The rapid global deployment of 5G technology has ushered in an era of unprecedented connectivity, but it has also brought the energy management of 5G base stations (BSs) to the forefront of operational challenges. Compared to their predecessors, 5G BSs, supporting higher data rates and broader coverage, exhibit significantly greater energy consumption. Efficiently managing this energy usage and improving overall energy efficiency have thus become critical imperatives. In this context, the coordinated operation of demand response (DR) mechanisms and Battery Energy Storage Systems (BESS) presents a promising solution for optimizing 5G base station operations within modern power grids characterized by increasing renewable energy penetration.
This article proposes a novel two-stage State of Charge (SoC) interval optimization method for 5G base stations, explicitly coordinating communication load migration (a form of information-centric demand response) with physical BESS scheduling. We first develop a comprehensive energy optimization model that integrates the costs associated with communication load transfer and battery degradation. Subsequently, we design a two-phase optimization framework: the first phase determines the optimal user association strategy and establishes robust SoC boundaries for the BESS over a 24-hour horizon, considering forecasted renewable generation and load patterns. The second phase then performs hourly, real-time BESS dispatch within these pre-defined SoC intervals to track the actual realization of uncertainties (like solar irradiance) and compensate for the first-stage decisions. Case studies demonstrate that our proposed SoC interval optimization strategy effectively leverages the flexibility of 5G base station battery energy storage system resources, enabling economical operation under varying communication loads and electricity market conditions.
1. Integrated 5G Base Station Model with Communication Load Migration and BESS
1.1 5G Base Station Power Consumption Model
A 5G base station primarily consists of two key systems: the power supply system and the communication system. The power supply system typically includes the battery energy storage system, rooftop solar photovoltaic (PV) panels, and a connection point to the distribution grid. The communication system comprises Active Antenna Units (AAUs), Baseband Units (BBUs), and network transmission equipment.
The total power consumption of a 5G BS can be categorized into active and sleep modes. In active mode, the power draw has both static and dynamic components, with the dynamic part being proportional to the communication traffic load. The power model is given by:
$$P^{5G} = \begin{cases} P^{Base} + P^{Active}, & \text{Active Mode} \\ P^{Sleep}, & \text{Sleep Mode} \end{cases}$$
where \(P^{Base} = P^{AAU} + P^{BBU} + P^{other}\) represents the fixed power of the BS in active mode, and \(P^{Active} = \epsilon \cdot P^{dy}\) is the load-dependent incremental power. Here, \(\epsilon\) is the energy efficiency coefficient and \(P^{dy}\) is the transmit power increment corresponding to user connections.
1.2 Communication Load Migration Model
Communication demand exhibits significant spatiotemporal variation across different geographical areas (e.g., residential, commercial, industrial districts). We propose a strategy where the 5G network can reconfigure its connections with mobile users based on real-time communication load, PV output, and battery energy storage system status, thereby facilitating the migration of information and data flows. Let \(I = \{1,2,…,i\}\) be the set of BSs, \(J = \{1,2,…,j\}\) be the set of mobile users, and \(T = \{1,2,…,t\}\) be the set of time slots. A binary variable \(c_{i,j,t}\) indicates the connection between BS \(i\) and user \(j\) in slot \(t\).
The constraint ensuring a user is connected to only one BS per slot is:
$$\sum_{i \in S_{j,t}} c_{i,j,t} = 1, \quad \forall j,t$$
where \(S_{j,t}\) is the set of BSs accessible to user \(j\) at time \(t\). The dynamic transmit power for a specific connection and the total dynamic power for a BS are:
$$P^{dy}_{i,j,t} = N_0 B (2^{\frac{D_{j,t}}{B}} – 1) \cdot 10^{\frac{\alpha + \beta \cdot \lg(d_{i,j})}{10}}$$
$$P^{dy}_{i,t} = \sum_{j \in U_i} \alpha_{i,j,t} \cdot P^{dy}_{i,j,t}$$
where \(N_0\) is noise power, \(B\) is bandwidth, \(D_{j,t}\) is user traffic demand, \(\alpha, \beta\) are channel fading coefficients, \(d_{i,j}\) is the distance, and \(U_i\) is the set of users within BS \(i\)’s coverage.
1.3 Battery Energy Storage System Model
The battery energy storage system in a 5G base station must ensure uninterrupted power supply. We model the battery degradation cost, focusing on the impact of Depth of Discharge (DoD). The DoD and cycle life \(L\) are calculated as:
$$DoD_{i,t} = \frac{|P^{B}_{i,t}| \cdot \tau}{E^{cap}_{i}}$$
$$L(DoD) = A \cdot DoD^{-B} \cdot e^{-C \cdot DoD}, \quad A, B, C > 0$$
where \(P^{B}_{i,t}\) is the battery power (positive for discharge), \(\tau\) is the time interval, and \(E^{cap}_{i}\) is the rated capacity. The degradation cost per cycle is:
$$C^{deg}_{i,t} = \frac{C^{B}_{i} \cdot P^{B}_{i,t} \cdot \tau}{2 \cdot L(DoD_{i,t}) \cdot E^{cap}_{i} \cdot (\eta^{ch} + \eta^{dis})}$$
where \(C^{B}_{i}\) is the replacement cost, and \(\eta^{ch}\), \(\eta^{dis}\) are charge/discharge efficiencies.
The operational constraints for the battery energy storage system include:
$$0 \leq \beta^{ch}_{i,t} + \beta^{dis}_{i,t} \leq 1$$
$$P^{B}_{i,t} = \eta^{ch}P^{ch}_{i,t} – \frac{1}{\eta^{dis}}P^{dis}_{i,t}$$
$$0 \leq P^{ch}_{i,t} \leq \beta^{ch}_{i,t}P^{ch,max}_{i}, \quad 0 \leq P^{dis}_{i,t} \leq \beta^{dis}_{i,t}P^{dis,max}_{i}$$
$$E^{B}_{i,t} = E^{B}_{i,t-1} + P^{B}_{i,t} \cdot \tau$$
$$SoC^{B}_{low,i,t} \leq \frac{E^{B}_{i,t}}{E^{cap}_{i}} \leq SoC^{B}_{up,i,t}$$
$$SoC_{min} \leq \frac{E^{B}_{i,t}}{E^{cap}_{i}} \leq SoC_{max}$$
$$E^{B}_{i,t} \geq E^{re,min}_{i,t}, \quad E^{B}_{i,0} = E^{B}_{i,24}$$
Here, \(\beta^{ch/dis}\) are binary charge/discharge status variables, \(P^{ch/dis}\) are the charge/discharge powers, \(E^{B}_{i,t}\) is the stored energy, and \(SoC^{B}_{low/up,i,t}\) are the optimized SoC interval bounds. \(E^{re,min}_{i,t}\) is the minimum backup energy reserve.
2. Two-Stage Optimization Scheduling Framework
2.1 Objective Function
The objective is to minimize the total operational cost of the 5G base station network:
$$\text{min } F^{total} = C^{PV} + C^{BESS} + C^{grid}$$
subject to the constraints (1) to (32) from the model. The cost components are:
$$C^{PV} = \sum_{t \in T} \sum_{i \in I} C^{PVOM}_{i} P^{PV}_{i,t}$$
$$C^{BESS} = \sum_{t \in T} \sum_{i \in I} C^{deg}_{i,t}$$
$$C^{grid} = \sum_{t \in T} \sum_{i \in I} \omega_t P^{grid}_{i,t}$$
where \(C^{PVOM}_{i}\) is the PV O&M cost, \(\omega_t\) is the time-of-use electricity price, and \(P^{grid}_{i,t} = P^{5G}_{i,t} + P^{B}_{i,t} – P^{PV}_{i,t}\) ensures power balance.
2.2 Two-Stage Robust Optimization Model
To handle the uncertainty in PV output \(P^{PV}_{i,t}\), we employ a two-stage robust optimization approach with interval forecasts: \( \underline{P}^{PV}_{i,t} \leq P^{PV}_{i,t} \leq \overline{P}^{PV}_{i,t} \).
Stage 1 (Master Problem): Decisions are made on user association \(x\) (variables \(c_{i,j,t}\)) and the robust SoC interval bounds \(y’\) (related to \(SoC^{B}_{low/up,i,t}\)), considering the worst-case realization of uncertainty from the previous iteration.
$$\min_{x, y’} \left( F(x, y’, u^*) + \max_{u \in \mathcal{U}} F(x, y’, u) \right)$$
Stage 2 (Subproblem): For given first-stage decisions \((x^*, y’^*)\), the subproblem finds the optimistic (\(u^*_{opt}\)) and pessimistic (\(u^*_{pes}\)) realizations of uncertainty (PV output) within the forecast interval \(\mathcal{U}\) that minimize and maximize the cost, respectively.
The algorithm iterates between the master and subproblems until convergence. Following this, the final battery energy storage system dispatch \(y\) (charge/discharge powers) is determined by solving a second-stage problem that minimizes the expected cost across the two extreme scenarios, while adhering to the optimized SoC bounds:
$$\min_{y_1, y_2} \left[ F(x^*, y_1, u^*_{opt}) + F(x^*, y_2, u^*_{pes}) \right]$$
subject to the power balance and BESS constraints for both scenarios. Additional constraints ensure the SoC interval is practical for real-time adjustment:
$$(SoC^{B}_{low,i,t} – SoC^{B}_{low,i,t-1}) \cdot E^{cap}_{i} \geq P^{dis,max}_{i} \cdot \tau$$
$$SoC^{B}_{up,i,t} – SoC^{B}_{low,i,t} \geq \zeta_{i,t}$$
where \(\zeta_{i,t}\) is a design parameter for the minimum interval width, ensuring sufficient flexibility for hourly dispatch.
3. Case Study and Analysis
We evaluate the proposed algorithm on a 2km×2km area divided into four typical regions: Administrative, Industrial, Commercial, and Residential. Each region hosts multiple 5G base stations with distinct PV generation profiles and communication demand patterns.
The scheduling results for the battery energy storage system in each region show distinct behaviors. Base stations in Industrial and Commercial areas, with higher PV output, tend to store excess energy during peak solar hours (before 16:00) for later use during high-price or high-demand periods. Base stations in Administrative and Residential areas, with lower PV generation, primarily charge their BESS during low-price periods and discharge during peak hours. Communication load migration allows shifting traffic from areas with lower PV generation to those with higher generation or putting BSs with low load into sleep mode, enhancing overall network energy efficiency.
To validate the effectiveness, we compare four algorithms:
- Algorithm 1 (Baseline): No load migration, hourly BESS dispatch based on forecast expectation.
- Algorithm 2: Includes communication load migration.
- Algorithm 3: Load migration + two-stage optimization with fixed SoC interval width.
- Algorithm 4 (Proposed): Load migration + two-stage optimization with dynamic SoC interval adjustment.
We perform a Monte Carlo simulation with 100 scenarios representing PV uncertainty. The results are summarized below:
| Algorithm | Average Cost ($) | Cost Range ($) | Avg. BS Power (W) | Avg. Grid Purch. (kWh) |
|---|---|---|---|---|
| Algorithm 1 | 921.3 | [882.8, 962.2] | 1620.4 | 987.3 |
| Algorithm 2 | 868.5 | [843.2, 891.3] | 1438.2 | 925.2 |
| Algorithm 3 | 782.4 | [751.5, 816.8] | 1313.6 | 732.5 |
| Algorithm 4 (Proposed) | 758.1 | [723.7, 791.9] | 1302.7 | 693.6 |
The analysis clearly shows that Algorithm 2 outperforms the baseline by reducing costs through load migration and BS sleep modes. Algorithm 3 provides a further significant cost reduction by robustly optimizing the battery energy storage system schedule against uncertainty. Our proposed Algorithm 4 achieves the best economic performance by dynamically adjusting the SoC interval, providing greater flexibility for the real-time dispatch of the battery energy storage system to track actual conditions, thereby minimizing total operational cost.
4. Conclusion
This article presented a coordinated optimization strategy for 5G base station energy management, integrating communication load migration (as a demand response tool) with the scheduling of an on-site battery energy storage system. The core of the method is a two-stage SoC interval optimization framework that first establishes robust energy reserves considering forecasts and then allows fine-grained, uncertainty-tracking dispatch within those bounds. The case study demonstrates that this approach significantly enhances the economic efficiency and operational flexibility of 5G base stations, providing a viable solution for their integration into future smart grids with high renewable energy penetration.