Peak-Constrained Energy Management of PV–BESS–EV Communities : A Reproducible Linear-Programming Benchmark

1. Introduction

The transition toward distributed renewable generation is changing the operating conditions of low-voltage networks. Residential PV systems can reduce grid demand during daylight hours, but their production is intermittent and often misaligned with evening consumption. Battery storage can shift electricity across time, while EVs introduce additional flexible demand and, in some configurations, a potential source of distributed storage. The resulting problem is not limited to maximum-power-point tracking or converter control; it also requires decisions about how multiple assets should be coordinated under tariffs, technical limits, and uncertainty.

The user’s previous research provides a relevant foundation for this direction. Fergani and co-authors studied PV pumping with a permanent-magnet synchronous motor and compared PSO, GWO, and CSA-based MPPT strategies 4. Other work examined a PSO-tuned ANN for maximum-power extraction from a DC microgrid under changing irradiance 3. The user’s group has also investigated efficient EV battery charging in a DC microgrid in Ain El Ibel, Algeria 2, as well as advanced PV and wind control strategies. Those studies focus primarily on energy harvesting, control, and device-level optimization. The present paper complements them by considering community-level coordination between PV generation, flexible storage, and grid exchange. The broader publication record also covers ANFIS control of grid-connected PV inverters 6, hybrid whale-optimization and cuckoo-search MPPT for PMSG wind systems 7, modified bacterial-foraging MPPT under partial shading 8, quantum marine-predator optimization for PV 9, advanced PMSG control 10, adaptive neural-network wind-energy control 12, and a second formulation of the hybrid whale–cuckoo-search wind MPPT method 16. The record further includes PV hotspot diagnosis using RT-DETR and YOLOv8 5, secure partial-shading detection under adversarial attacks 13, a Kitsune optimizer 14, FOPI–ANN PV extraction for the Ain El Ibel case 15, and a broader metaheuristic PV-control study 17. These works are cited here to connect the present scheduling benchmark with the author’s wider research program; they are not treated as direct methodological inputs to the present LP model.

The objective of this preliminary study is to establish a technically transparent benchmark that can be extended into a full robust optimization paper. The central research question is:

Can coordinated battery dispatch reduce the cost of operating a PV–EV energy community and increase PV self-consumption without increasing the community’s unmanaged feeder peak?

The main contributions are fourfold. First, an open energy-community dataset is converted into a reproducible community-scale optimization instance. Second, a linear-programming formulation is given for tariff-aware battery dispatch with a terminal state-of-charge condition. Third, a grid-import cap is introduced so that the economic comparison does not hide a worsening feeder peak. Fourth, the paper reports a complete baseline and optimized benchmark, together with limitations and a roadmap for stochastic and EV-aware extensions.

2. Related Work and Research Gap

Energy-management research has addressed PV self-consumption, battery scheduling, demand response, energy communities, and EV charging through mathematical programming, metaheuristic optimization, model predictive control, and machine learning. The appropriate method depends on the decision horizon, uncertainty, asset coupling, and whether the goal is real-time control or day-ahead scheduling. Linear programming is attractive for an initial benchmark because it is transparent, globally optimal for the stated linear model, and easy to reproduce across datasets.

At the device level, PV controllers seek to extract the maximum available power despite changing irradiance and partial shading. The user’s PSO-tuned ANN study 3 and comparative PV-pumping study 4 are examples of this control-oriented perspective. At the microgrid level, the problem includes energy balance, battery charging, and EV operation. The user’s DC-microgrid and EV-charging publication 2 provides a direct application context for the present community-scale formulation.

The open dataset used in this work was created specifically to support energy-community simulation. It combines household consumption and PV production profiles with BESS, EV, charging-station, tariff, and contractual-power information 1. Its structure makes it suitable for a reproducible benchmark, but it also imposes a methodological responsibility: conclusions must distinguish what is directly measured from what is generated or allocated within the dataset construction. Reviews of microgrid energy management emphasize that uncertainty in PV, load, EV availability, and battery behavior should be represented explicitly when moving from deterministic benchmarks to operational strategies 21–24. Stochastic optimization and predictive-control formulations are therefore natural next steps for this study 22,23,25.

The research gap addressed here is therefore methodological rather than a claim of a new optimization algorithm. Many studies report sophisticated controllers or metaheuristics, but reproducibility can be limited when source data, constraints, or parameter choices are not released together. This work provides a baseline against which future robust optimization, model predictive control, and metaheuristic methods can be compared. The resulting benchmark is especially relevant to the user’s research trajectory because it links PV extraction and EV charging studies to system-level coordination.

3. Materials and Methods

3.1. Dataset and Study Horizon

The study uses the Zenodo record A complete energy community dataset with photovoltaic generation, battery energy storage systems and electric vehicles 1. The dataset describes a European energy community with 250 households. According to the dataset documentation, 200 community members are assigned PV generation and 150 are assigned a BESS. The workbook includes a representative 96-period day with 15-minute resolution, corresponding to 24 hours because each period is 0.25 hours and the horizon contains 96 periods.

Table 1. Extracted study parameters and results. 

Dataset element	Value	Interpretation
Community size	250 households	Aggregate residential community
Time resolution	15 min	96 periods in the supplied day
Aggregate load	6135.30 kWh	Sum of the Load table over the day
Aggregate PV	2390.86 kWh	Sum of the PV table over the day
BESS capacity	1587.00 kWh	Sum of positive household capacities
Maximum charge/discharge	769.40 kW	Aggregate power limit
Weighted efficiency	0.9612	Capacity-weighted value from BESS table
Initial and terminal SOC	793.50 kWh	Terminal equality avoids end effects

Table 1. Extracted study parameters and results. 

Dataset element	Value	Interpretation
Community size	250 households	Aggregate residential community
Time resolution	15 min	96 periods in the supplied day
Aggregate load	6135.30 kWh	Sum of the Load table over the day
Aggregate PV	2390.86 kWh	Sum of the PV table over the day
BESS capacity	1587.00 kWh	Sum of positive household capacities
Maximum charge/discharge	769.40 kW	Aggregate power limit
Weighted efficiency	0.9612	Capacity-weighted value from BESS table
Initial and terminal SOC	793.50 kWh	Terminal equality avoids end effects

The household profiles are aggregated for this first benchmark. This choice makes the optimization small, transparent, and suitable for independent replication. It also means that the model does not yet preserve every household’s individual SOC, inverter, or contractual-power constraint. Those constraints are addressed as future work rather than being silently approximated as household-level behavior.

3.2. Baseline Operation

The unmanaged baseline directly consumes PV when it is available. For each period, positive residual demand is imported from the grid and positive PV surplus is exported. No battery dispatch is allowed. If L_t is aggregate load and P_t^("PV") is aggregate PV generation, net load is

n_t = L_t - P_t^("PV").

The baseline imports g_t^0 = max(n_t,0) and exports e_t^0 = max(-n_t,0). Its energy cost is

C_0 = sum_(t=1)^T Delta t (p_t^b g_t^0 - p_t^s e_t^0)

where p_t^b and p_t^s are the dataset purchase and sale prices, respectively, and Delta t=0.25 h.

3.3. Linear-Programming Dispatch Model

The optimized model uses five decision variables in each period: charge power c_t, discharge power d_t, state of charge s_t, grid import g_t, and grid export e_t. The community power balance is

g_t - e_t + c_t - d_t = n_t.

The SOC dynamics are represented by

s_t = s_(t-1) + eta c_t Delta t - (d_t Delta t)/eta

where eta=0.9612 is the capacity-weighted efficiency parameter. The bounds are

0 <= s_t <= S^("max"), quad 0 <= c_t <= C^("max"), quad 0 <= d_t <= D^("max")

with S^("max")=1587.0 kWh and C^("max")=D^("max")=769.4 kW. The initial condition is s_0=793.5 kWh and the terminal condition is s_T=793.5 kWh.

The objective is

min C = sum_(t=1)^T Delta t (p_t^b g_t - p_t^s e_t + lambda(c_t+d_t))

where lambda=0.01 EUR/kWh is a small throughput penalty used as a transparent proxy for battery wear. The model does not claim that this single coefficient is a validated electrochemical degradation model.

To avoid an economic schedule that increases the feeder peak, the optimized grid-import variable is capped at the unmanaged baseline peak:

0 <= g_t <= P^("cap"), quad P^("cap") = max_t g_t^0 = 453.85 "kW".

The optimization is solved using a standard linear-programming solver. Since the formulation is linear and all variables are continuous, the solution is globally optimal for the stated aggregate model and parameter set.

3.4. Evaluation Metrics

The following metrics are reported: operating cost, imported energy, exported energy, direct PV utilization, peak grid import, peak grid export, and battery throughput. Direct PV utilization is calculated as total PV generation less exported PV energy. Battery throughput is the sum of charging and discharging energy over the horizon. The study does not estimate emissions because an hourly carbon-intensity signal is not included in the selected input tables.

4. Results

4.1. Energy and Economic Performance

The unmanaged baseline has a daily operating cost of EUR 724.15. After coordinated battery dispatch, the modeled cost is EUR 537.22, representing a reduction of 25.81%. Grid imports decline by 742.58 kWh, while export declines from 620.27 kWh to zero under the optimized schedule.

Table 2. Extracted study parameters and results. 

Metric	Baseline	Optimized	Relative change
Operating cost (EUR/day)	724.15	537.22	−25.81%
Grid import (kWh/day)	4364.71	3622.13	−17.01%
Grid export (kWh/day)	620.27	0.00	−100.00%
Direct PV utilization (kWh/day)	1770.58	2390.86	+35.03%
Peak grid import (kW)	453.85	453.85	0.00%
Peak grid export (kW)	296.31	0.00	−100.00%
Battery throughput (kWh/day)	0.00	3094.60	Not applicable

Table 2. Extracted study parameters and results. 

Metric	Baseline	Optimized	Relative change
Operating cost (EUR/day)	724.15	537.22	−25.81%
Grid import (kWh/day)	4364.71	3622.13	−17.01%
Grid export (kWh/day)	620.27	0.00	−100.00%
Direct PV utilization (kWh/day)	1770.58	2390.86	+35.03%
Peak grid import (kW)	453.85	453.85	0.00%
Peak grid export (kW)	296.31	0.00	−100.00%
Battery throughput (kWh/day)	0.00	3094.60	Not applicable

The increase in direct PV utilization is the most direct operational effect of storage. In the baseline, PV surplus is exported whenever it exceeds instantaneous household demand. In the optimized schedule, the battery absorbs sufficient surplus to eliminate modeled export while maintaining the terminal SOC condition. The result is consistent with the role of storage as a temporal-shifting resource rather than as a generator.

4.2. Peak Constraint and Dispatch Behavior

The peak-import constraint is binding as a policy safeguard rather than as a claim that the battery necessarily reduces peak demand in every tariff setting. The baseline and optimized peak are both 453.85 kW. Without the cap, a purely cost-driven formulation can charge the battery during inexpensive periods and temporarily increase instantaneous grid import. The cap prevents that behavior and makes the economic comparison compatible with a feeder that cannot accept a higher maximum import.

Figure 1. Aggregate load, PV generation, grid exchange, and optimized battery state of charge. 

Figure 1. Aggregate load, PV generation, grid exchange, and optimized battery state of charge. 

The optimized schedule produces 3094.60 kWh of battery throughput over the representative day. This value is material relative to the aggregate capacity and should be examined in future work with a more rigorous degradation model. The result also demonstrates why reporting only energy cost would be insufficient: cost, peak import, PV self-consumption, export, and cycling must be interpreted together.

4.3. Sensitivity Interpretation

The current result depends on the dataset’s purchase and sale prices, battery efficiency, throughput penalty, and terminal SOC. The difference between purchase and sale prices creates an incentive to avoid exporting PV and to shift energy across tariff periods. The terminal SOC condition prevents the optimizer from treating the initial battery inventory as free energy. The throughput penalty discourages economically artificial cycling but is not intended to calibrate battery lifetime.

For a full journal-ready revision, the next sensitivity analysis should vary the throughput penalty, terminal SOC requirement, export remuneration, grid cap, and forecast error. The most informative extension would compare deterministic dispatch with scenario-based dispatch under multiple PV and load trajectories.

5. Discussion

The simulation provides three main insights. First, the selected open dataset is sufficiently rich to support a reproducible energy-community benchmark rather than a purely synthetic case. It combines load, PV, BESS, EV, tariff, and contractual-power information in a common workbook 1. Second, the community-level LP provides a clean reference point for more complex controllers and metaheuristics. A new method should be evaluated against this benchmark rather than only against a hand-designed rule. Third, a peak constraint changes the interpretation of cost savings: storage can reduce energy expenditure and PV export without being permitted to increase the maximum feeder import.

The relation to the user’s prior work is direct. The user’s MPPT and PV-pumping studies seek to increase harvested PV energy under dynamic operating conditions 4,8,9,15. The present model assumes that the available PV profiles have already been generated or measured and addresses what happens after generation is available to a community. The user’s DC-microgrid EV-charging study 2 and PV-to-EV comparative study 11 provide device and microgrid perspectives, whereas this paper aggregates those assets into a scheduling problem. The user’s work on UAV PID control 18 and the two incomplete profile entries involving broader engineering collaborations 19,20 are included in the bibliography for completeness but are not used to support claims about energy-community optimization. The two levels can be combined in a future hierarchical architecture: a supervisory optimizer schedules community power, while local MPPT and converter controllers track the setpoints.

A possible architecture for the next paper is therefore hierarchical. At the upper level, a robust or model-predictive scheduler would determine battery and EV charging trajectories using forecast distributions. At the lower level, local PV controllers and converters would track those trajectories subject to electrical constraints. This would allow the user’s established MPPT expertise to remain central while broadening the scientific contribution to system-level energy management.

6. Limitations

This study has important limitations. The evaluation covers one representative 24-hour profile rather than a year of operation. The battery is aggregated, so individual household allocation and network constraints are not preserved. EV charging is not yet optimized, despite EV-related fields being available in the dataset. The model uses a linear efficiency representation and a scalar throughput penalty rather than a validated battery-aging model. The tariff is taken from the dataset and is not treated as a stochastic or market-clearing variable. No power-flow model is included, so voltage, line loading, and phase imbalance are outside the scope of the current benchmark. Finally, because no carbon-intensity time series is used, the paper does not claim a reduction in operational emissions.

These limitations define the next experimental stage rather than invalidating the benchmark. A full extension should preserve household identity, add EV departure and arrival constraints, use rolling-horizon forecasts, test multiple days and seasons, and report confidence intervals across scenarios. It should also compare LP, mixed-integer formulations, model predictive control, and the metaheuristic approaches relevant to the user’s prior publications.

7. Conclusions

This paper presented a reproducible linear-programming benchmark for tariff-aware, peak-constrained battery dispatch in a PV–BESS–EV energy community. Using an open 250-household dataset, the optimized schedule reduced modeled daily operating cost by 25.81%, reduced grid imports by 17.01%, increased direct PV utilization by 35.03%, eliminated modeled export, and maintained the unmanaged peak-import limit of 453.85 kW.

The contribution is best understood as a baseline for a stronger preprint, not as a final universal performance claim. Its value lies in connecting the user’s prior work on PV MPPT, PV pumping, DC microgrids, and EV charging to a transparent community-scale optimization problem. The recommended next version is a robust, EV-aware, multi-day formulation with household-level constraints and forecast uncertainty. That extension would provide a stronger novelty claim and a more defensible basis for journal submission.