An intelligent dispatch and management system and method for energy storage power stations based on EMS
The EMS system, through multi-protocol adaptation and hybrid prediction models, solves the problems of rigid scheduling strategies and data silos in energy storage power stations, realizes dynamic power station management, improves profitability and security, and extends battery life.
Patent Information
- Application Number
- CN202611566867.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-10-10
- Publication Date
- 2026-11-17
AI Technical Summary
Existing energy storage power station EMS systems suffer from rigid scheduling strategies, lack of predictive capabilities, data silos, simplistic safety protection, and a lack of closed-loop execution of plans, resulting in low power station revenue, insufficient battery utilization, and safety hazards.
A multi-protocol adapter gateway and edge computing unit are used to achieve unified data collection and cleaning. A hybrid prediction model of LSTM and TCN is used for load and electricity price prediction. A multi-objective optimization scheduling module is constructed. Safety verification is performed based on the SOC-temperature two-dimensional safety plane. Closed-loop optimization is formed through rolling time-domain control and policy self-learning.
It enables dynamic scheduling based on electricity prices, load, and renewable energy output, thereby improving power plant revenue and battery utilization, extending battery life, enhancing scheduling security and robustness, breaking down data silos, and providing unified, high-quality data support.
Smart Images

Figure CN123122020A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of power energy storage technology, specifically to an intelligent dispatch management system and method for energy storage power stations based on EMS. Background Technology
[0002] With the advancement of the "dual carbon" target, the installed capacity of new energy sources, represented by wind power and photovoltaics, has grown rapidly. Due to the randomness, volatility, and intermittency of new energy output, large-scale grid integration places higher demands on the power system's peak-shaving and frequency regulation capabilities. Energy storage power stations, with their flexible charging and discharging characteristics, have become a key means to mitigate new energy fluctuations, participate in grid peak-shaving and frequency regulation, and improve grid operational stability, and have experienced rapid development in recent years.
[0003] The Energy Management System (EMS) is the brain of an energy storage power station, responsible for data acquisition, monitoring, and the formulation and execution of charging and discharging strategies. However, existing energy storage power station EMSs generally suffer from the following problems: First, the scheduling strategy is rigid. Most existing EMS systems use manually set fixed charging and discharging strategies (such as fixed charging and discharging periods and fixed SOC upper and lower limits), which are difficult to adapt to the dynamic changes in electricity price signals, load demand and renewable energy output, resulting in low returns and insufficient battery utilization of energy storage power stations.
[0004] Secondly, there is a lack of predictive capabilities. Most existing EMS systems only have data acquisition and monitoring functions, and do not have the ability to predict load, electricity prices and renewable energy output. They cannot achieve "prediction first, optimized dispatch", and dispatch decisions lag behind the actual changes in the power grid and power plants.
[0005] Third, data silos exist among subsystems. Subsystems such as BMS, PCS, metering, environment, and fire protection are usually provided by different vendors with different communication protocols, resulting in scattered and isolated data. EMS struggles to obtain comprehensive, unified, and high-quality data support, which hinders the realization of advanced analysis functions.
[0006] Fourth, the safety protection strategy is simple. Existing EMS mostly use fixed upper and lower limits of SOC and temperature thresholds for protection, without considering the degradation of battery state of health (SOH), differences in battery consistency, and the coupling effects of multiple factors. The protection threshold is applied in a one-size-fits-all manner, which either sacrifices usable capacity or poses safety hazards; moreover, there is a lack of simulation and verification capabilities for scheduling plans before the plan is executed.
[0007] Fifth, the execution of the plan lacks a closed loop. Once the scheduling plan is issued, it is no longer adjusted. Deviations caused by forecasting errors, equipment malfunctions, etc. during actual operation cannot be corrected in a timely manner, and it also lacks the ability to learn and continuously optimize strategies based on operational results. Summary of the Invention
[0008] To address the shortcomings of the existing technologies, this invention provides an intelligent dispatch and management system and method for energy storage power stations based on EMS.
[0009] The technical solution adopted in this invention is as follows: An intelligent dispatch and management system for energy storage power stations based on EMS includes a data acquisition layer, a data platform layer, an intelligent decision-making layer, an execution layer, and a human-machine interaction layer.
[0010] The data acquisition layer includes a multi-protocol adapter gateway and an edge computing unit. The multi-protocol adapter gateway supports protocols such as Modbus RTU / TCP, IEC 60870-5-104, CAN 2.0, and MQTT, and can connect to BMS, PCS, grid-connected meters, environmental monitoring devices (temperature, humidity, smoke, combustible gas sensors), and fire monitoring devices from different manufacturers, achieving unified protocol adaptation and data normalization. The edge computing unit is deployed locally at the power plant, performing edge-side cleaning (removing outliers and filling in missing values), time alignment, and feature extraction (such as charge / discharge rate, cell temperature difference, voltage range, etc.) on the acquired data, and performing breakpoint resume caching when communication is interrupted.
[0011] The data platform layer includes a real-time database, a time-series database, and a digital twin model library. The real-time database stores millisecond-level telemetry and telecontrol data to support monitoring and control; the time-series database stores historical operating data and alarm events to support statistical analysis and model training; the digital twin model library constructs an electro-thermal coupling model based on the power plant's series and parallel electrical topology, battery cell charge-discharge characteristic curves, and thermal characteristic parameters to provide model support for planned simulation verification.
[0012] The intelligent decision-making layer is the core of this invention, and includes a prediction module, a multi-objective optimization scheduling module, a security assessment and verification module, and a strategy self-learning module.
[0013] The prediction module includes a load prediction unit, an electricity price prediction unit, and an output prediction unit. The load and output prediction units employ a hybrid prediction model composed of LSTM and TCN: LSTM captures long-term dependencies in the sequence, while TCN extracts multi-scale local features through dilated causal convolution; the outputs of both are weighted and fused using an attention mechanism. The electricity price prediction unit uses a gradient boosting tree model, taking historical electricity prices, load prediction results, renewable energy prediction results, and calendar features as input. Different strategies are employed for different time horizons: ultra-short-term predictions (15 minutes to 4 hours ahead, 15-minute resolution) use an LSTM-TCN hybrid model; short-term predictions (4 hours to 72 hours ahead, 15-minute resolution) use a weighted fusion of the hybrid model and the gradient boosting tree model. The prediction module simultaneously outputs the predicted value and prediction range through quantile regression.
[0014] The multi-objective optimization scheduling module aims to maximize operational revenue, minimize battery degradation cost, and optimize grid-friendliness within the scheduling cycle. It constructs the following multi-objective optimization model: the objective function is min F = w1(-C_rev) + w2C_deg + w3P_dev, where C_rev is the sum of peak-shaving revenue and electricity price arbitrage revenue, calculated based on electricity price and charging / discharging power for each time period; C_deg is the battery degradation cost, calculated using the rainflow counting method to count the equivalent number of cycles, combined with an empirical battery degradation model (considering the nonlinear impact of depth of discharge (DOD) on cycle life); and P_dev is a grid-friendliness penalty term, penalizing the deviation between actual output and scheduled output. Constraints include: dynamic SOC constraints (energy balance relationship between SOC and charging / discharging power in adjacent time periods), SOC boundary constraints, upper and lower limits of PCS charging / discharging power constraints, charging / discharging mutual exclusion constraints (simultaneous charging / discharging is not allowed in the same time period), grid connection point capacity constraints, and planned power constraints. The model is solved using a mixed-integer linear programming solver, generating charging / discharging plan curves for the next day or the current day. The weighting coefficients w1, w2, and w3 can be configured according to operational strategies, such as guaranteed revenue mode, guaranteed lifespan mode, and guaranteed performance mode.
[0015] The safety assessment and verification module constructs a two-dimensional safety plane using SOC and cell temperature, and dynamically divides the system into three zones, taking into account SOH and battery consistency indicators: normal scheduling zone (operating point is within the SOC safety range and temperature is normal, operating at full power as planned), power-limited scheduling zone (operating point is close to the SOC boundary or temperature is too high, linearly limiting charging and discharging power according to a preset function and issuing an alarm), and emergency control zone (operating point exceeds the safety boundary or temperature exceeds the limit, executing step-by-step power reduction, shutdown, and fire alarm linkage). Before the plan is issued, the verification module calls the electro-thermal coupling model in the digital twin model library to simulate the plan execution process. If the simulation finds that the operating point falls into the power-limited scheduling zone at any time, the power for that period is corrected and the simulation is re-performed; if it falls into the emergency control zone, the power for that period is set to zero and an alarm is marked, and the system can only be rearranged after manual confirmation.
[0016] The execution layer employs rolling time-domain control (MPC): each control cycle (e.g., 15 minutes) optimizes the charging and discharging plan for the next 24 hours starting from the current moment, executing only the plan for the first time period; the next control cycle uses the latest measured data and updated prediction results to re-solve the plan, thus continuously absorbing the latest information to correct the plan and effectively suppressing the accumulation of prediction errors. The execution layer also has an AGC / AVC interface, which can receive grid dispatch commands and convert them into energy storage charging and discharging commands to realize ancillary services such as peak shaving and frequency regulation.
[0017] The strategy self-learning module uses the plan execution deviation rate, prediction accuracy (MAPE), and daily revenue per unit capacity as evaluation indicators to generate daily / weekly / monthly operation evaluation reports. When the indicators are lower than the preset threshold, the prediction model is automatically retrained (by incrementally updating the model parameters using the latest time-series data) and the target weight is adjusted. At the same time, the SOH decay data is fed back to the multi-objective optimization scheduling module to dynamically correct the battery decay cost coefficient, forming a closed loop of "operation-evaluation-learning-rescheduling".
[0018] The human-machine interaction layer includes a monitoring screen, a dispatch workstation, a mobile APP, and a web browser. It graphically displays the power plant overview, battery cluster status, SOC / SOH curves, dispatch plan curves, revenue statistics, and alarm information. It supports manual adjustment of plans, mode switching (automatic / semi-automatic / manual), and access control.
[0019] Compared with the prior art, the beneficial effects of the present invention are as follows: (1) The scheduling architecture of "prediction + optimization" is adopted. Based on the multi-time scale prediction results of electricity price, load and new energy output, the charging and discharging plan is generated, which can significantly improve the power station revenue and battery utilization rate compared with the fixed strategy. (2) Explicitly incorporate battery degradation costs into the optimization objective and combine them with SOH feedback for dynamic correction to achieve a balance between revenue and lifespan and extend battery lifespan; (3) Based on the two-dimensional safety plane of SOC-temperature and the dynamic partition verification mechanism that integrates SOH and consistency index, the available capacity of the battery is fully utilized under the premise of ensuring safety, and the capacity waste caused by one-size-fits-all protection is avoided. (4) Before the plan is issued, it is simulated and verified by a digital twin model. During the execution of the plan, it is corrected in real time by rolling time domain, forming a closed loop of the whole process of "prediction-optimization-verification-execution-evaluation-learning", which significantly improves the scheduling security and robustness; (5) Multi-protocol adapter gateways connect data silos in subsystems such as BMS, PCS, metering, environment and fire protection, providing a unified high-quality data foundation for intelligent decision-making. Attached Figure Description
[0020] Figure 1 This is a diagram showing the overall architecture of the system of the present invention; Figure 2 This is a flowchart of the intelligent scheduling and management method of the present invention. Detailed Implementation
[0021] The technical solution of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.
[0022] Example 1: System Deployment.
[0023] A 100MW / 200MWh lithium iron phosphate energy storage power station is equipped with an EMS system based on this invention. The multi-protocol adapter gateway of the data acquisition layer is deployed in the power station control room, connecting to 40 PCS units (each 2500kW) via Modbus TCP, 160 battery cluster management slave controllers via CAN bus, and connected to the gateway interface meters via IEC 60870-5-104. It also connects to the environmental and fire protection IoT gateway via MQTT. The edge computing unit uses an industrial-grade server, deploying a data cleaning and alignment engine with a data sampling cycle of 1 second and a feature extraction cycle of 1 minute. The data platform layer adopts a "real-time database + time-series database" architecture. The digital twin model library constructs an electro-thermal coupling model based on the characteristic curves provided by the cell manufacturer and the actual power station topology. The model parameters are identified and corrected using on-site measured data.
[0024] Example 2: Prediction function.
[0025] The forecasting module performs an ultra-short-term forecast every 15 minutes and a short-term forecast for the next day at 17:00 daily. Taking the ultra-short-term photovoltaic output forecast for a certain day in the summer of 2026 as an example, the root mean square error (RMSE) of the LSTM-TCN hybrid model is 4.2% of the rated capacity, which is about 1.8 percentage points lower than that of the single LSTM model; the mean absolute percentage error (MAPE) of the day-ahead electricity price forecast is 6.5%. The forecast results also output the P10 / P50 / P90 quantile forecast intervals.
[0026] Example 3: Optimized scheduling.
[0027] The multi-objective optimization scheduling module generates 96 charging and discharging plans (15-minute resolution) daily based on the next day's electricity price curve, load forecast, and new energy forecast. In the objective function, C_rev is calculated based on the peak-valley price difference in the local spot market, C_deg uses the rainflow counting method to count the equivalent number of cycles per day and then calculates the cost using an empirical attenuation model, and P_dev penalizes plan deviations. The optimization result for a certain day is: charging at 0.5C during off-peak hours (23:00-07:00 the next day) and discharging at 0.5C during peak hours (10:00-12:00, 18:00-21:00). After execution, the daily arbitrage profit is increased by approximately 18% compared to a manual fixed strategy, and the equivalent number of cycles is controlled within 0.9.
[0028] Example 4: Security Verification.
[0029] In the two-dimensional safety plane of the safety assessment and verification module, SOC (State of Charge) is used as the horizontal axis (5%~95% is the nominal safety range) and cell temperature is used as the vertical axis (0~45℃ is the normal range). The boundary is dynamically narrowed as SOH (State of Harm) decreases: when the SOH of a certain cluster of batteries is below 90%, its SOC safety range is narrowed by 3 percentage points, and the temperature difference limit is lowered by 2℃. Before the plan is issued, a digital twin model is used to simulate the 96-point plan. On a high-temperature day, the simulation found that the temperature of the 7th cluster of cells would reach 46.2℃ at 11:30, falling into the power-limited scheduling zone. The system linearly corrected the discharge power from 50MW to 42MW during the 11:00-12:00 period and allowed the plan to be issued only after the simulation was passed again, thus avoiding high-temperature alarms in actual operation.
[0030] Example 5: Rolling Correction and Self-Learning.
[0031] The execution layer operates on a rolling control cycle of 15 minutes. If, at a certain moment, the actual photovoltaic output is 10MW lower than the predicted value, the execution layer will increase the discharge plan by 8MW in the next cycle to make up for the shortfall. The daily plan execution deviation rate decreases from 7.3% without rolling correction to 1.9%. Monthly statistics from the strategy self-learning module reveal a systematically large prediction deviation during charge / discharge switching, automatically triggering incremental retraining of the prediction model. After retraining, the predicted MAPE for that period decreases from 9.1% to 6.4%.
[0032] Those skilled in the art should understand that the specific parameters (power station capacity, sampling period, prediction model structure, safety boundary threshold, etc.) in the above embodiments are merely examples and can be adjusted according to the power station scale, battery type, grid connection requirements and operation mode, without departing from the protection scope of the present invention.
Claims
1. An intelligent dispatch and management system for energy storage power stations based on EMS, characterized in that, The system comprises a data acquisition layer, a data platform layer, an intelligent decision-making layer, an execution layer, and a human-machine interaction layer. The data acquisition layer includes a multi-protocol adapter gateway and an edge computing unit. The multi-protocol adapter gateway collects operational data from the battery management system (BMS), energy storage converter (PCS), grid-connected meters, environmental monitoring devices, and fire monitoring devices via at least one communication protocol selected from Modbus, IEC 60870-5-104, CAN, and MQTT. The edge computing unit performs edge-side cleaning, alignment, and feature extraction on the collected data. The data platform layer includes a real-time database, a time-series database, and a digital twin model library for storing and managing the collected data, and constructing digital twin models based on the electrical topology and battery characteristics of the energy storage power station. The intelligent decision-making layer includes a prediction module, a multi-objective optimization scheduling module, a safety assessment and verification module, and a strategy self-learning module. The prediction module performs ultra-short-term and short-term predictions of load demand, electricity prices, and renewable energy output based on historical operational data and meteorological data. The multi-objective optimization scheduling module aims to maximize the operating benefits of the energy storage power station, minimize battery degradation costs, and optimize grid-friendliness. It constructs and solves a mixed-integer linear programming model based on constraints such as SOC, power, charge / discharge mutual exclusion, and grid-connected capacity to generate a charge / discharge plan. The safety assessment and verification module performs a safety verification of the charge / discharge plan based on SOC, SOH, cell temperature, and battery consistency indicators. The strategy self-learning module updates the prediction model parameters and optimization target weights online based on the plan execution deviation and actual operation effect; the execution layer is used to decompose the charging and discharging plan that has passed the safety verification into control commands and send them to the PCS for execution, and to make real-time corrections to the charging and discharging plan based on the rolling time domain control method during the execution process; the human-machine interaction layer is used to display the operating status, scheduling plan and alarm information of the energy storage power station, and to receive manual intervention commands.
2. The intelligent dispatch and management system for energy storage power stations based on EMS according to claim 1, characterized in that, The prediction module includes a load prediction unit, an electricity price prediction unit, and an output prediction unit. The load prediction unit and the output prediction unit adopt a hybrid prediction model composed of a Long Short-Term Memory (LSTM) network and a Temporal Convolutional Network (TCN). The electricity price prediction unit adopts a gradient boosting tree model. Different model combinations are used according to the prediction period: the ultra-short-term prediction adopts the LSTM-TCN hybrid model, and the short-term prediction adopts the weighted fusion result of the LSTM-TCN hybrid model and the gradient boosting tree model.
3. The intelligent dispatch and management system for energy storage power stations based on EMS according to claim 1, characterized in that, The objective function of the optimization model constructed by the multi-objective optimization scheduling module is: min F = w1(-C_rev) + w2C_deg + w3P_dev, where C_rev is the sum of peak shaving revenue and electricity price arbitrage revenue, C_deg is the battery degradation cost, P_dev is the penalty term for the deviation between actual output and planned output, w1, w2, and w3 are weight coefficients, and w1+w2+w3=1; the battery degradation cost is calculated based on the equivalent number of cycles and the depth of discharge using the rainflow counting method and the battery degradation empirical model.
4. The intelligent dispatch and management system for energy storage power stations based on EMS according to claim 1, characterized in that, The safety assessment and verification module constructs a two-dimensional safety plane using SOC and cell temperature, dividing it into three zones: a normal scheduling zone, a power-limited scheduling zone, and an emergency control zone. When the operating point is in the normal scheduling zone, the system operates at full power according to the charging and discharging plan. When the operating point is in the power-limited scheduling zone, the charging and discharging power is linearly limited according to a preset function, and an alarm is issued. When the operating point is in the emergency control zone, the system performs step-by-step power reduction, shutdown, and fire-fighting linkage control. The boundaries of the three zones are dynamically adjusted based on SOH and battery consistency indicators.
5. The intelligent dispatch and management system for energy storage power stations based on EMS according to claim 1, characterized in that, The execution layer adopts a rolling time-domain control method: within each control cycle, the charging and discharging plan for a future preset duration is optimized and solved starting from the current time. Only the plan for the first time period is executed. In the next control cycle, the plan is re-solved based on the latest measured data and predicted data to achieve rolling correction of the charging and discharging plan.
6. The intelligent dispatch and management system for energy storage power stations based on EMS according to claim 1, characterized in that, The strategy self-learning module uses the plan execution deviation rate, prediction accuracy, and unit capacity revenue as evaluation indicators. When the evaluation indicators are lower than the preset threshold, it triggers the retraining of the prediction model and the adjustment of the target weights. It also feeds back the SOH decay data to the multi-objective optimization scheduling module to dynamically correct the battery decay cost coefficient.
7. The intelligent dispatch and management system for energy storage power stations based on EMS according to claim 1, characterized in that, The digital twin model library in the data platform layer constructs an electro-thermal coupling model based on the series and parallel topology of the energy storage power station, the characteristic curves of individual battery cells, and the thermal characteristic parameters. The safety assessment and verification module calls the electro-thermal coupling model to perform offline simulation verification of the charge and discharge plan. Only after the simulation passes can the charge and discharge plan be issued for execution.
8. A smart dispatch and management method for energy storage power stations based on EMS, applied to the system as described in any one of claims 1 to 7, characterized in that, The process includes the following steps: S1, The data acquisition layer collects the operating data of BMS, PCS, grid-connected meters, environmental monitoring devices and fire monitoring devices through a multi-protocol adapter gateway, and writes it into the data platform layer after cleaning and processing by the edge computing unit; S2, The forecasting module performs ultra-short-term and short-term forecasts of load demand, electricity price and renewable energy output based on historical data and meteorological data. S3. The multi-objective optimization scheduling module constructs and solves a multi-objective optimization model with the objectives of maximizing operational benefits, minimizing battery degradation costs, and optimizing grid-friendliness, thereby generating a charging and discharging plan. S4. The safety assessment and verification module performs a safety verification of the charge and discharge plan based on SOC, SOH, cell temperature and battery consistency indicators. If the verification fails, the charge and discharge plan is corrected according to the power limit and SOC boundary constraints and then re-verified. S5. The verified charge and discharge plan is decomposed into control instructions by the execution layer and sent to the PCS for execution. S6. During execution, monitor the deviation between actual output and planned output in real time, and use rolling time-domain control to correct the charging and discharging plan in real time; S7. Evaluate the operating effect according to the preset cycle, and update the prediction model and optimization target weights through self-learning based on the evaluation results; S8. Feed back the battery SOH decay data to the multi-objective optimization model, and dynamically adjust the battery decay cost coefficient and SOC safety boundary.
9. The method according to claim 8, characterized in that, The safety verification in step S4 specifically includes: forming a two-dimensional safety plane with SOC and cell temperature and dividing it into a normal scheduling zone, a power-limited scheduling zone, and an emergency control zone; simulating the execution process of the charge and discharge plan; if the operating point at any time falls into the power-limited scheduling zone, the charge and discharge power of that period is corrected according to a preset function; if the operating point at any time falls into the emergency control zone, the charge and discharge power of that period is set to zero and an alarm is marked.
10. The method according to claim 8, characterized in that, The inputs for the ultra-short-term forecast in step S2 include historical load sequences, historical power output sequences, numerical weather predictions, and calendar features. The forecast is performed using an LSTM-TCN hybrid model, and the forecast interval is output through quantile regression. In step S3, optimistic and conservative plans are generated based on the upper and lower bounds of the forecast interval, respectively, and the weighted result of the two plans is used as the final charging and discharging plan.
