Photovoltaic energy storage and charging stations participate in the power grid peak shaving and valley filling control system based on demand response.

By constructing a control system for photovoltaic-storage-charging stations and combining source-load forecasting and real-time optimized scheduling, the problem of photovoltaic-storage-charging stations being unable to actively respond to grid scheduling has been solved, achieving precise peak shaving and valley filling and new energy consumption, thereby improving the operating efficiency and economy of the power station.

CN122292469BActive Publication Date: 2026-11-17HENAN SUNSHINE LONGRUI NEW ENERGY CO LTD
View PDF 1 Cites 0 Cited by

Patent Information

Application Number
CN202610325459.1
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2026-03-17
Publication Date
2026-11-17
Estimated Expiration
2046-03-17

AI Technical Summary

Technical Problem

The existing scheduling and control strategies for photovoltaic, energy storage and charging stations cannot proactively respond to grid dispatch instructions, resulting in poor peak shaving and valley filling effects. Furthermore, they fail to effectively adapt to the uncertainties in photovoltaic output and electric vehicle charging demand, leading to significant deviations between operation and planning.

Method used

The photovoltaic-storage-charging station control system based on demand response is adopted, which includes a grid demand response interaction layer, a station-level optimization scheduling layer, and an equipment-level distributed execution layer. Through source-load prediction module, power dispatch prediction module, full-quantity perception module, and real-time rolling correction module, combined with dynamic weight adjustment mechanism and genetic algorithm, real-time optimization and control are achieved.

Benefits of technology

This has enabled photovoltaic-storage-charging stations to transform from passive load providers to proactive and flexible resource regulators, accurately responding to the grid's peak shaving and valley filling needs, improving the absorption of new energy sources, optimizing the economic efficiency of power station operation, and reducing equipment wear and maintenance costs.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122292469B_ABST
    Figure CN122292469B_ABST
Patent Text Reader

Abstract

Based on demand response, photovoltaic-storage-charging stations participate in the power grid peak shaving and valley filling control system. The system adopts a three-layer architecture: a power grid demand response interaction layer, a station-level optimized scheduling layer, and a device-level distributed execution layer, enabling bidirectional demand response interaction with the power grid. The station-level layer constructs modules for source-load prediction, power dispatch estimation using an improved dual-objective genetic algorithm, full-scale sensing, and real-time rolling correction, and designs a dynamic weight adjustment and closed-loop self-learning optimization mechanism. At the device level, based on a charging demand-level strategy and a genetic algorithm, optimal power allocation for charging pile clusters is achieved. This invention solves problems such as low coupling between existing technologies and power grid demand response, poor adaptability to source-load uncertainty, and imbalance of interests among multiple stakeholders. It can achieve global optimization and precise execution of peak shaving and valley filling, significantly improving the proactive support capability of photovoltaic-storage-charging stations for the power grid.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of power system automation technology, specifically to a demand-response-based photovoltaic, energy storage, and charging station-based power grid peak shaving and valley filling control system. Background Technology

[0002] The integrated photovoltaic, energy storage and charging power station combines distributed photovoltaic, energy storage system and electric vehicle charging pile cluster. It can realize the local consumption of new energy and the spatial and temporal transfer of energy. It is an important flexible regulation resource in the new power system and has the natural advantage of participating in grid peak shaving and valley filling and demand response.

[0003] In existing technologies, the scheduling and control strategies for photovoltaic energy storage and charging stations still have the following core shortcomings:

[0004] 1. The strategy has a singular objective and low coupling with the power grid demand response system. Existing control systems mostly focus on the economic efficiency of power plant operation, passively responding only to fixed peak-valley electricity prices. They cannot proactively connect with the dynamic peak-shaving and valley-filling instructions of the power grid dispatch and the market demand for ancillary services. Their proactive support capability for the power grid is insufficient, and they cannot match the real-time adjustment needs of the multi-timescale power system.

[0005] 2. Weak adaptability to uncertainty, resulting in less than expected peak shaving and valley filling effects. Most existing strategies are based on fixed forecast curves to formulate scheduling plans, without fully considering the randomness of photovoltaic power output and the spatiotemporal uncertainty of electric vehicle charging demand. They also lack real-time rolling correction mechanisms, leading to large deviations between actual operation and plans, and even the occurrence of reverse peak shaving, failing to effectively smooth out the peak-valley difference in the power grid. Summary of the Invention

[0006] To address the aforementioned technical problems, the present invention aims to provide a demand-response-based photovoltaic energy storage and charging station participation grid peak shaving and valley filling control system, including a control center, which is equipped with a grid demand response interaction layer, a station-level optimized scheduling layer, and an equipment-level distributed execution layer.

[0007] The grid demand response interaction layer is used to receive grid demand response signals, report adjustable capacity data of photovoltaic, energy storage and charging stations and response feedback data;

[0008] The station-level optimized scheduling layer includes a source-load prediction module, a power dispatch estimation module, a full-scale sensing module, and a real-time rolling correction module;

[0009] The source-load prediction module is used to collect the full operation data of the equipment-level distributed execution layer of each photovoltaic storage and charging station and the demand response signal of the grid connection point in real time, construct the source-load prediction model of the photovoltaic storage and charging station, and output the source-load prediction curve within the long-scale prediction period.

[0010] The power dispatching prediction module is used to improve the dual-objective genetic algorithm by introducing a dynamic weight adjustment mechanism. Based on the source-load prediction curve, full operation data and demand response signal, the improved dual-objective genetic algorithm is used to predict power dispatching and obtain the optimal power dispatching plan.

[0011] Among them, the dynamic weight adjustment mechanism is used to adjust the weight coefficients of the dual-objective optimization function and the global objective function in the process of power dispatch prediction and real-time rolling correction, and to perform self-learning optimization of weight execution and effect feedback based on response feedback data;

[0012] The full-scale perception module is used to perceive the full status and demand response of each optical storage and charging station, and determines whether to generate a station-level fluctuation report and update the source load prediction curve based on the perception results.

[0013] The real-time rolling correction module is used to perform real-time rolling correction of the optimal power dispatch plan based on station-level fluctuation reports or demand response signals, and to obtain real-time control command sequences.

[0014] The device-level distributed execution layer is used to optimally distribute the real-time control command sequence issued by the station-level optimized scheduling layer to each individual pile based on the built-in charging demand classification strategy.

[0015] Furthermore, the power grid demand response interaction layer includes a demand response signal receiving module, a photovoltaic-storage-charging station adjustable capacity reporting module, and a response effect feedback module;

[0016] The demand response signal receiving module is used to receive daytime peak and valley period and time-of-use electricity price signals, intraday peak warning signals, real-time peak shaving demand instructions, ancillary service market signals, and power grid safety constraint limits issued by the power grid.

[0017] The adjustable capacity reporting module for photovoltaic-storage-charging stations is connected to each photovoltaic-storage-charging station in the target area to report to the power grid the photovoltaic expected output curve, energy storage charging and discharging adjustable range, adjustable upper and lower limits of the charging pile cluster, day-ahead peak shaving and valley filling response plan, and intraday adjustable margin of each photovoltaic-storage-charging station.

[0018] The response effect feedback module communicates with each photovoltaic, energy storage and charging station in the target area to collect statistics and report to the power grid the peak load reduction, valley renewable energy consumption increase, demand response command tracking deviation, and peak shaving service completion rate of each photovoltaic, energy storage and charging station.

[0019] Furthermore, the process of outputting the source load prediction curve over a long-scale prediction period includes:

[0020] Real-time collection of full operational data, environmental data, and demand response signals from grid connection points of each optical storage and charging station within the target area at the equipment-level distributed execution layer, marking the collection time, and setting the collection cycle;

[0021] The full operational data, environmental data, and demand response signals of grid-connected points of each optical storage and charging station within several historical acquisition periods in the target area are obtained as training and test sets. A source-load prediction model for the optical storage and charging station is constructed. The training set is input into the source-load prediction model for the optical storage and charging station for training until the loss function is stable and the model parameters are saved. The source-load prediction model for the optical storage and charging station is tested through the test set until it meets the preset requirements, and then the source-load prediction model for the optical storage and charging station is output.

[0022] A long-scale prediction period is set, and the selected timestamp of the long-scale prediction period is used as the prediction time point. At the prediction time point, weather forecast data, user charging reservation data, and demand response signals for the next long-scale prediction period of the photovoltaic-storage-charging station are obtained. The weather forecast data, user charging reservation data, and demand response signals for the next long-scale prediction period are input into the photovoltaic-storage-charging station source-load prediction model. Based on the photovoltaic-storage-charging station source-load prediction model, the photovoltaic output prediction curve and the spatiotemporal distribution prediction curve of charging demand for the next long-scale prediction period are output.

[0023] Furthermore, the process of obtaining the optimal power dispatch plan includes:

[0024] Based on the photovoltaic output prediction curve and charging demand spatiotemporal distribution prediction curve of the photovoltaic-storage-charging station in the next long-scale prediction period, the full operation data of the equipment-level distributed execution layer and the demand response signal of the grid connection point, several correlation quantization functions and constraints are constructed. Based on the correlation quantization functions, a bi-objective optimization function is constructed. The initial conventional weights of the bi-objective optimization function are set. Several power dispatch plans are preset. The power dispatch plans are chromosome-encoded and the population is initialized to generate the initial population. A dynamic weight adjustment mechanism is introduced to improve the bi-objective genetic algorithm.

[0025] Based on the initial population, the bi-objective optimization function, and the constraints, the optimal power dispatch plan is obtained through an improved bi-objective genetic algorithm. The optimal power dispatch plan includes the planned values ​​of energy storage charging and discharging power of the photovoltaic-storage-charging station in each acquisition cycle within the next long-scale prediction cycle, the baseline planned value of the power purchased from the grid by the grid connection point, and the planned charging power and adjustable capacity boundary of the charging pile cluster.

[0026] Furthermore, the constraints include grid-side constraints, equipment-side constraints, safety protection constraints, and power balance constraints. Among them, the condition for satisfying the power balance constraint is that the planned value of the energy storage charging and discharging power of the photovoltaic-storage-charging station is equal to the baseline planned value of the power purchased from the grid by the grid connection point. This equals the planned charging power of the charging pile cluster.

[0027] Furthermore, the process of achieving full-scale status and demand response awareness for each photovoltaic-storage-charging station includes:

[0028] The system compares the full operational data of the equipment-level distributed execution layer of the photovoltaic-storage-charging station collected in real time during the acquisition period with the photovoltaic output prediction curve and the spatiotemporal distribution prediction curve of charging demand during the acquisition period. It obtains the photovoltaic output deviation time series and the charging demand spatiotemporal distribution deviation time series, presets deviation thresholds and time thresholds, and obtains the cumulative time when the photovoltaic output deviation time series and the charging demand spatiotemporal distribution deviation time series are greater than the deviation threshold. If there is a cumulative time corresponding to the photovoltaic output deviation time series or the charging demand spatiotemporal distribution deviation time series that is greater than the time threshold, a station-level fluctuation report is generated. The system inputs the full operational data of the equipment-level distributed execution layer of the photovoltaic-storage-charging station collected in real time during the acquisition period, environmental data, and the demand response signal of the grid connection point into the photovoltaic-storage-charging station source-load prediction model, sets the first time scale, and outputs the photovoltaic output prediction curve and the charging demand spatiotemporal distribution prediction curve for the next segment of the first time scale according to the photovoltaic-storage-charging station source-load prediction model. The system updates the photovoltaic output prediction curve and the charging demand spatiotemporal distribution prediction curve for the acquisition period covered by the first time scale according to the photovoltaic output prediction curve and the charging demand spatiotemporal distribution prediction curve for the first time scale.

[0029] Similarly, if a real-time peak-shaving demand instruction is generated in the demand response signal, the full amount of operating data, environmental data, and demand response signals of the grid-connected points of the photovoltaic, energy storage, and charging stations collected in real time during the collection period will be input into the photovoltaic, energy storage, and charging station source-load prediction model. Based on the photovoltaic output prediction curve and the spatiotemporal distribution prediction curve of charging demand in the first time scale output by the photovoltaic, energy storage, and charging station source-load prediction model, the photovoltaic output prediction curve and the spatiotemporal distribution prediction curve of charging demand in the collection period covered by the first time scale will be updated.

[0030] Furthermore, the process of acquiring the real-time control command sequence includes:

[0031] When a real-time peak-shaving demand instruction is generated in the station-level fluctuation report or demand response signal, a global objective function is constructed based on the photovoltaic output prediction curve, the spatiotemporal distribution prediction curve of charging demand, the full operation data of the equipment-level distributed execution layer in the current collection period, the demand response signal of the grid connection point, and the optimal power dispatch plan of the photovoltaic, energy storage and charging station in the first time scale. The initial conventional weights of the global objective function are set, several real-time control instruction sequences are preset, chromosome encoding and population initialization are performed on the real-time control instruction sequences, an initial population is generated, and a dynamic weight adjustment mechanism is introduced to improve the genetic algorithm.

[0032] Based on the initial population, global objective function, and adjustable capability boundary, an improved genetic algorithm is used to obtain the optimal real-time control command sequence within the current acquisition cycle.

[0033] Furthermore, the process of dynamically adjusting the weights of the bi-objective optimization function and the global objective function in the power dispatch forecasting and real-time rolling correction process includes:

[0034] A pre-defined quantitative standard library is provided, which includes three types of triggering dimensions and the judgment criteria for different priority levels corresponding to the three types of triggering dimensions. The three types of triggering dimensions include the urgency of the power grid demand response command, the power station operation status, and the level and concentration of user charging demand.

[0035] When a real-time peak shaving demand instruction is generated in the station-level fluctuation report or demand response signal, the full operation data of the distributed execution layer of the photovoltaic storage and charging station equipment level and the demand response signal of the grid connection point are obtained in the current collection period. The full operation data and demand response signal are input into the quantitative standard library for comparison, and the priority level corresponding to the three trigger dimensions is obtained. The priority level corresponding to the three trigger dimensions is marked as the trigger scenario.

[0036] A pre-defined weight rule library is used, which includes weight adjustment rules for the dual-objective optimization function and the global objective function corresponding to different triggering scenarios. The weight adjustment rules for the dual-objective optimization function and the global objective function are obtained based on the triggering scenario and the weight rule library. The initial regular weights of the dual-objective optimization function and the global objective function are adjusted according to the weight adjustment rules, and then the corresponding optimal power dispatch plan and optimal real-time control instruction sequence are regenerated.

[0037] Furthermore, the self-learning optimization process for weighted execution and effect feedback based on the data collected by the response effect feedback module includes:

[0038] The corresponding optimal power dispatch plan and optimal real-time control command sequence will be regenerated and sent to the equipment-level distributed execution layer for execution. Simultaneously, the peak load reduction, valley renewable energy consumption improvement, demand response command tracking deviation and peak shaving service completion rate will be obtained from the response effect feedback module after execution.

[0039] The system presets expected threshold ranges for peak load reduction, valley renewable energy consumption improvement, demand response instruction tracking deviation, and peak shaving service completion rate. If any of these values ​​are outside the corresponding expected threshold ranges, a fishbone diagram is constructed for root cause analysis and control variable simulation verification. This optimizes the weight adjustment range in the weight rule library and the grading thresholds in the quantification standard library.

[0040] Furthermore, based on the built-in charging demand tiering strategy, the process of optimally allocating the real-time control command sequence issued by the station-level optimized scheduling layer to each individual charging pile includes:

[0041] The total power limit instruction of the charging pile cluster is obtained according to the received real-time control instruction sequence. The single pile data is obtained according to the full operation data of the distributed execution layer of the photovoltaic energy storage charging station equipment level. The single pile data includes the demand level, user demand power and single pile rated power. The sum of user demand power of all charging piles with demand level of level 1 is calculated. The sum of user demand power of all charging piles with level 1 demand is subtracted from the total power limit to obtain the total adjustable power limit. Constraints are set according to the total power limit instruction of the charging pile cluster and the demand level.

[0042] A power allocation objective function is constructed, several power allocation schemes are preset, chromosome encoding and population initialization are performed on the power allocation schemes to generate an initial population, and the optimal power allocation scheme is obtained through a genetic algorithm based on the initial population, the power allocation objective function and constraints, and power allocation control is executed based on the optimal power allocation scheme.

[0043] Compared with the prior art, the beneficial effects of the present invention are:

[0044] 1. This invention realizes the fundamental transformation of photovoltaic-storage-charging stations from passive load subjects of the power grid to active and flexible regulation resources. It deeply connects the two-way interactive links between photovoltaic, storage, and charging resources and the power grid dispatching system and the power market system, and can fully adapt to the grid's multi-timescale and multi-scenario regulation needs. Through the full-process linkage of source-load forecasting, global optimized scheduling, and real-time rolling correction, it can accurately respond to the core requirements of the power grid for peak shaving and valley filling, effectively smooth the peak-valley difference of the distribution network, significantly alleviate the power supply pressure of the distribution network during peak load periods, and significantly improve the absorption level of distributed renewable energy during valley load periods, solving the grid regulation problems brought about by the high proportion of renewable energy and high penetration rate of electric vehicles. The multi-station collaborative sensing and reporting mechanism allows the power grid dispatching side to accurately grasp the overall regulation potential and real-time operating status of photovoltaic, storage, and charging resources in the target area, providing refined data support for global optimized scheduling of the power grid, and effectively improving the scientific and refined level of regional power grid scheduling.

[0045] 2. The dual-objective optimized scheduling system can maximize the full-cycle operational value of photovoltaic-storage-charging stations while adapting to grid regulation needs. Through globally optimized scheduling plans, it achieves a stable increase in station revenue. Scientific optimization of energy storage charging and discharging strategies effectively reduces cyclic losses in energy storage equipment, extends the lifespan of core equipment, and significantly reduces the station's overall lifecycle maintenance and equipment replacement costs, resulting in a long-term improvement in the station's operational economics. The dynamic weight adjustment mechanism and self-learning optimization system can adaptively adjust the core orientation of the optimization strategy based on dynamic changes in grid demand urgency, real-time station operating status, and user charging demand characteristics. This completely solves the pain point of traditional fixed scheduling strategies being unable to adapt to complex and ever-changing operating scenarios, ensuring the operational stability and revenue reliability of the station under various operating conditions. Attached Figure Description

[0046] Figure 1 This is a flowchart illustrating the demand-response-based photovoltaic energy storage and charging station participating in the power grid peak shaving and valley filling control system according to an embodiment of this application. Detailed Implementation

[0047] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this application. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0048] like Figure 1 As shown, the photovoltaic-storage-charging station participating in the power grid peak shaving and valley filling control system based on demand response includes a control center, which is equipped with a power grid demand response interaction layer, a station-level optimization scheduling layer, and a device-level distributed execution layer.

[0049] The power grid demand response interaction layer is used to receive demand response signals from the power grid at multiple time scales, and to report the adjustable capacity data and response feedback data of the photovoltaic, energy storage and charging stations.

[0050] The station-level optimized scheduling layer includes a source-load prediction module, a power dispatch estimation module, a full-scale sensing module, and a real-time rolling correction module;

[0051] The source-load prediction module is used to collect the full operation data of the equipment-level distributed execution layer of each photovoltaic storage and charging station and the demand response signal of the grid connection point in real time, construct the source-load prediction model of the photovoltaic storage and charging station, and output the source-load prediction curve within the long-scale prediction period.

[0052] The power dispatching prediction module is used to improve the dual-objective genetic algorithm by introducing a dynamic weight adjustment mechanism. Based on the source-load prediction curve, full operation data and demand response signal, the improved dual-objective genetic algorithm is used to predict power dispatching and obtain the optimal power dispatching plan.

[0053] Among them, the dynamic weight adjustment mechanism is used to adjust the weight coefficients of the dual-objective optimization function and the global objective function in the process of power dispatch prediction and real-time rolling correction, and to perform self-learning optimization of weight execution and effect feedback based on response feedback data;

[0054] The full-scale perception module is used to perceive the full status and demand response of each optical storage and charging station, and determines whether to generate a station-level fluctuation report and update the source load prediction curve based on the perception results.

[0055] The real-time rolling correction module is used to perform real-time rolling correction of the optimal power dispatch plan based on station-level fluctuation reports or demand response signals, and to obtain real-time control command sequences.

[0056] The device-level distributed execution layer is used to optimally distribute the real-time control command sequence issued by the station-level optimized scheduling layer to each individual pile based on the built-in charging demand classification strategy.

[0057] It should be further explained that, in the specific implementation process, the power grid demand response interaction layer includes a demand response signal receiving module, a photovoltaic-storage-charging station adjustable capacity reporting module, and a response effect feedback module.

[0058] The demand response signal receiving module is used to receive daytime peak and valley period and time-of-use price signals, intraday peak warning signals, real-time peak shaving demand instructions, ancillary service market signals and grid security constraint limits issued by the power grid through the power dedicated communication network.

[0059] The adjustable capacity reporting module for photovoltaic-storage-charging stations is connected to each photovoltaic-storage-charging station in the target area. It is used to report the expected output curve of photovoltaic power generation, adjustable range of energy storage charging and discharging, adjustable upper and lower limits of the charging pile cluster, day-ahead peak shaving and valley filling response plan, and intraday adjustable margin of each photovoltaic-storage-charging station to the power grid, so as to support the power grid in issuing demand response instructions at multiple time scales.

[0060] The response effect feedback module communicates with each photovoltaic, energy storage and charging station in the target area to collect statistics and report to the power grid the peak load reduction, valley renewable energy consumption increase, demand response command tracking deviation, and peak shaving service completion rate of each photovoltaic, energy storage and charging station, which serve as the basis for ancillary service settlement.

[0061] It should be further explained that, in the specific implementation process, the process of collecting full operational data of the equipment-level distributed execution layer of each photovoltaic-storage-charging station and the demand response signals of the grid connection points in real time, constructing the source-load prediction model of the photovoltaic-storage-charging station, and outputting the source-load prediction curve for the long-scale prediction period includes:

[0062] Through the industrial Ethernet within the station, real-time collection of full operational data (including: output, voltage, and current data of the photovoltaic array; SOC, charging and discharging power, battery temperature, and SOH data of the energy storage system; operating status, charging power, vehicle battery SOC, and user charging demand data of each charging pile in the charging pile cluster), environmental data (including light intensity, ambient temperature, and humidity), and demand response signals from grid connection points (including voltage, frequency, active / reactive power, and power factor data of the grid connection points) are collected, and the collection time is marked and the collection cycle is set (15 minutes).

[0063] The full operational data, environmental data, and demand response signals of grid-connected points of each optical storage and charging station within several historical acquisition periods in the target area are obtained as training and test sets. A source-load prediction model for the optical storage and charging station is constructed. The training set is input into the source-load prediction model for the optical storage and charging station for training until the loss function is stable and the model parameters are saved. The source-load prediction model for the optical storage and charging station is tested through the test set until it meets the preset requirements, and then the source-load prediction model for the optical storage and charging station is output.

[0064] A long-scale prediction period (24h) is set, and the selected timestamp of the long-scale prediction period (e.g., 16:00 every day) is used as the prediction time point. At the prediction time point, weather forecast data, user charging reservation data, and demand response signals of the photovoltaic-storage-charging station for the next long-scale prediction period are acquired. The weather forecast data, user charging reservation data, and demand response signals of the next long-scale prediction period are input into the photovoltaic-storage-charging station source-load prediction model. Based on the photovoltaic-storage-charging station source-load prediction model, the photovoltaic output prediction curve and the spatiotemporal distribution prediction curve of charging demand for the next long-scale prediction period are output. The time resolution is 15min, and several acquisition cycles are divided within the next long-scale prediction period.

[0065] Building a source-load prediction model for a photovoltaic-storage-charging station is a complex process involving multiple steps such as model selection, training, validation, and testing. The following is a detailed supplementary explanation of this process:

[0066] It collects full operational data, environmental data, and demand response signals from grid-connected points at the device-level distributed execution layer, covering the entire "source-load-grid-environment-user" dimension. Unlike traditional prediction models that only collect single environmental / device data, it has more comprehensive feature dimensions. The specific collection content is as follows:

[0067] Photovoltaic output data: historical / real-time output, voltage, and current of the photovoltaic array; on-site environmental data such as irradiance, ambient temperature, humidity, and wind speed; weather forecast data (irradiance, temperature, and weather conditions).

[0068] Charging demand-related data: historical / real-time charging power of charging pile clusters, single pile operating status; electric vehicle user charging reservation data (reservation time, charging power, estimated charging duration); vehicle battery SOC data; distribution of charging user types at different times (commercial vehicles / private cars).

[0069] Grid demand response signals: voltage, frequency, active / reactive power, and power factor at the grid connection point; grid peak-valley time division and time-of-use pricing signals;

[0070] Historical time series data: at least one year of the above full data (this embodiment of the invention uses one year of historical data) to ensure that the model can learn the seasonal, time period, weekday / holiday and other time series variation patterns of the source load.

[0071] Traditional LSTM neural networks suffer from problems such as long-term gradient vanishing, weak feature extraction targeting, and poor coupling between dual-objective predictions. This invention improves the LSTM structure to address the dual-objective nature, temporal correlation, and multi-feature coupling of source-load prediction for photovoltaic power generation and charging stations. An improved network architecture of "feature fusion layer + bidirectional LSTM + attention mechanism + dual output layer" is designed. This architecture retains the LSTM's ability to model long-term data while improving the efficiency of extracting key source-load features, adapting to the dual-objective collaborative prediction of photovoltaic power output and charging demand. The design of each network layer is as follows:

[0072] Input layer: The input layer is the feature entry point of the model, with a dimension of T×N, where T is the time step (set according to the historical data pattern, such as selecting the time series data of the previous 24 hours as input, with a time resolution of 15 minutes, then T=96), and N is the total number of preprocessed multi-dimensional features (including photovoltaic, charging, environment, power grid, user, etc.). The input layer passes the preprocessed time series feature matrix to the feature fusion layer to complete the initial coupling of multiple features.

[0073] Feature Fusion Layer: A Temporal Convolutional Network (TCN) is introduced to construct a feature fusion layer. The one-dimensional convolutional kernel of the TCN is used to extract local temporal features of the source and load data (such as short-term sudden rises and falls in photovoltaic output and sudden increases in charging load), which makes up for the shortcomings of traditional LSTM in capturing local temporal features. At the same time, through the multi-scale receptive field of the convolutional kernel, the coupled features of the grid side and the environment side (such as the impact of grid connection point voltage constraints on photovoltaic output) are fused, and the fused feature vector is passed to the bidirectional LSTM layer.

[0074] Bidirectional LSTM layer: A bidirectional LSTM is used to replace the traditional unidirectional LSTM. It learns the variation patterns of source-load time series data from both the forward (past to future) and backward (future to past) directions to capture the full time series correlation between photovoltaic power output and charging demand (such as the correlation between morning peak charging demand and the previous night's off-peak charging, and the correlation between afternoon peak photovoltaic power output and morning irradiance). Multiple bidirectional LSTM hidden layers are set (the number of layers is optimized according to the amount of data, generally 2-3 layers), and the number of neurons in each layer is 64 / 128 / 256 (adjusted as needed) to achieve deep modeling of long time series and multi-coupling features.

[0075] Attention Mechanism Layer: An additive attention mechanism is added after the bidirectional LSTM layer to assign weights to the hidden layer features of the bidirectional LSTM output: higher weights are given to key features that affect source-load prediction (such as light intensity on photovoltaic output, user reservations on charging demand, and grid connection voltage on photovoltaic power limitation), and lower weights are given to secondary features (such as ambient humidity on fast charging demand). This eliminates the interference of invalid features and improves the model's prediction accuracy for core factors. This is the key improvement point for this model to achieve a prediction accuracy of ≥90%.

[0076] Fully connected layer: The weighted feature vector output from the attention mechanism layer is input into the fully connected layer. Through linear transformation, the high-dimensional features are mapped to low dimensions and integrated into a core feature vector that can represent the source load change law, providing a prediction basis for the output layer.

[0077] Dual Output Layer: For the dual-objective prediction of photovoltaic power output and charging demand, an independent dual output layer is designed to output the predicted value of photovoltaic power output and the predicted value of total charging demand, respectively. At the same time, the spatiotemporal distribution coefficient of charging demand (i.e. the load ratio of each charging pile) is also output, which directly provides the spatial allocation basis of charging load for station-level scheduling. The output layer adopts a linear activation function to adapt to the continuous value output demand of power regression prediction.

[0078] This model is a bi-objective regression prediction model, using the weighted fusion of root mean square error (RMSE) as the core loss function, taking into account the error convergence of both prediction objectives. The formula is as follows: in, , The weighting of losses for photovoltaic power output and charging demand (can be dynamically adjusted according to the priority of grid demand response, such as increasing during peak hours). ), The photovoltaic power output is the predicted value (unit: kW). The actual output of photovoltaic power (unit: kW). The total power demand prediction (unit: kW) is the output value of the charging demand dimension in the dual output layer of the improved LSTM model. It includes the prediction of the total active power of the charging pile cluster, and is associated with the spatiotemporal distribution coefficient of charging demand (which can be decomposed into the predicted charging power of each individual charging pile). It is time series data (15 min). This represents the actual charging demand (unit: kW).

[0079] During training, the weights are continuously updated using the backpropagation algorithm, causing the loss function to gradually decrease until it reaches a stable state. During this period, techniques such as early stopping are used to avoid overfitting. In addition to the basic training process, various model parameters, including the learning rate, batch size, and regularization coefficient, are fine-tuned using grid search.

[0080] Once the model training is complete and the parameters have been tuned, a final evaluation is performed using a test set to obtain the model's evaluation results. These results include classification metrics such as accuracy, recall, and F1 score. Based on the evaluation results on the test set, it is determined whether the model has met the expected standards. If the requirements are met, the model parameters are saved and deployment is prepared; otherwise, it is necessary to return to a previous stage to re-examine issues such as data quality, model structure, or training strategy.

[0081] It should be further explained that, in the specific implementation process, a dynamic weight adjustment mechanism is introduced to improve the bi-objective genetic algorithm. Based on the source-load prediction curve, full operational data, and demand response signals, the improved bi-objective genetic algorithm is used to predict power dispatch and obtain the optimal power dispatch plan. The process includes:

[0082] Based on the photovoltaic output forecast curve and charging demand spatiotemporal distribution forecast curve of the next long-scale forecast period for the photovoltaic-storage-charging station, the full operation data of the equipment-level distributed execution layer (including rated parameters of photovoltaic / energy storage / charging piles, safe operating range of energy storage SOC (20%~80%), charging and discharging power limits, energy storage life loss cost coefficient, photovoltaic-storage-charging station operation and maintenance cost coefficient, final value of energy storage SOC of the previous day, equipment health status (SOH)) and the demand response signals of the grid connection point (the next day's peak and valley time division, time-of-use electricity price, demand response subsidy standard, ancillary service market rules, upper and lower limits of grid connection point power, and grid safety constraint limits), several correlation quantification functions and constraints are constructed. Based on the correlation quantification functions, a bi-objective optimization function is constructed. The initial conventional weights of the bi-objective optimization function are set, several power dispatch plans are preset, chromosome encoding and initialization of the population are performed on the several power dispatch plans, an initial population is generated, and a dynamic weight adjustment mechanism is introduced to improve the bi-objective genetic algorithm.

[0083] Based on the initial population, the bi-objective optimization function, and constraints, an improved bi-objective genetic algorithm is used to obtain the optimal power dispatch plan. The optimal power dispatch plan includes the planned energy storage charging and discharging power values ​​of the photovoltaic-storage-charging station in each acquisition cycle within the next long-scale prediction period. The baseline planned value of the power purchased from the grid by the grid connection point. and charging pile cluster planned charging power And adjustable capacity boundaries (including: adjustable upper and lower limits of energy storage charging and discharging power, upper and lower limits of total power of charging pile cluster, upper and lower limits of adjustable load of cluster, and upper and lower limits of grid connection point power).

[0084] It should be further explained that, in the specific implementation process, the bi-objective optimization function is as follows:

[0085] ;

[0086] ;

[0087] ;

[0088] in, For fitness value, Let t be the power purchased from the grid at the grid connection point (positive value for purchased power, negative value for power supplied to the grid). To minimize the difference between the maximum and minimum power purchased by the grid connection point throughout the day, thus fulfilling the core requirement of smoothing out peak-valley differences in the power grid, For charging service revenue, For demand response subsidy revenue, For the cost of purchasing electricity from the power grid, Costs related to energy storage lifespan depletion; To maximize the net revenue of the photovoltaic-storage-charging station throughout the day; , These are the weight coefficients of the two objective functions. , These are the theoretical maximum values ​​of the two objective optimization functions, used for normalization to eliminate the influence of dimensions.

[0089] The specific correlation quantization functions are as follows:

[0090] ;

[0091] ;

[0092] ;

[0093] ;

[0094] ;

[0095] First, the time dimension is unified (specifically, a long-scale prediction period of 24 hours and 15 minutes resolution, with a total number of time periods T = 24 × 4 = 96, and the duration of a single time period). =0.25h), serving as the universal basis for all formulas. The station's power consumption and line loss power at time t are fixed predicted values. This represents the photovoltaic array output value at time t, predicted in the ultra-short term of the previous 24 hours. This represents the total charging demand power of the charging pile cluster at time t, predicted in the ultra-short-term 24-hour forecast. The energy storage charging and discharging power at time t (positive value for discharging, negative value for charging). The planned charging power (kW) of the charging pile cluster during time period t. The planned power purchase capacity (kW) for the grid connection point during time period t. The planned charging and discharging power of energy storage during time period t (kW, with discharge being positive and charging being negative). The time-of-use electricity price for period t is (yuan / kWh). The unit charging service price for charging piles is (yuan / kWh), and σ(t) is the unit subsidy price for demand response during time period t (yuan / kWh). This refers to the set of peak periods (the day-ahead peak-valley period segmentation signal in the demand response signal). This refers to the set of off-peak periods (the daily peak-valley period segmentation signal in the demand response signal). Price subsidy per unit for peak shaving (RMB / kWh) The subsidy price for renewable energy consumption units in the valley section (yuan / kWh). The maximum power purchase capacity (kW) at the grid connection point during peak hours (grid safety constraint limit). The lower limit of the power purchase capacity at the grid connection point during off-peak hours (kW) (grid safety constraint limit), and k is the life loss cost of energy storage unit charge and discharge capacity (yuan / kWh).

[0096] The specific process of obtaining the optimal power dispatch plan using the improved bi-objective genetic algorithm based on the initial population, bi-objective optimization function, and constraints is as follows:

[0097] The power dispatch plan is encoded as chromosomes. For example, a string of numbers represents the planned power charging and discharging power of photovoltaic-storage-charging stations in each time period, the baseline planned power purchased from the grid by grid connection points in each time period, and other information. A certain number of chromosomes are randomly generated to form an initial population, with each chromosome representing a possible power dispatch plan. The power dispatch plan is then constrained and validated based on constraints, including:

[0098] Energy storage SOC trajectory verification throughout the entire time period: ensures that the SOC after charging does not exceed 80% and the SOC after discharging does not fall below 20%, with no risk of exceeding the limit;

[0099] Grid connection point power verification throughout the day: Ensure that the power purchased during peak hours does not exceed the grid limit, there is no reverse power, and it meets the requirements for preventing backflow;

[0100] Equipment power limit verification: Ensure that the planned power of all equipment does not exceed the rated parameters and there is no risk of overload;

[0101] Power dispatch plans that do not meet the constraints are eliminated. Then, a tournament selection method is used to select chromosomes with lower fitness values ​​from the current population as parents. The parent chromosomes are crossovered, exchanging some genes to generate new offspring chromosomes. This simulates the exchange of genes in biological genetics to generate new power dispatch plans. The offspring chromosomes are mutated, randomly changing some genes to increase population diversity and avoid getting trapped in local optima. The above steps are repeated iteratively until the termination condition is met, such as reaching the maximum number of iterations or the fitness value no longer decreasing significantly. The optimal power dispatch plan is then output.

[0102] It should be further explained that, in the specific implementation process, the constraints include grid-side constraints: upper and lower limits of grid connection power, voltage / frequency deviation, reverse current prevention, and power factor constraints; equipment-side constraints: energy storage SOC operating range, charging and discharging power, and photovoltaic / charging pile output limit constraints; and safety protection constraints: battery over-temperature / over-current / over-voltage protection, grid fault ride-through constraints, and power balance constraints. Among these, the condition for meeting the power balance constraint is that the planned value of energy storage charging and discharging power of the photovoltaic-energy storage-charging station is equal to the baseline planned value of the power purchased from the grid at the grid connection point. This equals the planned charging power of the charging pile cluster.

[0103] It should be further explained that, in the specific implementation process, the process of conducting full-scale status and demand response sensing for each photovoltaic storage and charging station, and determining whether to generate a station-level fluctuation report and update the source-load prediction curve based on the sensing results includes:

[0104] The system compares the full operational data of the equipment-level distributed execution layer of the photovoltaic-storage-charging station collected in real time during the acquisition period with the photovoltaic output prediction curve and the spatiotemporal distribution prediction curve of charging demand during the acquisition period to obtain the photovoltaic output deviation time series and the charging demand spatiotemporal distribution deviation time series. Preset deviation thresholds and time thresholds are set, and the cumulative time when the photovoltaic output deviation time series or the charging demand spatiotemporal distribution deviation time series exceeds the deviation threshold is obtained. If the cumulative time corresponding to either the photovoltaic output deviation time series or the charging demand spatiotemporal distribution deviation time series exceeds the time threshold, a station-level fluctuation report is generated. The system then inputs the full operational data of the equipment-level distributed execution layer of the photovoltaic-storage-charging station collected in real time during the acquisition period, environmental data, and the demand response signal from the grid connection point into the photovoltaic-storage-charging station's source and load data. The prediction model sets a first time scale (4h). Based on the source-load prediction model of the photovoltaic-storage-charging station, it outputs the photovoltaic output prediction curve and the spatiotemporal distribution prediction curve of charging demand within the first time scale at the current moment. Based on the photovoltaic output prediction curve and the spatiotemporal distribution prediction curve of charging demand within the first time scale, the photovoltaic output prediction curve and the spatiotemporal distribution prediction curve of charging demand within the collection period covered by the first time scale are updated. Based on the data within 15 minutes, the source-load rolling prediction curve for the next 4 hours is dynamically generated. The deviation of the photovoltaic output prediction curve and the spatiotemporal distribution prediction curve of charging demand in the long-scale prediction period is continuously corrected, which effectively smooths the randomness of photovoltaic output (such as cloud cover) and the spatiotemporal uncertainty of charging demand (such as the simultaneous start of multiple fast charging piles).

[0105] Similarly, if a real-time peak-shaving demand instruction is generated in the demand response signal, the full amount of operating data, environmental data, and demand response signals of the grid-connected points of the photovoltaic, energy storage, and charging stations collected in real time during the collection period will be input into the photovoltaic, energy storage, and charging station source-load prediction model. Based on the photovoltaic output prediction curve and the spatiotemporal distribution prediction curve of charging demand in the first time scale output by the photovoltaic, energy storage, and charging station source-load prediction model, the photovoltaic output prediction curve and the spatiotemporal distribution prediction curve of charging demand in the collection period covered by the first time scale will be updated.

[0106] Because photovoltaic, energy storage, and charging stations submit their optimal power dispatch plans for the following day to the power grid before 16:00 every day, these plans are part of the daily dispatch plan of the power grid dispatch center and are also the core basis for demand response ancillary service settlement and electricity spot market transactions. They have strict power grid compliance requirements. If the filed optimal power dispatch plan is frequently changed, the power grid will be unable to calculate the demand response completion rate and peak-shaving service volume of the photovoltaic, energy storage, and charging stations, which will directly affect the settlement of ancillary service subsidies and may even be judged as a breach of contract by the power grid, resulting in the cancellation of their demand response participation qualification. Therefore, a real-time rolling correction module is needed to smooth out the random uncertainty of source and load and solve the problem of day-ahead plan execution deviation.

[0107] It should be further explained that, in the specific implementation process, the process of performing real-time rolling correction of the optimal power dispatch plan based on station-level fluctuation reports or demand response signals to obtain the real-time control command sequence includes:

[0108] When a real-time peak-shaving demand instruction is generated in the station-level fluctuation report or demand response signal, a global objective function is constructed based on the photovoltaic output prediction curve, the spatiotemporal distribution prediction curve of charging demand, the full operation data of the equipment-level distributed execution layer in the current collection period, the demand response signal of the grid connection point, and the optimal power dispatch plan of the photovoltaic, energy storage and charging station in the first time scale. The initial conventional weights of the global objective function are set, several real-time control instruction sequences are preset, chromosome encoding and population initialization are performed on the real-time control instruction sequences, an initial population is generated, and a dynamic weight adjustment mechanism is introduced to improve the genetic algorithm.

[0109] Based on the initial population, global objective function, and adjustable capability boundary, the improved genetic algorithm obtains the optimal real-time control command sequence within the current acquisition cycle. The optimal real-time control command sequence includes the real-time bidirectional charging and discharging power commands (positive values ​​for discharging and negative values ​​for charging) of the energy storage units of the photovoltaic-storage-charging station within the acquisition cycle, the real-time command of the total power limit of the charging pile cluster, the power control target value of the grid connection point, and the photovoltaic output limit control value.

[0110] It should be further explained that, in the specific implementation process, the global objective function is as follows:

[0111] ;

[0112] Among them, based on a 15-minute prediction resolution =16, This represents the power purchased from the grid at the grid connection point at time t+k (positive values ​​represent purchased power, and negative values ​​represent power supplied to the grid). This represents the planned power purchase value at the grid connection point at time t+k in the optimal power dispatch plan. The energy storage charging and discharging power at time t+k (positive value for discharging, negative value for charging). This represents the planned energy storage charging and discharging power at time t+k in the optimal power dispatch plan. Total adjustment of charging power of the charging pile cluster at time t+k. , , These are the weighting coefficients;

[0113] The specific process of obtaining the optimal real-time control command sequence using an improved genetic algorithm based on the initial population, global objective function, and adjustable capability boundary is as follows:

[0114] Real-time control command sequences are encoded as chromosomes. A certain number of chromosomes are randomly generated to form an initial population, with each chromosome representing a possible real-time control command sequence. The real-time control command sequences are constrained and verified based on the adjustable capability boundary, and power dispatch plans that do not meet the adjustable capability boundary are eliminated. Then, a tournament selection method is used to select chromosomes with lower fitness values ​​from the current population as parents. The parent chromosomes are crossbred, exchanging some genes to generate new offspring chromosomes. This simulates biological gene exchange and generates new real-time control command sequences. The offspring chromosomes are mutated, randomly changing some genes to increase population diversity and avoid getting trapped in local optima. The above steps are repeated iteratively until the termination condition is met, such as reaching the maximum number of iterations or the fitness value no longer decreasing significantly. The optimal real-time control command sequence is then output.

[0115] It should be further explained that, in the specific implementation process, the dynamic adjustment of weights mechanism for adjusting the weight coefficients of the bi-objective optimization function and the global objective function in the power dispatch forecasting and real-time rolling correction process includes:

[0116] A pre-defined quantitative standard library is provided, which includes three types of triggering dimensions and the judgment criteria for different priority levels corresponding to the three types of triggering dimensions. The three types of triggering dimensions include the urgency of the power grid demand response command, the power station operation status, and the level and concentration of user charging demand.

[0117] When a real-time peak shaving demand instruction is generated in the station-level fluctuation report or demand response signal, the full operation data of the distributed execution layer of the photovoltaic storage and charging station equipment level and the demand response signal of the grid connection point are obtained in the current collection period. The full operation data and demand response signal are input into the quantitative standard library for comparison, and the priority level corresponding to the three trigger dimensions is obtained. The priority level corresponding to the three trigger dimensions is marked as the trigger scenario.

[0118] A pre-defined weight rule library is used, which includes weight adjustment rules for the dual-objective optimization function and the global objective function corresponding to different triggering scenarios. The weight adjustment rules for the dual-objective optimization function and the global objective function are obtained according to the triggering scenario and the weight rule library. The initial regular weights of the dual-objective optimization function and the global objective function are adjusted according to the weight adjustment rules. Then, the corresponding optimization calculation is performed to regenerate the corresponding optimal power dispatch plan and the optimal real-time control instruction sequence.

[0119] It should be further explained that, in the specific implementation process, before the photovoltaic-storage-charging station is put into operation, the benchmark weight combination for normal operating conditions is determined based on the power station design parameters (photovoltaic installed capacity, energy storage capacity, charging pile scale), the demand response rules of the local power grid, electricity price and subsidy policies, and historical operating data.

[0120] =0.4, =0.6 (normal operating condition setting). =0.4, =0.3, =0.3 (balancing the fundamental requirements of real-time power grid command tracking, equipment safety, and user experience), and , satisfy + =1, , , satisfy .

[0121] The three trigger dimensions for weight adjustment are pre-quantified and graded, and the judgment criteria and priority rules for different levels are clarified to provide a trigger basis for dynamic adjustment, as shown in Table 1 below:

[0122] Table 1

[0123] Adjust dimensions Level 1 (Regular / Low Priority) Level 2 (Warning / Medium Priority) Level 3 (Urgent / High Priority) urgency of power grid demand response instructions Only the day-ahead peak-valley electricity price and routine peak-shaving plan are issued, ensuring the safety of the power grid load. Issued intraday peak load warning and power grid peak load gap warning, but no mandatory emergency orders. The issuance of real-time emergency peak-shaving orders tightens grid security constraints, posing risks of load gaps and voltage / frequency exceeding limits. Power plant operating status Photovoltaic output is normal, energy storage SOC is within the safe range of 20%-80%, and equipment is functioning without fault. Photovoltaic output fell short of expectations, energy storage SOC was close to its upper or lower limits (<30% or >70%), and there were non-core equipment failures. Core equipment failure, energy storage SOC exceeding limit User charging demand level and concentration Charging pile access rate <30%, Level 1 emergency charging demand accounts for <10%, and there is sufficient adjustment margin. Charging pile access rate is 30%-70%, with first-level emergency demand accounting for 10%-20%, leaving room for adjustment. Charging pile access rate >70%, Level 1 emergency demand accounts for >20%, and adjustment margin is insufficient.

[0124] Subsequently, real-time, multi-dimensional status perception is performed, obtaining data from full operational data and demand response signals from grid connection points:

[0125] Grid-side data: Real-time acquisition of demand response signal types, command urgency, grid security constraints, and peak-shaving demand completion time limits from the grid demand response interaction layer;

[0126] Power station side data: Obtain real-time photovoltaic output, energy storage SOC / SOH, equipment operating status, current adjustable capacity, and revenue completion status from full operation data;

[0127] User-side data: Real-time charging demand classification data, charging pile access rate, charging demand fulfillment rate, and user charging reservation data are obtained from the full operational data.

[0128] Subsequently, the weight adjustment of the bi-objective optimization function is applied to the 24-hour global scheduling plan. The core is to balance the long-term control needs of peak shaving and valley filling on the grid side with the full-cycle operation benefits on the power plant side. The adjustment rules and logic for different scenarios are shown in Table 2 below:

[0129] Table 2

[0130] Triggering Scenario Weighting adjustment rules Standard operating conditions (Grid Level 1, Power Plant Level 1, User Level 1-2) Maintain baseline values: w1=0.4, w2=0.6 Level 2 power grid early warning condition W1 is revised upward to 0.5-0.6, and W2 downward to 0.4-0.5. Level 3 emergency status of power grid W1 has been significantly revised upwards to 0.7-0.8; W2 has been correspondingly revised downwards to 0.2-0.3. Level 2-3 restricted / abnormal operating conditions of the power plant W2 was revised upward to 0.7-0.9, and W1 was revised downward to 0.1-0.3.

[0131] The global objective function adjustment is for real-time control with a 15-minute rolling cycle. The core is to adapt to sudden fluctuations in real-time operating conditions and ensure accurate execution of minute-level demand response. The adjustment rules and logic for different scenarios are shown in Table 3 below:

[0132] Table 3

[0133] Triggering Scenario Weighting adjustment rules Standard operating conditions (Grid Level 1, Power Plant Level 1, User Level 1-2) Maintain the baseline values: β1=0.4, β2=0.3, β3=0.3 Level 3 Emergency Peak Shaving Condition of Power Grid β1 was significantly revised upwards to 0.7-0.8, β2 downwards to 0.1-0.15, and β3 downwards to 0.1-0.15. Abnormal operating conditions of power plant level 2-3 equipment β2 was significantly increased to 0.7-0.8, β1 was decreased to 0.1-0.2, and β3 remained at 0.1-0.15. User Level 3 High Load / Urgent Demand Concentrated Working Conditions β3 was significantly revised upwards to 0.7-0.8, β1 downwards to 0.1-0.15, and β2 downwards to 0.1-0.15.

[0134] It should be further explained that, in the specific implementation process, the self-learning optimization process of weight execution and effect feedback based on the data collected by the response effect feedback module includes:

[0135] The corresponding optimal power dispatch plan and optimal real-time control command sequence will be regenerated and sent to the equipment-level distributed execution layer for execution. Simultaneously, the peak load reduction, valley renewable energy consumption improvement, demand response command tracking deviation and peak shaving service completion rate will be obtained from the response effect feedback module after execution.

[0136] The system presets expected threshold ranges for peak load reduction, valley renewable energy consumption improvement, demand response instruction tracking deviation, and peak shaving service completion rate. If any of these values ​​are outside the corresponding expected threshold ranges, a fishbone diagram is constructed for root cause analysis and control variable simulation verification. This optimizes the weight adjustment range in the weight rule library and the grading thresholds in the quantification standard library.

[0137] It should be further explained that, in the specific implementation process, the process of optimizing the weight adjustment range in the weight rule base and the hierarchical threshold in the quantization standard base includes:

[0138] After regenerating the corresponding optimal power dispatch plan and optimal real-time control command sequence, the response feedback results are as follows:

[0139] On the power grid side: Peak shaving and valley filling completion rate is 92% (expected ≥95%).

[0140] Power plant side: Monthly operating revenue increased by 18% (expected ≥20%).

[0141] On the user side: the charging demand satisfaction rate is 94% (expected ≥95%).

[0142] Subsequently, using the "fishbone diagram analysis method + controlled variable method," the root cause of the failure to meet expectations was identified: unreasonable weight adjustment range (e.g., adjustment range too small to achieve the desired effect, adjustment range too large to harm the interests of other entities) or unscientific setting of tiered thresholds (e.g., threshold too wide leading to misjudgment of the scenario, threshold too narrow leading to excessively frequent adjustments).

[0143] The fishbone diagram is constructed from two dimensions: "weight adjustment range" and "tiered threshold setting," as shown in Table 4.

[0144] Table 4

[0145] Dimension Potential sub-cause Investigation methods Weight adjustment range The increase in w1 is too small (not sufficient to guarantee peak shaving for the power grid). Controlled variable method: Keeping other conditions constant, only adjusting the amplitude of w1, and simulating the change in peak shaving completion rate. The increase in β1 is too large (excessively sacrificing the interests of users and power plants). Controlled variable method: Keeping other conditions constant, only the magnitude of β1 is reduced, and the changes in user satisfaction rate and power plant revenue are simulated. Hierarchical threshold settings The "load gap threshold" for Level 3 emergency peak shaving in the power grid is set too wide (misclassifying non-emergency scenarios as Level 3). Statistical analysis method: Analyze the correlation between historical power grid load gaps and peak-shaving effects, and redetermine the threshold. The "access rate threshold" for Level 2 users with medium load is set too narrowly (misclassifying actual high load as medium load). Statistical analysis method: Analyze the correlation between historical access rate and charging satisfaction rate, and redetermine the threshold.

[0146] A simulation model of a photovoltaic energy storage and charging station was built under the current triggering scenario, and the potential causes were verified one by one using the controlled variable method.

[0147] Verification 1: Is the adjustment range of the upper layer w1 reasonable?

[0148] Original adjustment plan: w1 will be increased from the baseline of 0.4 to 0.75, an increase of 87.5%;

[0149] Simulation adjustment: Keeping other conditions unchanged, adjust w1 by 75% (w1=0.7), 87.5% (original scheme), and 100% (w1=0.8) respectively.

[0150] Simulation results:

[0151] When w1=0.7, the peak shaving completion rate is 93% (still below expectations).

[0152] When w1=0.75, the peak shaving completion rate was 92% (not meeting expectations).

[0153] When w1=0.8, the peak shaving completion rate was 96% (as expected), but the user satisfaction rate dropped to 92% (below expectations).

[0154] Preliminary conclusion: Insufficient increase in w1 resulted in the grid side not meeting expectations, but excessive increase would harm the interests of users, so the tiered thresholds need to be optimized simultaneously.

[0155] Verification 2: Is the tiered threshold setting reasonable?

[0156] The original power grid level 3 threshold is: "load gap ≥ 1500kW"; the original user level 2 threshold is: "access rate 30%-70%";

[0157] Historical data analysis revealed that when the load gap is ≥1600kW, a larger w1 is required to achieve the expected results; when the connection rate is ≥45%, the user-side adjustment margin is clearly insufficient, and it should be classified as a Level 3 high load.

[0158] Simulation adjustments: Modify the power grid level 3 threshold to "load gap ≥ 1600kW", modify the user level 2 threshold to "30%-45%", and modify the level 3 threshold to "≥ 45%";

[0159] Optimization results: Under the current triggering scenario, the judgment was re-evaluated as "Grid Level 3 + User Level 3", and the weights were adjusted to w1=0.8 and λ3=0.75. The final peak shaving completion rate was 96%, the user satisfaction rate was 95%, and the power plant revenue increased by 21%. All three main indicators met expectations.

[0160] It should be further explained that, in the specific implementation process, the process of optimally allocating the real-time control command sequence issued by the station-level optimization scheduling layer to each individual charging pile based on the built-in charging demand classification strategy includes:

[0161] The system obtains the total power limit command for the charging pile cluster based on the received real-time control command sequence, and acquires individual pile data based on the full operational data of the distributed execution layer at the photovoltaic-storage-charging station equipment level. The individual pile data includes the demand level and the user's required power. Rated power of a single pile The requirements include vehicle SOC, charging cutoff time, etc., with the following levels: Level 1 demand: Emergency charging (commercial / emergency vehicles), where users set a specific charging cutoff time, and the charging power is guaranteed in full with priority. It does not participate in load adjustment and is a rigid constraint with the highest priority (cannot be lowered); Level 2 demand: Normal charging (private cars), where there is no specific charging cutoff time, and the charging power can be adjusted within a set range, with limited adjustment flexibility, and a medium priority (limited downward adjustment); Level 3 demand: Scheduled charging, which can be completely shifted to off-peak hours, and the power can be adjusted in the full range from 0 to rated power, with the lowest maximum adjustment flexibility (can be paused / shifted). The total adjustable power limit is obtained by summing the user demand power of all charging piles with a demand level of Level 1 demand and subtracting the sum of the user demand power of all charging piles with Level 1 demand from the total power limit. The constraints are set according to the total power limit instruction of the charging pile cluster and the demand level.

[0162] A power allocation objective function is constructed, several power allocation schemes are preset, chromosome encoding and population initialization are performed on the power allocation schemes to generate an initial population, and the optimal power allocation scheme (the optimal power allocation for each charging pile) is obtained through a genetic algorithm based on the initial population, the power allocation objective function and constraints, and power allocation control is executed based on the optimal power allocation scheme.

[0163] It should be further noted that the optimal power allocation scheme of the genetic algorithm satisfies the following constraints:

[0164] The sum of the actual output power of all charging piles must not exceed the total power limit issued; the power of a single charging pile must be fully met for Level 1 requirements. = , The actual output power of the charging pile is used, and there is no room for adjustment. If this is violated, the solution will be directly judged as infeasible.

[0165] Secondary requirements: ≤ ≤ ,in Minimum guaranteed power (default is no less than 20% of the rated power of a single pile to avoid frequent start-stop).

[0166] Level 3 requirement: 0≤ ≤ Charging can be paused;

[0167] The output power of a single pile must not exceed its rated power, nor be lower than the minimum allowable output power of the equipment, in order to avoid damage to the equipment.

[0168] The power allocation objective function is as follows:

[0169] ;

[0170] in, , , The charging demand fulfillment rates for single charging piles at levels one, two, and three are respectively. , , , The number of charging stations for Level 1, 2, and 3 demand levels are respectively. = + + ,

[0171] The specific process of obtaining the optimal power allocation scheme using a genetic algorithm based on the initial population, the power allocation objective function, and constraints is as follows:

[0172] The power allocation scheme is encoded as chromosomes. For example, each gene position uses a real number ki in the interval [0,1] to represent the power allocation coefficient. A certain number of chromosomes are randomly generated to form an initial population. Each chromosome represents a possible power allocation scheme. The power allocation schemes are constrained and verified based on constraints, and those that do not meet the constraints are eliminated. Then, a tournament selection method is used to select chromosomes with higher fitness values ​​from the current population as parents. The parent chromosomes are crossbred, exchanging some genes to generate new offspring chromosomes. This simulates the exchange of genes in biological genetics to generate new power allocation schemes. The offspring chromosomes are mutated, randomly changing some genes to increase population diversity and avoid getting trapped in local optima. The above steps are repeated iteratively until the termination condition is met, such as reaching the maximum number of iterations or the fitness value no longer significantly improving. The optimal power allocation scheme is then output. The genetic algorithm realizes the optimal allocation of total power commands among charging piles, maximizing the satisfaction rate of user charging needs and minimizing the impact on user charging experience while meeting the station-level total power constraints.

[0173] The above embodiments are only used to illustrate the technical methods of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical methods of the present invention without departing from the spirit and scope of the technical methods of the present invention.

Claims

1. A demand response based optical storage and charging station participating in grid peak shaving and filling control system, characterized in that, This includes a control center, which is equipped with a power grid demand response interaction layer, a station-level optimized scheduling layer, and a device-level distributed execution layer. The grid demand response interaction layer is used to receive grid demand response signals, report adjustable capacity data of photovoltaic, energy storage and charging stations and response feedback data; The station-level optimized scheduling layer includes a source-load prediction module, a power dispatch estimation module, a full-scale sensing module, and a real-time rolling correction module; The source-load prediction module is used to collect the full operation data of the equipment-level distributed execution layer of each photovoltaic storage and charging station and the demand response signal of the grid connection point in real time, construct the source-load prediction model of the photovoltaic storage and charging station, and output the source-load prediction curve within the long-scale prediction period. The power dispatching prediction module is used to improve the dual-objective genetic algorithm by introducing a dynamic weight adjustment mechanism. Based on the source-load prediction curve, full operation data and demand response signal, the improved dual-objective genetic algorithm is used to predict power dispatching and obtain the optimal power dispatching plan. The dynamic weight adjustment mechanism is used to adjust the weight coefficients of the bi-objective optimization function and the global objective function during power dispatch forecasting and real-time rolling correction, including: A pre-defined quantitative standard library is provided, which includes three types of triggering dimensions and the judgment criteria for different priority levels corresponding to the three types of triggering dimensions. The three types of triggering dimensions include the urgency of the power grid demand response command, the power station operation status, and the level and concentration of user charging demand. When a real-time peak shaving demand instruction is generated in the station-level fluctuation report or demand response signal, the full operation data of the distributed execution layer of the photovoltaic storage and charging station equipment level and the demand response signal of the grid connection point are obtained in the current collection period. The full operation data and demand response signal are input into the quantitative standard library for comparison, and the priority level corresponding to the three trigger dimensions is obtained. The priority level corresponding to the three trigger dimensions is marked as the trigger scenario. A pre-defined weight rule library is used, which includes weight adjustment rules for the dual-objective optimization function and the global objective function corresponding to different triggering scenarios. The weight adjustment rules for the dual-objective optimization function and the global objective function are obtained according to the triggering scenario and the weight rule library. The initial regular weights of the dual-objective optimization function and the global objective function are adjusted according to the weight adjustment rules, and then the corresponding optimal power dispatch plan and optimal real-time control instruction sequence are regenerated. And it performs self-learning optimization of weight execution and effect feedback based on response feedback data; The full-scale perception module is used to perceive the full status and demand response of each optical storage and charging station, and determines whether to generate a station-level fluctuation report and update the source-load prediction curve based on the perception results. The real-time rolling correction module is used to perform real-time rolling correction of the optimal power dispatch plan based on station-level fluctuation reports or demand response signals, and to obtain real-time control command sequences. The device-level distributed execution layer is used to optimally distribute the real-time control command sequence issued by the station-level optimized scheduling layer to each individual pile based on the built-in charging demand classification strategy.

2. The demand response based optical storage and charging station participating in grid peak load shifting control system according to claim 1, characterized in that, The power grid demand response interaction layer includes a demand response signal receiving module, a photovoltaic, energy storage, and charging station adjustable capacity reporting module, and a response effect feedback module. The demand response signal receiving module is used to receive daytime peak and valley period and time-of-use electricity price signals, intraday peak warning signals, real-time peak shaving demand instructions, ancillary service market signals, and power grid safety constraint limits issued by the power grid. The adjustable capacity reporting module for photovoltaic-storage-charging stations is connected to each photovoltaic-storage-charging station in the target area to report to the power grid the photovoltaic expected output curve, energy storage charging and discharging adjustable range, adjustable upper and lower limits of the charging pile cluster, day-ahead peak shaving and valley filling response plan, and intraday adjustable margin of each photovoltaic-storage-charging station. The response effect feedback module communicates with each photovoltaic, energy storage and charging station in the target area to collect statistics and report to the power grid the peak load reduction, valley renewable energy consumption increase, demand response command tracking deviation, and peak shaving service completion rate of each photovoltaic, energy storage and charging station.

3. The demand response based optical storage and charging station participating in grid peak load shifting control system according to claim 2, characterized in that, The process of outputting the source load prediction curve over a long-scale prediction period includes: Real-time collection of full operational data, environmental data, and demand response signals from grid connection points of each optical storage and charging station within the target area at the equipment-level distributed execution layer, marking the collection time, and setting the collection cycle; The full operational data, environmental data, and demand response signals of grid-connected points of each optical storage and charging station within several historical acquisition periods in the target area are obtained as training and test sets. A source-load prediction model for the optical storage and charging station is constructed. The training set is input into the source-load prediction model for the optical storage and charging station for training until the loss function is stable and the model parameters are saved. The source-load prediction model for the optical storage and charging station is tested through the test set until it meets the preset requirements, and then the source-load prediction model for the optical storage and charging station is output. A long-scale prediction period is set, and the selected timestamp of the long-scale prediction period is used as the prediction time point. At the prediction time point, weather forecast data, user charging reservation data, and demand response signals for the next long-scale prediction period of the photovoltaic-storage-charging station are obtained. The weather forecast data, user charging reservation data, and demand response signals for the next long-scale prediction period are input into the photovoltaic-storage-charging station source-load prediction model. Based on the photovoltaic-storage-charging station source-load prediction model, the photovoltaic output prediction curve and the spatiotemporal distribution prediction curve of charging demand for the next long-scale prediction period are output.

4. The demand response based optical storage and charging station participating in grid peak load shifting control system according to claim 3, characterized in that, The process of obtaining the optimal power dispatch plan includes: Based on the photovoltaic output prediction curve and charging demand spatiotemporal distribution prediction curve of the photovoltaic-storage-charging station in the next long-scale prediction period, the full operation data of the equipment-level distributed execution layer and the demand response signal of the grid connection point, several correlation quantization functions and constraints are constructed. Based on the correlation quantization functions, a bi-objective optimization function is constructed. The initial conventional weights of the bi-objective optimization function are set. Several power dispatch plans are preset. The power dispatch plans are chromosome-encoded and the population is initialized to generate the initial population. A dynamic weight adjustment mechanism is introduced to improve the bi-objective genetic algorithm. Based on the initial population, the bi-objective optimization function, and the constraints, the optimal power dispatch plan is obtained through an improved bi-objective genetic algorithm. The optimal power dispatch plan includes the planned values ​​of energy storage charging and discharging power of the photovoltaic-storage-charging station in each acquisition cycle within the next long-scale prediction cycle, the baseline planned value of the power purchased from the grid by the grid connection point, and the planned charging power and adjustable capacity boundary of the charging pile cluster.

5. The demand-response-based photovoltaic-storage-charging station participation in grid peak shaving and valley filling control system according to claim 4, characterized in that, The constraints include grid-side constraints, equipment-side constraints, safety protection constraints, and power balance constraints. Among them, the condition for satisfying the power balance constraint is that the planned value of the energy storage charging and discharging power of the photovoltaic-storage-charging station is equal to the baseline planned value of the power purchased from the grid by the grid connection point. This equals the planned charging power of the charging pile cluster.

6. The demand-response-based photovoltaic-storage-charging station participation in grid peak shaving and valley filling control system according to claim 5, characterized in that, The process of achieving full status and demand response awareness for each photovoltaic storage and charging station includes: The system compares the full operational data of the equipment-level distributed execution layer of the photovoltaic, energy storage, and charging stations collected in real time during the acquisition period with the photovoltaic output prediction curve and the spatiotemporal distribution prediction curve of charging demand during the acquisition period. It obtains the photovoltaic output deviation time series and the charging demand spatiotemporal distribution deviation time series, presets deviation thresholds and time thresholds, and obtains the cumulative time in the photovoltaic output deviation time series and the charging demand spatiotemporal distribution deviation time series that exceeds the deviation threshold. If the cumulative time corresponding to either the photovoltaic output deviation time series or the charging demand spatiotemporal distribution deviation time series exceeds the time threshold, a station-level fluctuation report is generated. The system also inputs the full operational data of the equipment-level distributed execution layer of the photovoltaic, energy storage, and charging stations collected in real time during the acquisition period, environmental data, and the demand response signal from the grid connection point into the photovoltaic, energy storage, and charging station source-load prediction model. A first time scale is set, and the photovoltaic output prediction curve and the charging demand spatiotemporal distribution prediction curve for the next segment of the first time scale are output based on the photovoltaic output prediction curve and the charging demand spatiotemporal distribution prediction curve within the first time scale. Finally, the photovoltaic output prediction curve and the charging demand spatiotemporal distribution prediction curve within the acquisition period covered by the first time scale are updated based on the photovoltaic output prediction curve and the charging demand spatiotemporal distribution prediction curve within the first time scale. Similarly, if a real-time peak-shaving demand instruction is generated in the demand response signal, the full amount of operating data, environmental data, and demand response signals of the grid-connected points of the photovoltaic, energy storage, and charging stations collected in real time during the collection period will be input into the photovoltaic, energy storage, and charging station source-load prediction model. Based on the photovoltaic output prediction curve and the spatiotemporal distribution prediction curve of charging demand in the first time scale output by the photovoltaic, energy storage, and charging station source-load prediction model, the photovoltaic output prediction curve and the spatiotemporal distribution prediction curve of charging demand in the collection period covered by the first time scale will be updated.

7. The demand-response-based photovoltaic-storage-charging station participation in grid peak shaving and valley filling control system according to claim 6, characterized in that, The process of obtaining a real-time control command sequence includes: When a real-time peak-shaving demand instruction is generated in the station-level fluctuation report or demand response signal, a global objective function is constructed based on the photovoltaic output prediction curve, the spatiotemporal distribution prediction curve of charging demand, the full operation data of the equipment-level distributed execution layer in the current collection period, the demand response signal of the grid connection point, and the optimal power dispatch plan of the photovoltaic, energy storage and charging station in the first time scale. The initial conventional weights of the global objective function are set, several real-time control instruction sequences are preset, chromosome encoding and population initialization are performed on the real-time control instruction sequences, an initial population is generated, and a dynamic weight adjustment mechanism is introduced to improve the genetic algorithm. Based on the initial population, global objective function, and adjustable capability boundary, an improved genetic algorithm is used to obtain the optimal real-time control command sequence within the current acquisition cycle.

8. The demand-response-based photovoltaic-storage-charging station participation grid peak shaving and valley filling control system according to claim 7, characterized in that, The self-learning optimization process based on the data collected by the response effect feedback module for weight execution and effect feedback includes: The corresponding optimal power dispatch plan and optimal real-time control command sequence will be regenerated and sent to the device-level distributed execution layer for execution. Simultaneously, the peak load reduction, valley renewable energy consumption improvement, demand response command tracking deviation and peak shaving service completion rate will be obtained from the post-execution response effect feedback module. The system presets expected threshold ranges for peak load reduction, valley renewable energy consumption improvement, demand response instruction tracking deviation, and peak shaving service completion rate. If any of these values ​​are outside the corresponding expected threshold ranges, a fishbone diagram is constructed for root cause analysis and control variable simulation verification. This optimizes the weight adjustment range in the weight rule library and the grading thresholds in the quantification standard library.

9. The demand-response-based photovoltaic-storage-charging station participation in grid peak shaving and valley filling control system according to claim 8, characterized in that, Based on the built-in charging demand tiering strategy, the process of optimally allocating the real-time control command sequence issued by the station-level optimized scheduling layer to each individual charging pile includes: The total power limit instruction of the charging pile cluster is obtained according to the received real-time control instruction sequence. The single pile data is obtained according to the full operation data of the distributed execution layer of the photovoltaic energy storage charging station equipment level. The single pile data includes the demand level, user demand power and single pile rated power. The sum of user demand power of all charging piles with demand level of level 1 is calculated. The sum of user demand power of all charging piles with level 1 demand is subtracted from the total power limit to obtain the total adjustable power limit. Constraints are set according to the total power limit instruction of the charging pile cluster and the demand level. A power allocation objective function is constructed, several power allocation schemes are preset, chromosome encoding and population initialization are performed on the power allocation schemes to generate an initial population, and the optimal power allocation scheme is obtained through a genetic algorithm based on the initial population, the power allocation objective function and constraints, and power allocation control is executed based on the optimal power allocation scheme.

Citation Information

Patent Citations

  • Intelligent scheduling system of photovoltaic energy storage charging station

    CN121566644A