage system. This research employs the rainflow counting method to analyze the complete charge-discharge cycles and their corresponding depth-of-discharge (DOD) values. The rainflow counting algorithm processes the SOC time series by rotating it 90 degrees clockwise and identifying half-cycles and full cycles based on the flow path of virtual raindrops along the “roof” surfaces formed by the SOC curve. This method accurately captures both closed hysteresis loops (representing full charge-discharge cycles) and open paths (representing half-cycles consisting of either charging or discharging).
The relationship between cycle life and DOD was established through Gaussian function fitting of experimental data:
$$N(D) = 4.45 \times 10^4 e^{2.685} + 15.16 \times 10^{-10} e^{45.27D}$$
| DOD | 0.1 | 0.2 | 0.3 | 0.4 | 0.5 | 0.6 | 0.7 | 0.8 | 0.9 | 1.0 |
|---|---|---|---|---|---|---|---|---|---|---|
| Cycles | 15000 | 50000 | 30000 | 14000 | 10000 | 8000 | 7500 | 6000 | 5000 | 4000 |
The equivalent cycle count is calculated by normalizing to the lifecycle at full discharge depth:
$$N_{eq} = \sum_{i=1}^{k} \frac{N_c(D_i)}{N_c(D_i)} \times \frac{N_c(1)}{1} = \sum_{i=1}^{k} \frac{N_c(1)}{N_c(D_i)}$$
The battery’s operational lifespan is estimated as:
$$Y = \frac{N_{eq}}{365 \times N_d}$$
where $N_d$ represents the daily equivalent cycles. My analysis reveals an important relationship: battery cycle life correlates positively with rated energy capacity. For a given power command profile, increasing the rated capacity reduces the DOD per cycle, thereby decreasing cumulative damage and extending operational life. This relationship highlights the economic trade-off between higher initial investment in capacity and extended replacement intervals.
Outer-Layer Objective Function
The outer-layer optimization minimizes the total lifecycle cost of the hybrid energy storage system, encompassing initial investment, replacement costs, operation and maintenance expenses, and residual value:
$$\min f = C_{in} + C_{pre} + C_{om} + C_{rec} + C_{sv}$$
The initial investment covers power and capacity costs:
$$C_{in} = k_{pi} P_{HN} + k_{ei} E_{HN}$$
The replacement cost accounts for the present value of future replacements over the planning horizon:
$$C_{pre} = \sum_{n=1}^{N_p} (k_{pp} P_{HN} + k_{ep} E_{HN}) (1+r)^{-nT}$$
The operation and maintenance costs are expressed as:
$$C_{om} = \sum_{t=1}^{T} \frac{(k_{po} P_{HN} + k_{eo} W)(1+r)^t – (1+r)^{-1}}{r(1+r)^T}$$
The constraints include power limits, SOC bounds, and grid fluctuation standards:
$$\begin{cases} |P_{en}(t)| \leq P_{EN} \\ |P_{pn}(t)| \leq P_{PN} \\ SOC_{en}^{min} \leq SOC_{en}(t) \leq SOC_{en}^{max} \\ SOC_{pn}^{min} \leq SOC_{pn}(t) \leq SOC_{pn}^{max} \\ \lambda_1 \leq \lambda_{1S} \\ \lambda_{10} \leq \lambda_{10S} \end{cases}$$
Inner-Layer Objective Function
The inner-layer optimization minimizes the daily scheduling cost to determine the optimal operation strategy for given capacity configurations:
$$\min W_j = \frac{1}{30} \sum_{i=1}^{n} h \cdot P_l(i) \cdot \Delta t$$
where $h$ represents the electricity price and $P_l(i)$ is the purchased power at time $i$. Three typical days with distinct wind and load profiles are selected to represent seasonal variations and reduce uncertainty effects on the optimization results.
Sparrow Search Algorithm
The Sparrow Search Algorithm (SSA) was employed to solve the bi-level optimization problem. Inspired by the foraging and anti-predation behavior of sparrows, this metaheuristic algorithm partitions the population into producers, scroungers, and sentinels. Producers explore for food with high energy reserves, scroungers follow producers to access discovered food sources, and sentinels monitor environmental threats. The producer position update is:
$$X_{i,j}^{t+1} = \begin{cases} X_{i,j}^t \cdot \exp\left(-\frac{i}{\alpha \cdot i_{max}}\right) & R_2 < S_T \\ X_{i,j}^t + Q \cdot L & R_2 \geq S_T \end{cases}$$
Through iterative optimization, the SSA efficiently explores the search space and converges to optimal capacity configurations. The bi-level interaction allows the inner layer to feed back realistic operational costs to the outer layer, ensuring that the final capacity solution reflects both investment economics and operational efficiency.
Case Study Results and Economic Analysis
This section presents the case study results for the 50 MW wind farm using the proposed optimization framework. Three representative days were selected to capture seasonal wind patterns and load variations. The original and smoothed wind power curves for these typical days demonstrate that the weighted filtering approach effectively reduces fluctuations while maintaining good tracking performance of the original power trajectory.
| Configuration | LiB+SC | LiB+FES | VRB+SC | VRB+FES |
|---|---|---|---|---|
| P_EN (MW) | 7.80 | 7.80 | 8.84 | 8.84 |
| E_EN (MWh) | 8.80 | 9.43 | 14.25 | 14.25 |
| P_PN (MW) | 2.40 | 2.40 | 2.40 | 2.40 |
| E_PN (MWh) | 1.70 | 1.84 | 1.70 | 1.84 |
| Annual cost (yuan) | 1.805×10⁷ | 2.033×10⁷ | 1.666×10⁷ | 1.922×10⁷ |
| Configuration | LiB+SC | LiB+FES | VRB+SC | VRB+FES |
|---|---|---|---|---|
| P_EN (MW) | 7.74 | 7.74 | 8.77 | 8.77 |
| E_EN (MWh) | 4.41 | 5.14 | 9.32 | 9.32 |
| P_PN (MW) | 2.98 | 2.95 | 2.98 | 2.95 |
| E_PN (MWh) | 0.46 | 0.39 | 0.46 | 0.39 |
| Annual cost (yuan) | 1.237×10⁷ | 1.380×10⁷ | 1.270×10⁷ | 1.318×10⁷ |
Comparing the results across decomposition methods reveals that PSO-VMD achieves significantly lower annual costs for all four HESS configurations. The EMD method, due to its mode-mixing limitations, requires larger storage capacities, particularly for the energy-type storage component. The superior frequency separation of PSO-VMD enables more precise power allocation, reducing the required battery capacity while increasing the power-type storage allocation slightly. This optimization directly translates to reduced investment requirements.
Among the four configurations studied, the lithium-ion battery and supercapacitor combination achieves the lowest annual cost of 1.237×10⁷ yuan under the PSO-VMD approach. This configuration effectively balances the high energy density requirements of sustained power compensation with the rapid response capability needed for transient fluctuation suppression. The combined system demonstrates excellent technical performance while maintaining economic viability. The flow battery configurations show competitive performance due to their longer projected lifetimes and lower replacement frequency, but their higher initial costs and lower efficiency offset these advantages in the lifecycle economic analysis.
The typical day scheduling results further illustrate the operational benefits of the optimized energy storage system configuration. During low-load periods when wind power exceeds local demand, the energy storage system charges to absorb excess renewable generation, preventing curtailment and storing energy for later use. During peak load periods when wind output is insufficient, the storage system discharges to supplement supply, reducing grid power purchases and improving overall system economics. This operational pattern enhances renewable energy utilization while reducing both operating costs and environmental impact.
The battery capacity versus cycle life analysis demonstrates that increasing the energy rating from the minimum required capacity significantly extends the operational lifespan. At the 10 MW system scale, a 8 MWh battery bank achieves approximately 5200 cycles while a 15 MWh bank extends to over 7500 cycles under the same power profile. This relationship confirms that the capacity optimization must consider not only initial investment but also the lifetime extension benefit that higher capacity provides, thereby achieving improved lifecycle economics.
Conclusion
This thesis presents a comprehensive investigation into wind power smoothing and capacity optimization using hybrid energy storage systems. The main contributions and findings are summarized as follows:
First, a weighted filtering algorithm that combines recursive moving average and exponential smoothing methods effectively reduces wind power fluctuations while minimizing time lag. The simulation results demonstrate that this approach satisfies grid connection standards for both 1-minute and 10-minute fluctuation limits, providing more responsive power regulation than conventional filtering methods.
Second, the PSO-VMD decomposition framework significantly improves the accuracy of power allocation between energy-type and power-type storage devices. The particle swarm optimization automatically determines optimal VMD parameters, eliminating the limitations of manual parameter selection. Compared with EMD, the PSO-VMD method exhibits superior frequency separation and reduced mode mixing, enabling more precise identification of the frequency boundary between storage types. This precision leads to better-optimized capacity configurations with reduced storage requirements and improved economic performance.
Third, the fuzzy control strategy for SOC regulation prevents storage devices from exceeding their safe operating ranges during dynamic operation. By monitoring both the current SOC and its rate of change, the fuzzy controller adjusts power commands in real time to maintain SOC within acceptable limits, thereby protecting storage devices from degradation and extending their useful life.
Fourth, the bi-level optimization model effectively coordinates technical performance and economic objectives. The outer layer minimizes lifecycle costs through optimized capacity selection, while the inner layer determines the minimum daily operation cost for each candidate configuration. The Sparrow Search Algorithm efficiently solves this complex optimization problem, converging to high-quality solutions.
Finally, the comparative analysis of four HESS configurations reveals that the lithium-ion battery and supercapacitor combination achieves the best overall performance, combining technical effectiveness with economic efficiency. This research provides a practical framework for energy storage system deployment in wind farms, supporting the transition toward higher renewable energy penetration while maintaining grid stability and economic viability.
Future research directions include extending the analysis to grid-side and user-side energy storage configurations, investigating multi-objective optimization approaches that incorporate system resilience and power quality metrics, and validating the proposed algorithms with field measurements from operating wind-storage systems. The integration of degradation-aware control strategies and market participation models could further enhance the economic competitiveness of hybrid energy storage systems in evolving electricity markets.