Multi-source coordinated charging and discharging strategy optimization method and system for energy storage system

Abstract

This invention provides a method and system for optimizing multi-source coordinated charging and discharging strategies for energy storage systems. The method includes: predicting electricity prices based on acquired multi-source historical data to obtain predicted electricity prices; outputting an optimized load curve with the objective of minimizing the weighted sum of maximum demand power and electricity cost within a scheduling cycle; using the optimized load curve as a consumption benchmark, determining the existence of curtailment risk and identifying a dynamic operating range when the predicted photovoltaic power generation exceeds the sum of the consumption benchmark and the current rechargeable capacity of the energy storage system; constructing a scheduling curve model based on the predicted electricity prices to maximize net revenue; constructing constraints for the scheduling curve model based on the optimized load curve and the dynamic operating range; and solving the scheduling curve model based on the constraints to obtain a charging and discharging scheduling curve and control the operation of the energy storage system. This invention improves energy utilization efficiency and reduces electricity costs.

Description

Technical Field

[0001] This invention relates to the field of energy storage system scheduling technology, and specifically to a method and system for optimizing multi-source coordinated charging and discharging strategies for energy storage systems. Background Technology

[0002] With the widespread application of distributed photovoltaic and energy storage systems in industrial parks, how to minimize electricity costs and efficiently absorb new energy sources through multi-source synergistic optimization has become an important issue in the field of energy management. Current mainstream energy management strategies typically include multiple functional modules such as electricity price forecasting, load regulation, curtailment avoidance, and energy storage charging and discharging scheduling. However, these modules are often designed in isolation or simply combined in series, lacking in-depth synergistic optimization.

[0003] Specifically, existing technologies have the following shortcomings: 1. Separation of electricity price forecasting and load dispatching: Traditional demand control methods typically shift load power based on fixed or day-ahead electricity prices, failing to dynamically integrate high-precision electricity price forecasting results into the flexible reconstruction process of the load curve, thus limiting peak-valley arbitrage effects. 2. Static binding of curtailment avoidance and energy storage operating range: Most photovoltaic-storage systems use fixed upper and lower limits of state of charge (SoC), making it impossible to dynamically adjust the rechargeable capacity of energy storage according to real-time photovoltaic over-generation risks, often resulting in curtailment where there is electricity available for charging but no capacity to accept it. Although a few methods introduce reserved capacity strategies, they do not link the curtailment risk index with the optimized load absorption benchmark, resulting in insufficient accuracy in reserved decision-making. 3. Lack of multi-constraint closed-loop charging and discharging dispatching: Existing energy storage dispatching models often only consider price differences and basic energy storage constraints, rarely introducing load optimization curves and dynamic SoC operating ranges as constraints simultaneously, leading to a disconnect between the dispatching curve and the actual electricity consumption characteristics of the park, making it difficult to achieve global optimization of economic objectives.

[0004] Therefore, there is an urgent need for a multi-source coordinated charging and discharging strategy optimization method and system for energy storage systems, which can perform end-to-end coordinated optimization of electricity price forecasting, flexible load regulation, power curtailment risk early warning and energy storage charging and discharging scheduling, so as to improve energy utilization and reduce electricity costs. Summary of the Invention

[0005] In view of this, it is necessary to provide a multi-source coordinated charging and discharging strategy optimization method and system for energy storage systems, in order to solve the technical problems in the existing technology that do not perform end-to-end coordinated optimization of electricity price prediction, flexible load regulation, power curtailment risk warning and energy storage charging and discharging scheduling, resulting in low energy utilization and high electricity costs.

[0006] To address the aforementioned technical problems, in a first aspect, the present invention provides a method for optimizing multi-source coordinated charging and discharging strategies for energy storage systems, comprising: Acquire multi-source historical data of the area where the energy storage system is located, and predict the electricity price at multiple future times based on the multi-source historical data to obtain the predicted electricity price value; Based on the predicted electricity price, high-price and low-price periods are identified. Industrial loads are divided into rigid loads and flexible loads according to their adjustability. A time window for allowing power shifting is set for flexible loads. Within the time window, part of the flexible load power in the high-price period is transferred to the low-price period. The goal is to minimize the weighted sum of the maximum demand power and electricity cost within the scheduling cycle, and the optimized load curve is output. The photovoltaic power generation forecast value is obtained, and the optimized load curve is used as the absorption benchmark. When the photovoltaic power generation forecast value exceeds the sum of the absorption benchmark and the current rechargeable capacity of the energy storage system, it is determined that there is a curtailment risk, and a curtailment risk index is determined. Based on the curtailment risk index and the current state of charge of the energy storage system, a dynamic operating range is determined from multiple preset state of charge operating ranges. Based on the predicted electricity price, a scheduling curve model is constructed with the goal of maximizing net revenue. Based on the optimized load curve and the dynamic operating range, constraints are constructed for the scheduling curve model. The scheduling curve model is then solved based on the constraints to obtain the charging and discharging scheduling curve. The operation of the energy storage system is controlled based on the charge and discharge scheduling curve.

[0007] In one possible implementation, the step of predicting electricity prices for multiple future times based on the multi-source historical data to obtain predicted electricity prices includes: A multi-source feature vector is constructed based on the multi-source historical data. The multi-source feature vector includes electricity price time-series features, temperature features, and interaction features between date type and temperature. The multi-source feature vector is input into the electricity price prediction model, which includes a feature extraction module, a hybrid attention module, and a prediction module. The hybrid attention module includes a Flash Attention unit and a temporal convolutional attention unit. Based on the feature extraction module, nonlinear transformation and feature filtering are performed on the multi-source feature vectors to generate enhanced feature vectors; The global dependency features of the enhanced feature vector are extracted based on the Flash Attention unit, and the local temporal pattern features of the enhanced feature vector are extracted based on the temporal convolutional attention unit. The global dependency features and the local temporal pattern features are fused to obtain fused features; The fused features are input into the prediction module to obtain electricity price predictions for multiple future time points.

[0008] In one possible implementation, the loss function used during training of the electricity price prediction model is:

[0009] In the formula, For multiple prediction times; Huber loss function; These are the actual electricity price and the predicted electricity price at the i-th prediction time, respectively; This represents the predicted electricity price at time i+1. This is the smoothing coefficient.

[0010] In one possible implementation, the step of outputting an optimized load curve with the objective of minimizing the weighted sum of maximum power demand and electricity cost within a scheduling period includes: The objective function is constructed by minimizing the weighted sum of the maximum power demand and electricity cost within the scheduling period. The objective function is as follows: ; ; Construct the objective constraints for the objective function. The objective constraints include power balance constraints, grid capacity constraints, and flexible load operation constraints. The flexible load operation constraints include total demand fulfillment constraints, shift time window constraints, and power regulation rate limit constraints. The power balance constraint is: ; The power grid capacity constraint is: ; The total demand fulfillment constraint is as follows: ; The translation time window constraint is: ; The power regulation rate limit constraint is: ; The objective function is solved based on the objective constraints, and the optimized load curve is output. In the formula, Let be the total power consumption at time t; Power for rigid loads; The actual operating power of the k-th type of flexible load at time t; This represents the actual operating power of the k-th type of flexible load at time t-1. Based on the standard commissioning ratio; Reference power; This is a projected electricity price. , , These are the weighting coefficients; Power purchased from the power grid; The upper limit of the power purchase capacity of the power grid; The maximum power change rate of the k-th type of flexible load; , These are the start and end times of the time window, respectively. The maximum allowable translation duration; Photovoltaic power generation; This refers to the charging and discharging power of the energy storage system.

[0011] In one possible implementation, obtaining the photovoltaic power generation prediction value includes: A photovoltaic power generation prediction model is constructed. The input of the photovoltaic power generation prediction model includes the photovoltaic panel attitude, and the output is the photovoltaic power generation prediction value. With the goal of minimizing the weighted sum of the risk of curtailment and the cost of attitude adjustment, the attitude of the photovoltaic panels in the future is optimized to obtain the optimized attitude; The optimized posture is input into the photovoltaic power generation prediction model to obtain the photovoltaic power generation prediction value.

[0012] In one possible implementation, the power curtailment risk indicator is:

[0013] In the formula, As an indicator of power curtailment risk; To optimize the load curve; For charging efficiency; To the theoretical maximum photovoltaic output; This represents the maximum charging power of the energy storage system.

[0014] In one possible implementation, the scheduling curve model is as follows:

[0015] In the formula, For charging efficiency; For discharge efficiency; Let be the predicted electricity price at time t; Let be the charging power at time t; Let be the discharge power at time t; Peak penalty weight; This is the power smoothing coefficient; Power purchased from the power grid The peak penalty item.

[0016] In one possible implementation, the constraints include energy storage dynamic constraints, capacity constraints, power constraints, time-period constraints, and economic constraints. The energy storage dynamic constraints are:

[0017] The capacity constraint is:

[0018] The power constraint is:

[0019] The time period constraint is as follows:

[0020]

[0021] The economic constraints are as follows:

[0022] In the formula, The stored energy amount at time t+1; Let be the stored energy quantity at time t; For time intervals; Maximum energy storage capacity; Minimum energy storage capacity; This is the maximum charging power; The maximum discharge power; t For a specified time; The minimum discharge power at a specified time; The discharge power at a specified time; This is a set of charging periods.

[0023] In one possible implementation, after obtaining the charge / discharge scheduling curve, the method further includes: According to the preset rolling optimization cycle, the model prediction control method based on the scenario tree is adopted to correct the charging and discharging scheduling curve based on the current actual energy storage capacity, electricity price probability prediction and photovoltaic probability prediction, so as to obtain the optimized scheduling curve. The variance of the electricity price prediction error is calculated in real time. When the variance exceeds a preset threshold, a pre-trained deep Q-network is activated. The deep Q-network outputs discrete charging and discharging action commands and maps these commands to real-time charging and discharging power commands. The real-time charging and discharging power commands are then weighted and summed with the charging and discharging power commands of the optimized scheduling curve at the current moment to obtain the target charging and discharging power command. The weight of the real-time charging and discharging power command increases linearly with the increase of the variance of the electricity price prediction error. When the variance of the electricity price prediction error is less than or equal to the preset threshold, the charging and discharging power command of the optimized scheduling curve at the current moment is used as the target charging and discharging power command.

[0024] Secondly, the present invention also provides a multi-source coordinated charging and discharging strategy optimization system for energy storage systems, comprising: The electricity price prediction unit is used to acquire multi-source historical data of the area where the energy storage system is located, and to predict the electricity price at multiple future times based on the multi-source historical data to obtain the predicted electricity price value. The load curve optimization unit is used to identify high-electricity-price periods and low-electricity-price periods based on the electricity price forecast, classify industrial loads into rigid loads and flexible loads according to their adjustability, and set a time window for allowing power shifting for flexible loads. Within the time window, a portion of the flexible load power in the high-electricity-price period is transferred to the low-electricity-price period for execution, with the goal of minimizing the weighted sum of the maximum demand power and electricity cost within the scheduling cycle, and outputting the optimized load curve. The dynamic operating range determination unit is used to obtain the photovoltaic power generation prediction value, and use the optimized load curve as the absorption benchmark. When the photovoltaic power generation prediction value exceeds the sum of the absorption benchmark and the current rechargeable capacity of the energy storage system, it is determined that there is a curtailment risk, and a curtailment risk index is determined. Based on the curtailment risk index and the current state of charge of the energy storage system, the dynamic operating range is determined from multiple preset state of charge operating ranges. The charging and discharging scheduling curve determination unit is used to construct a scheduling curve model with the goal of maximizing net revenue based on the electricity price forecast, construct the constraints of the scheduling curve model based on the optimized load curve and the dynamic operating range, and solve the scheduling curve model based on the constraints to obtain the charging and discharging scheduling curve. A control unit is used to control the operation of the energy storage system based on the charge-discharge scheduling curve.

[0025] The beneficial effects of this invention are as follows: The multi-source coordinated charging and discharging strategy optimization method for energy storage systems provided by this invention achieves coordinated peak shaving and valley filling between the demand and supply sides in the time dimension by simultaneously using electricity price forecasts to transfer flexible loads from high-price periods to low-price periods and to construct the scheduling curve model. Compared to independently executing load control or energy storage charging and discharging, this invention can make unified decisions under the same objective function of minimizing the weighted sum of maximum demand power and electricity cost within the scheduling cycle, avoiding conflicts between sub-objectives. Actual test data shows that the overall electricity cost can be reduced by more than 12%, and the peak demand can be reduced by about 25%.

[0026] Furthermore, this invention uses the optimized load curve as a benchmark for photovoltaic (PV) power consumption and introduces a curtailment risk index and the current state of charge (SOC) of energy storage to jointly determine the dynamic operating range. Compared with the traditional method of fixing the upper and lower limits of SoC, this invention can proactively expand the energy storage acceptance window or lower the lower limit to reserve more charging space when the risk of PV over-generation is predicted, while avoiding excessive reservation that would reduce the peak-shaving flexibility of energy storage, thus improving the PV power consumption rate.

[0027] Furthermore, the scheduling curve model constructed in this invention with the goal of maximizing net revenue incorporates constraints from load optimization curves and dynamic operating ranges. This ensures that charging and discharging decisions are not only based on electricity price forecasts but also jointly constrained by actual electricity consumption patterns and the risk of power curtailment. Compared to traditional models that only consider price differences and basic energy storage constraints, this model enables optimal decisions on charging timing and power, maximizing peak-valley arbitrage profits. Attached Figure Description

[0028] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0029] Figure 1 A schematic flowchart of an embodiment of the multi-source coordinated charging and discharging strategy optimization method for energy storage systems provided by the present invention; Figure 2 For the present invention Figure 1 A schematic diagram of an embodiment of step S101; Figure 3 This is a schematic diagram of an embodiment of the present invention for obtaining photovoltaic power generation prediction values; Figure 4 This is a schematic diagram of an embodiment of the multi-source coordinated charging and discharging strategy optimization system for energy storage systems provided by the present invention. Detailed Implementation

[0030] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0031] It should be understood that the illustrative drawings are not drawn to scale. The flowcharts used in this invention illustrate operations implemented according to some embodiments of the invention. It should be understood that the operations in the flowcharts may be implemented out of order, and steps without logical contextual relationships may be reversed or performed simultaneously. Furthermore, those skilled in the art, guided by the content of this invention, may add one or more other operations to the flowcharts, or remove one or more operations from the flowcharts. Some block diagrams shown in the drawings are functional entities and do not necessarily correspond to physically or logically independent entities. These functional entities may be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor systems and / or microcontroller systems.

[0032] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of the invention. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a mutually exclusive, independent, or alternative embodiment. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0033] This invention provides a method and system for optimizing multi-source coordinated charging and discharging strategies for energy storage systems, which will be described below.

[0034] Figure 1 This is a schematic flowchart of an embodiment of the multi-source coordinated charging and discharging strategy optimization method for energy storage systems provided by the present invention, as shown below. Figure 1 As shown, the optimization method for multi-source coordinated charging and discharging strategies for energy storage systems includes: S101. Obtain multi-source historical data of the area where the energy storage system is located, and predict the electricity price at multiple future moments based on the multi-source historical data to obtain the predicted electricity price value.

[0035] Among them, multi-source historical data includes, but is not limited to, historical electricity price series, temperature data, date type (weekday / holiday), solar irradiance, and historical load power.

[0036] It should be understood that existing time-series forecasting neural network models can be used for electricity price forecasting.

[0037] S102. Based on the electricity price forecast, identify the high electricity price period and the low electricity price period. Divide the industrial load into rigid load and flexible load according to the adjustability. Set a time window for the flexible load to allow power shifting. Within the time window, transfer part of the flexible load power in the high electricity price period to the low electricity price period. With the goal of minimizing the weighted sum of the maximum demand power and electricity cost in the scheduling cycle, output the optimized load curve.

[0038] Rigid loads refer to loads that are not adjustable or must operate strictly according to time periods, such as continuous production lines. Flexible loads refer to loads that allow for shifting or power adjustment within a time window, such as charging stations, air conditioners, and interruptible production processes.

[0039] Specifically, by shifting some flexible load power from high-price periods to low-price periods within a time window, instantaneous peak electricity consumption can be reduced, thus achieving peak reduction.

[0040] S103. Obtain the photovoltaic power generation forecast value, use the optimized load curve as the consumption benchmark, and when the photovoltaic power generation forecast value exceeds the sum of the consumption benchmark and the current rechargeable capacity of the energy storage system, determine that there is a curtailment risk and determine the curtailment risk index. Based on the curtailment risk index and the current state of charge of the energy storage system, determine the dynamic operating range from multiple preset state of charge operating ranges. S104. Construct a scheduling curve model with the goal of maximizing net revenue based on the electricity price forecast. Construct the constraints of the scheduling curve model based on the optimized load curve and dynamic operating range. Solve the scheduling curve model based on the constraints to obtain the charging and discharging scheduling curve. S105. Control the operation of the energy storage system based on the charge and discharge scheduling curve.

[0041] It should be understood that the multi-source coordinated charging and discharging strategy optimization method for energy storage systems in this embodiment of the invention can be implemented in any energy management device based on the multi-source coordinated charging and discharging strategy optimization method for energy storage systems, such as industrial park energy storage dispatch controllers, photovoltaic-storage-charging integrated energy management systems, microgrid central controllers, virtual power plant aggregation and control units, and other application devices. Specifically, the multi-source coordinated charging and discharging strategy optimization method for energy storage systems is stored in the aforementioned device as a pre-programmed program. When the device starts up, the program is invoked, and the multi-source coordinated charging and discharging strategy optimization method for energy storage systems is implemented.

[0042] Compared with existing technologies, the multi-source coordinated charging and discharging strategy optimization method for energy storage systems provided in this invention achieves coordinated peak shaving and valley filling between the demand and supply sides in the time dimension by simultaneously using electricity price forecasts to transfer flexible loads from high-price periods to low-price periods and to construct the scheduling curve model. Compared with independently executing load control or energy storage charging and discharging, this invention can make unified decisions under the same objective function of minimizing the weighted sum of maximum demand power and electricity cost within the scheduling cycle, avoiding conflicts between sub-objectives. Actual test data shows that the overall electricity cost can be reduced by more than 12%, and the peak demand can be reduced by about 25%.

[0043] Furthermore, this embodiment of the invention uses the optimized load curve as the benchmark for photovoltaic (PV) power consumption and introduces a curtailment risk index and the current state of charge of energy storage to jointly determine the dynamic operating range. Compared with the traditional method of fixing the upper and lower limits of SoC (Power Consumption Scale), this method can proactively expand the energy storage acceptance window or lower the lower limit to reserve more charging space when the risk of PV over-generation is predicted, while avoiding excessive reservation that would reduce the peak-shaving flexibility of energy storage and improve the PV power consumption rate.

[0044] Furthermore, the scheduling curve model constructed in this embodiment of the invention, with the goal of maximizing net revenue, incorporates constraints from load optimization curves and dynamic operating ranges. This ensures that charging and discharging decisions are not only based on electricity price forecasts but also subject to the joint constraints of actual electricity consumption patterns and the risk of power curtailment. Compared to traditional models that only consider electricity price differences and basic energy storage constraints, this model can achieve optimal decisions on charging timing and power, maximizing peak-valley arbitrage profits.

[0045] Since electricity price forecasting is the foundation for subsequent steps, but electricity prices are affected by a variety of dynamic factors, these factors result in electricity prices having high real-time and strong nonlinear characteristics. Traditional electricity price forecasting methods, which rely on a single time series forecasting model (such as ARIMA) or a simple machine learning model, are unable to effectively capture the deep correlations and long-term dependencies of multi-source heterogeneous data, resulting in insufficient forecast accuracy and affecting the effectiveness of peak-shaving scheduling.

[0046] Therefore, in a preferred embodiment of the present invention, such as Figure 2 As shown, step S101 includes: S201. Construct a multi-source feature vector based on multi-source historical data. The multi-source feature vector includes electricity price time series features, temperature features, and interaction features between date type and temperature.

[0047] It should be noted that, in order to improve data quality, the multi-source historical data is standardized before constructing the multi-source feature vector.

[0048] Among them, the time series characteristics of electricity price can be determined by calculating moving average, difference series and periodic characteristics based on the electricity price series; the temperature characteristics are determined by constructing a nonlinear mapping characteristic between temperature data and electricity price and transforming it through a kernel function; and the interaction characteristics between date type and temperature can supplement the temperature sensitivity under different dates.

[0049] In a specific embodiment of the present invention, multi-source feature vectors for:

[0050]

[0051] In the formula, The electricity price for the 24 hours preceding the current moment; The electricity price is the price for the hour preceding the current time. This is the 7-day moving average of electricity prices; The temperature at the current moment; Date type; Solar irradiance.

[0052] S202. Input the multi-source feature vector into the electricity price prediction model. The electricity price prediction model includes a feature extraction module, a hybrid attention module, and a prediction module. The hybrid attention module includes a Flash Attention unit and a temporal convolutional attention unit. S203. Based on the feature extraction module, perform nonlinear transformation and feature selection on the multi-source feature vectors to generate enhanced feature vectors.

[0053] In a specific embodiment of the present invention, the feature extraction module is a Gradient Boosting Decision Tree (GBDT). The GBDT performs importance filtering and non-linear feature extraction on features to generate enhanced feature vectors. It can be represented as:

[0054] In the formula, This indicates a vector concatenation operation.

[0055] S204. Extract global dependency features of the enhanced feature vector based on the Flash Attention unit, and extract local temporal pattern features of the enhanced feature vector based on the temporal convolutional attention unit.

[0056] Among them, for time series composed of enhanced feature vectors The global dependency feature is:

[0057] In the formula, , , This is the query, key, and value matrix after linear projection. For dimensions.

[0058] Temporal Convolutional Attention Module: Extracts local temporal patterns through one-dimensional convolution and generates attention weights to enhance sensitivity to recent electricity price fluctuations. The local temporal pattern features are as follows:

[0059] In the formula, This indicates element-wise multiplication.

[0060] S205. The global dependency features and local temporal pattern features are fused to obtain the fused features.

[0061] Among them, fusion features for:

[0062] In the formula, This is a layer normalization operation.

[0063] S206. Input the fusion features into the prediction module to obtain the predicted electricity price values ​​for multiple future times.

[0064] Specifically, the prediction module consists of a fully connected layer and a time distribution layer.

[0065] In this embodiment of the invention, electricity prices are predicted for three future time points.

[0066] The Flash Attention unit in this embodiment of the invention can efficiently capture the long-range global dependencies of the electricity price sequence, while the temporal convolutional attention unit enhances the model's sensitivity to local patterns of recent electricity price fluctuations. The fusion of the two enables the prediction model to grasp the overall trend and respond to rapid changes, thereby significantly improving the accuracy and generalization ability of electricity price prediction at multiple future moments, and providing a more reliable decision-making basis for subsequent load power shifting and energy storage charging and discharging scheduling.

[0067] In a specific embodiment of the present invention, the loss function used during training of the electricity price prediction model is:

[0068] In the formula, For multiple prediction times; Huber loss function; These are the actual electricity price and the predicted electricity price at the i-th prediction time, respectively; This represents the predicted electricity price at time i+1. This is the smoothing coefficient.

[0069] By setting the aforementioned loss function, this invention achieves two objectives: firstly, the Huber loss is insensitive to outliers, effectively mitigating the negative impact of outliers in electricity price data on model training and enhancing prediction robustness; secondly, the time-series smoothing term penalizes drastic changes in predicted values ​​at adjacent times, ensuring the output electricity price sequence maintains continuity and stability over time. The synergistic effect of these two factors enables the trained electricity price prediction model to output smooth and reasonable prediction curves even when faced with noisy data or sudden price fluctuations, preventing subsequent scheduling strategy instability caused by large oscillations in predicted values.

[0070] In some embodiments of the present invention, step S102, which aims to minimize the weighted sum of the maximum demand power and electricity cost within the scheduling period and outputs the optimized load curve, includes: The objective function is constructed by minimizing the weighted sum of the maximum power demand and electricity cost within the scheduling period. The objective function is: ; ; The objective constraints of the objective function are constructed. The objective constraints include power balance constraints, grid capacity constraints, and flexible load operation constraints. The flexible load operation constraints include total demand fulfillment constraints, shift time window constraints, and power regulation rate limit constraints. The power balance constraint is: ; The grid capacity constraint is: ; The total demand fulfillment constraint is: ; The translation time window constraint is: ; The power regulation rate limit constraint is: ; The objective function is solved based on the objective constraints, and the optimized load curve is output. In the formula, Let be the total power consumption at time t; Power for rigid loads; The actual operating power of the k-th type of flexible load at time t; This represents the actual operating power of the k-th type of flexible load at time t-1. Based on the standard commissioning ratio; Reference power; This is a projected electricity price. , , These are the weighting coefficients; Power purchased from the power grid; The upper limit of the power purchase capacity of the power grid; The maximum power change rate of the k-th type of flexible load; , These are the start and end times of the time window, respectively. The maximum allowable translation duration; Photovoltaic power generation; This refers to the charging and discharging power of the energy storage system.

[0071] This invention constructs an objective function that minimizes the weighted sum of maximum demand power and electricity cost within a scheduling cycle. The objective function incorporates a scheduling deviation penalty term for flexible load power shifting, achieving a dynamic balance between demand shaving and electricity cost reduction, thus avoiding extreme scheduling caused by solely pursuing one objective. Simultaneously, the established power balance constraints, grid capacity constraints, and flexible load operation constraints jointly ensure that the optimized load curve, while meeting the total electricity demand of production processes, can safely and smoothly transfer flexible load power from high-electricity-price periods to low-electricity-price periods without exceeding the grid's power purchase limit or causing frequent equipment start-ups and shutdowns. This technical solution makes the load curve reconstruction both economical and engineering feasible, providing a clear and executable absorption benchmark for subsequent energy storage system charging and discharging scheduling.

[0072] Since the objective function is a high-dimensional, non-convex combinatorial optimization problem with constraints, in order to improve the solution efficiency and accuracy, in some embodiments of the present invention, a hybrid strategy of adaptive dynamic programming (ADP) and improved particle swarm optimization (PSO) is used to solve the objective function.

[0073] Specifically, based on the adaptive dynamic programming algorithm, the multi-time period problem is decomposed into a series of single-time period sub-problems, and the value function... Defined as from state (Including unused power and current power of each load) The lowest cost from start to end of the cycle.

[0074] The recursive formula is:

[0075] In the formula, As decision variables, The instantaneous cost for time period t (including peak penalty and electricity cost).

[0076] For each time period An improved particle swarm optimization algorithm is used for efficient search: Each particle represents a possible load distribution scheme.

[0077] A Flash attention mechanism is introduced to update particle velocity and position: the learning factor is dynamically adjusted based on the particle's historical best and global best attention scores to avoid premature convergence. The update formula is as follows:

[0078]

[0079] Among them, attention weight Calculated by Flashattention, focusing on dimensions with significant differences; Let be the current velocity vector of the i-th particle; Inertial weight; , Individual learning factors and social learning factors; , It is a random number; For an individual, the best historical position; The globally optimal position; This is the updated velocity vector of the particle; The current position vector of the particle represents a candidate load allocation scheme; This is the updated position vector of the particle.

[0080] The outer ADP and inner PSO are executed iteratively until the total cost converges, obtaining the optimal load power allocation sequence, and thus the optimized load curve.

[0081] The PSO algorithm in this embodiment of the invention replaces the traditional fixed coefficient term by introducing an attention mechanism, i.e. It can dynamically identify key dimensions with significant differences and assign them higher update weights, thereby accelerating convergence and avoiding premature convergence. It is especially suitable for high-dimensional, non-convex load distribution optimization problems.

[0082] Existing photovoltaic (PV) power generation forecasting methods typically treat PV panel orientation as a fixed parameter or rely solely on historical irradiance data for static predictions, neglecting the dynamic adjustability of PV panel azimuth and tilt angles. When the predicted PV output exceeds the sum of the load absorption capacity and available energy storage capacity, failure to adjust PV panel orientation in advance to proactively reduce or shift the peak output will increase the risk of power curtailment.

[0083] To solve this technical problem, in some embodiments of the present invention, such as Figure 3 As shown, obtaining the photovoltaic power generation prediction value in step S103 includes: S301. Construct a photovoltaic power generation prediction model. The input of the photovoltaic power generation prediction model includes the photovoltaic panel attitude, and the output is the photovoltaic power generation prediction value.

[0084] Among them, the photovoltaic power generation prediction model is a deep learning network model that integrates spatiotemporal attention, which is used to provide photovoltaic power generation prediction values ​​1-4 hours in advance.

[0085] It should be noted that, in addition to the photovoltaic panel attitude, which includes the azimuth and tilt angles, the inputs to the photovoltaic power generation prediction model also include historical irradiance sequences, cloud cover, temperature, and humidity.

[0086] S302. With the goal of minimizing the weighted sum of the risk of curtailment and the cost of attitude adjustment, the attitude of the photovoltaic panels in the future is optimized to obtain the optimized attitude.

[0087] In a specific embodiment of the present invention, the objective function aimed at minimizing the weighted sum of the risk of power curtailment and the cost of attitude adjustment is as follows:

[0088]

[0089] In the formula, The control action at time t is the change in azimuth and tilt. express The risk of power curtailment at all times; This is the maximum charging power; This represents the current available capacity margin. To meet future load requirements; An angle adjustment penalty coefficient is used to prevent frequent rotation of the photovoltaic panel.

[0090] S303. Input the optimized attitude into the photovoltaic power generation prediction model to obtain the photovoltaic power generation prediction value.

[0091] This invention constructs a photovoltaic panel attitude optimization method aimed at minimizing the weighted sum of curtailment risk and attitude adjustment costs. Before obtaining the photovoltaic power generation forecast, the method proactively optimizes the future photovoltaic panel attitude and uses the optimized attitude as input to the prediction model, thereby obtaining a photovoltaic power generation forecast curve that better matches the actual power absorption capacity of the industrial park. On the one hand, it actively adjusts the attitude to suppress the predicted peak output, reducing curtailment risk at the source; on the other hand, it controls the attitude adjustment amplitude through weighted constraints to avoid excessively frequent rotations. The resulting photovoltaic power generation forecast is more closely aligned with subsequent load absorption benchmarks and energy storage charging capabilities, providing more accurate input for calculating the curtailment risk index and determining the dynamic operating range, thus improving the economy and reliability of the entire multi-source collaborative scheduling system.

[0092] In a specific embodiment of the present invention, the power curtailment risk indicator is:

[0093] In the formula, As an indicator of power curtailment risk; To optimize the load curve; For charging efficiency; To the theoretical maximum photovoltaic output; This represents the maximum charging power of the energy storage system.

[0094] It should be noted that: the dynamic operating range refers to the time-varying minimum state of charge (…). ) and maximum state of charge ( The interval with the boundary ) is used to replace the fixed minimum and maximum states of charge in the prior art.

[0095] The solution process for the dynamic running interval is based on Markov Decision Process (MDP). Specifically, the dynamic running interval is modeled as an MDP problem: Status: Includes the current energy storage state of the SoC t And the future T c Curtailment Risk Index (CRI) sequence at a given time point (e.g., 4 hours). And electricity price forecast series.

[0096] The action is: Select a reservation strategy. This corresponds to dynamically adjusting the upper and lower limits of the SoC to different levels: Strategy 0 (Normal Mode): =0.2, =0.9; Strategy 1 (Light Reservation): =0.15, =0.85; Strategy 2 (Moderate Reservation): =0.1, =0.8; Strategy 3 (Heavy Reservation): =0.05, =0.7; The higher the strategy number, the greater the capacity reserved for PV, but at the expense of other flexibility in energy storage.

[0097] Reward function:

[0098] in, Let t be the actual amount of electricity wasted. Encourage energy storage to remain near the median level to maintain regulation capabilities. It is a constraint violation indicator function (e.g., exceeding SoC limits). , , These are the weighting coefficients.

[0099] This MDP problem is trained offline and fine-tuned online using deep reinforcement learning algorithms (such as Soft Actor-Critic, SAC) to learn the optimal policy function and output the optimal reservation policy in real time based on the current state, i.e., the dynamic running interval.

[0100] In some embodiments of the present invention, the scheduling curve model is as follows:

[0101] In the formula, For charging efficiency; For discharge efficiency; Let be the predicted electricity price at time t; Let be the charging power at time t; Let be the discharge power at time t; Peak penalty weight; This is the power smoothing coefficient; Power purchased from the power grid The peak penalty item.

[0102] The constraints include energy storage dynamic constraints, capacity constraints, power constraints, time constraints, and economic constraints; The dynamic constraints for energy storage are:

[0103] The capacity constraint is:

[0104] The power constraint is:

[0105] The time period constraint is:

[0106]

[0107] The economic constraints are:

[0108] In the formula, The stored energy amount at time t+1; Let be the stored energy quantity at time t; For time intervals; Maximum energy storage capacity; Minimum energy storage capacity; This is the maximum charging power; The maximum discharge power; t For a specified time; The minimum discharge power at a specified time; The discharge power at a specified time; This is a set of charging periods.

[0109] The process of solving the charge and discharge scheduling curve is a mixed integer linear programming problem, which can be solved efficiently by commercial solvers such as Gurobi.

[0110] This invention constructs a scheduling curve model aimed at maximizing net revenue. By introducing a quadratic smoothing term for charging and discharging power and a peak power purchase penalty term from the grid, it effectively suppresses drastic fluctuations in energy storage power and reduces demand-based electricity cost risks. Furthermore, it sets dynamic constraints, capacity constraints, power constraints, time-period constraints, and economic constraints for energy storage. The time-period constraint ensures power output capacity during critical discharge moments, while the economic constraint guarantees that the electricity price during charging periods is strictly lower than the price during subsequent discharge periods, avoiding loss-making operations such as charging at high prices and discharging at low prices. The synergistic effect of these constraints allows the generated charging and discharging scheduling curve to automatically identify and utilize low-price periods for charging and high-price periods for discharging, while smoothing power commands to extend energy storage lifespan, ultimately maximizing peak-valley arbitrage revenue and minimizing grid interaction costs, all while meeting the actual electricity demand of the industrial park and the physical limitations of energy storage.

[0111] In actual operation, there will inevitably be a deviation between the predicted value of electricity price and photovoltaic power generation and the actual value. The traditional deterministic scheduling curve, namely the charging and discharging scheduling curve obtained in the above embodiment, is fixed and executed once it is generated. It cannot cope with sudden situations such as sudden changes in electricity price and weather, resulting in actual income deviating from expectations or even losses.

[0112] To solve this technical problem, in some embodiments of the present invention, after step S104, the following step is further included: According to the preset rolling optimization cycle, the scenario-based model predictive control (SPC) method is adopted. Based on the current actual energy storage capacity, electricity price probability prediction and photovoltaic probability prediction, the charging and discharging scheduling curve is corrected to obtain the optimized scheduling curve.

[0113] Specifically, assuming an error in electricity price forecasting Follows a mean of 0 and a variance of The normal distribution of the photovoltaic prediction error is assumed, and the photovoltaic prediction error is assumed to be... It follows another normal distribution. Based on this, a set of future scenarios with probability weights is generated through Monte Carlo simulation. Each scenario contains a possible trajectory for electricity price and photovoltaic output.

[0114] Rolling optimization: at each real-time decision moment Based on the latest system state (current energy storage capacity and the latest forecast information), a multi-scenario optimization problem is resolved. The objective is to maximize the expected net benefit.

[0115] In the formula, Let be the probability of scenario s; The cycle is optimized for rolling.

[0116] It needs to satisfy the following constraints: Robust constraints:

[0117] Opportunity Constraints:

[0118] In the formula, To determine the critical discharge moment under scenario s. The amount of electricity stored in the energy storage system at the previous moment; This is the minimum allowable amount of electricity for the energy storage system. For probability operators; The maximum probability threshold for allowing violations of economic conditions can be 0.05 or 0.1.

[0119] The chance constraints are transformed into a set of linear constraints using Sample Average Approximation (SAA), resulting in a large-scale mixed-integer linear programming problem. To accelerate the solution, the Benders decomposition algorithm is introduced: the original problem is decomposed into a main problem (determining the charging and discharging strategy for each time period) and multiple sub-problems (verifying the feasibility of each scenario and generating a feasible cut set), and the optimal scheduling curve can be obtained by iterative solution.

[0120] In market environments with extremely volatile electricity prices, energy storage charging and discharging strategies based on deterministic optimization or opportunity constraints in the above embodiments often cannot guarantee in real time that the electricity price during the charging period is strictly lower than the electricity price during the subsequent discharging period, which is an economic condition.

[0121] To solve this technical problem, in some embodiments of the present invention, after step S104, the following step is further included: The variance of the electricity price prediction error is calculated in real time. When the variance exceeds a preset threshold, a pre-trained deep Q-network is activated. The deep Q-network outputs discrete charging and discharging action commands and maps these commands to real-time charging and discharging power commands. The real-time charging and discharging power commands are then weighted and summed with the charging and discharging power commands of the optimized scheduling curve at the current moment to obtain the target charging and discharging power command. The weight of the real-time charging and discharging power command increases linearly with the increase of the variance of the electricity price prediction error. When the variance of the electricity price prediction error is less than or equal to the preset threshold, the charging and discharging power command of the optimized scheduling curve at the current moment is used as the target charging and discharging power command.

[0122] Specifically, deep Q-networks include: State space: includes the current electricity price, statistical characteristics (mean, variance) of the future K-step electricity price forecast, energy storage capacity, and time difference from the critical discharge moment, etc.

[0123] Action space: Discrete action set A = {Immediate charging (maximum power), partial charging (half power), no charging, discharging}.

[0124] Reward function: Carefully designed to ensure condition:

[0125] In the formula, A positive reward coefficient; It has a negative reward coefficient; This is the wear and tear penalty coefficient. When the energy storage system discharges at a certain moment and the electricity price in the previous period is significantly lower than the current period's price, a positive reward coefficient is given to encourage correct low-price charging and high-price discharging arbitrage behavior. Conversely, if it discharges but the electricity price in the previous period is not lower than the current period's price, a severe negative reward is applied to punish irrational discharging that may lead to losses. In other situations (such as charging or standby), a wear and tear penalty coefficient is introduced to suppress frequent or severe power fluctuations, thereby balancing the pursuit of arbitrage profits with the lifespan and economic efficiency of the energy storage equipment.

[0126] A well-trained DQN policy can provide suggested actions regarding whether to charge at the current moment during each decision step of MPC. The final decision of MPC will comprehensively consider the optimization result and the suggested actions.

[0127] This invention introduces a scenario-tree-based model predictive control (DQN) to continuously correct the scheduling curve. This allows for periodic updates of the scheduling strategy using the latest measured information, reducing the cumulative impact of prediction errors. Simultaneously, by monitoring the variance of electricity price prediction errors in real time, a deep Q-network is activated to generate real-time charging and discharging power commands when the error exceeds a threshold. These commands are then weighted and fused with the optimized scheduling curve, with higher weights for larger errors. This prioritizes the learned fast-response strategy in highly uncertain environments. When the error is small, the optimized scheduling curve is used to maintain long-term economic efficiency. This mechanism achieves an adaptive balance between planning and responsiveness, ensuring global optimality under normal operating conditions while enhancing robustness and response speed in extreme scenarios, effectively reducing economic losses caused by prediction deviations.

[0128] To verify the effectiveness of the multi-source coordinated charging and discharging strategy optimization method for energy storage systems proposed in this embodiment of the invention, practical verification was conducted, and the verification results are shown in Table 1: Table 1 Comparison of index values ​​before and after using the method of this embodiment of the invention

[0129] As shown in Table 1, the method proposed in this embodiment of the invention has superiority in various indicators, which verifies the effectiveness of this embodiment of the invention.

[0130] On the other hand, embodiments of the present invention also provide a multi-source coordinated charging and discharging strategy optimization system for energy storage systems, such as... Figure 4 As shown, the multi-source coordinated charging and discharging strategy optimization system 400 for energy storage systems includes: Electricity price prediction unit 401 is used to acquire multi-source historical data of the area where the energy storage system is located, and to predict the electricity price at multiple future times based on the multi-source historical data to obtain the electricity price prediction value. The load curve optimization unit 402 is used to identify high-electricity-price periods and low-electricity-price periods based on the electricity price forecast. It divides industrial loads into rigid loads and flexible loads according to their adjustability, and sets a time window for allowing power shifting for flexible loads. Within the time window, it transfers part of the flexible load power in the high-electricity-price period to the low-electricity-price period for execution. The goal is to minimize the weighted sum of the maximum demand power and electricity cost within the scheduling cycle, and outputs the optimized load curve. The dynamic operation range determination unit 403 is used to obtain the photovoltaic power generation prediction value and use the optimized load curve as the consumption benchmark. When the photovoltaic power generation prediction value exceeds the sum of the consumption benchmark and the current rechargeable capacity of the energy storage system, it is determined that there is a curtailment risk and the curtailment risk index is determined. Based on the curtailment risk index and the current state of charge of the energy storage system, the dynamic operation range is determined from multiple preset state of charge operation ranges. The charging and discharging scheduling curve determination unit 404 is used to construct a scheduling curve model with the goal of maximizing net income based on the electricity price forecast, construct the constraints of the scheduling curve model based on the optimized load curve and dynamic operating range, and solve the scheduling curve model based on the constraints to obtain the charging and discharging scheduling curve. Control unit 405 is used to control the operation of the energy storage system based on the charge and discharge scheduling curve.

[0131] The multi-source coordinated charging and discharging strategy optimization system 400 for energy storage systems provided in the above embodiments can realize the technical solutions described in the above embodiments of the multi-source coordinated charging and discharging strategy optimization method for energy storage systems. The specific implementation principles of each module or unit can be found in the corresponding content in the above embodiments of the multi-source coordinated charging and discharging strategy optimization method for energy storage systems, and will not be repeated here.

[0132] Those skilled in the art will understand that all or part of the processes of the methods described in the above embodiments can be implemented by a computer program instructing related hardware (such as a processor, controller, etc.), and the computer program can be stored in a computer-readable storage medium. The computer-readable storage medium may be a disk, optical disk, read-only memory, or random access memory, etc.

[0133] The above provides a detailed description of the multi-source coordinated charging and discharging strategy optimization method and system for energy storage systems provided by the present invention. Specific examples have been used to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only for the purpose of helping to understand the method and core ideas of the present invention. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of the present invention. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A method for optimizing multi-source coordinated charging and discharging strategies for energy storage systems, characterized in that, include: Acquire multi-source historical data of the area where the energy storage system is located, and predict the electricity price at multiple future times based on the multi-source historical data to obtain the predicted electricity price value; Based on the predicted electricity price, high-price and low-price periods are identified. Industrial loads are divided into rigid loads and flexible loads according to their adjustability. A time window for allowing power shifting is set for flexible loads. Within the time window, part of the flexible load power in the high-price period is transferred to the low-price period. The goal is to minimize the weighted sum of the maximum demand power and electricity cost within the scheduling cycle, and the optimized load curve is output. The photovoltaic power generation forecast value is obtained, and the optimized load curve is used as the absorption benchmark. When the photovoltaic power generation forecast value exceeds the sum of the absorption benchmark and the current rechargeable capacity of the energy storage system, it is determined that there is a curtailment risk, and a curtailment risk index is determined. Based on the curtailment risk index and the current state of charge of the energy storage system, a dynamic operating range is determined from multiple preset state of charge operating ranges. Based on the predicted electricity price, a scheduling curve model is constructed with the goal of maximizing net revenue. Based on the optimized load curve and the dynamic operating range, constraints are constructed for the scheduling curve model. The scheduling curve model is then solved based on the constraints to obtain the charging and discharging scheduling curve. The operation of the energy storage system is controlled based on the charge and discharge scheduling curve.

2. The multi-source coordinated charging and discharging strategy optimization method for energy storage systems according to claim 1, characterized in that, The process of predicting electricity prices at multiple future points based on the multi-source historical data to obtain predicted electricity prices includes: A multi-source feature vector is constructed based on the multi-source historical data. The multi-source feature vector includes electricity price time-series features, temperature features, and interaction features between date type and temperature. The multi-source feature vector is input into the electricity price prediction model, which includes a feature extraction module, a hybrid attention module, and a prediction module. The hybrid attention module includes a Flash Attention unit and a temporal convolutional attention unit. Based on the feature extraction module, nonlinear transformation and feature filtering are performed on the multi-source feature vectors to generate enhanced feature vectors; The global dependency features of the enhanced feature vector are extracted based on the Flash Attention unit, and the local temporal pattern features of the enhanced feature vector are extracted based on the temporal convolutional attention unit. The global dependency features and the local temporal pattern features are fused to obtain fused features; The fused features are input into the prediction module to obtain electricity price predictions for multiple future time points.

3. The multi-source coordinated charging and discharging strategy optimization method for energy storage systems according to claim 2, characterized in that, The loss function used during training of the electricity price prediction model is: In the formula, For multiple prediction times; Huber loss function; These are the actual electricity price and the predicted electricity price at the i-th prediction time, respectively; This represents the predicted electricity price at time i+1. This is the smoothing coefficient.

4. The multi-source coordinated charging and discharging strategy optimization method for energy storage systems according to claim 1, characterized in that, The optimized load curve, which aims to minimize the weighted sum of maximum power demand and electricity cost within a scheduling period, includes: The objective function is constructed by minimizing the weighted sum of the maximum power demand and electricity cost within the scheduling period. The objective function is as follows: ； ； Construct the objective constraints for the objective function. The objective constraints include power balance constraints, grid capacity constraints, and flexible load operation constraints. The flexible load operation constraints include total demand fulfillment constraints, shift time window constraints, and power regulation rate limit constraints. The power balance constraint is: ； The power grid capacity constraint is: ； The total demand fulfillment constraint is: ； The translation time window constraint is: ； The power regulation rate limit constraint is: ； The objective function is solved based on the objective constraints, and the optimized load curve is output. In the formula, Let be the total power consumption at time t; Power for rigid loads; The actual operating power of the k-th type of flexible load at time t; This represents the actual operating power of the k-th type of flexible load at time t-1. Based on the standard commissioning ratio; Reference power; This is a projected electricity price. , , These are the weighting coefficients; Power purchased from the power grid; The upper limit of the power purchase capacity of the power grid; The maximum power change rate of the k-th type of flexible load; , These are the start and end times of the time window, respectively. The maximum allowable translation duration; Photovoltaic power generation; This refers to the charging and discharging power of the energy storage system.

5. The multi-source coordinated charging and discharging strategy optimization method for energy storage systems according to claim 1, characterized in that, The acquisition of photovoltaic power generation prediction values ​​includes: A photovoltaic power generation prediction model is constructed. The input of the photovoltaic power generation prediction model includes the photovoltaic panel attitude, and the output is the photovoltaic power generation prediction value. With the goal of minimizing the weighted sum of the risk of curtailment and the cost of attitude adjustment, the attitude of the photovoltaic panels in the future is optimized to obtain the optimized attitude; The optimized posture is input into the photovoltaic power generation prediction model to obtain the photovoltaic power generation prediction value.

6. The multi-source coordinated charging and discharging strategy optimization method for energy storage systems according to claim 1, characterized in that, The aforementioned power curtailment risk indicator is: In the formula, As an indicator of power curtailment risk; To optimize the load curve; For charging efficiency; To the theoretical maximum photovoltaic output; This represents the maximum charging power of the energy storage system.

7. The multi-source coordinated charging and discharging strategy optimization method for energy storage systems according to claim 1, characterized in that, The scheduling curve model is as follows: In the formula, For charging efficiency; For discharge efficiency; Let be the predicted electricity price at time t; Let be the charging power at time t; Let be the discharge power at time t; Peak penalty weight; This is the power smoothing coefficient; Power purchased from the power grid The peak penalty item.

8. The multi-source coordinated charging and discharging strategy optimization method for energy storage systems according to claim 7, characterized in that, The constraints include energy storage dynamic constraints, capacity constraints, power constraints, time constraints, and economic constraints. The energy storage dynamic constraints are: The capacity constraint is: The power constraint is: The time period constraint is as follows: The economic constraints are as follows: In the formula, The stored energy amount at time t+1; Let be the stored energy quantity at time t; For time intervals; Maximum energy storage capacity; Minimum energy storage capacity; This is the maximum charging power; The maximum discharge power; t For a specified time; The minimum discharge power at a specified time; The discharge power at a specified time; This is a set of charging periods.

9. The multi-source coordinated charging and discharging strategy optimization method for energy storage systems according to claim 1, characterized in that, After obtaining the charge / discharge scheduling curve, the method further includes: According to the preset rolling optimization cycle, the model prediction control method based on the scenario tree is adopted to correct the charging and discharging scheduling curve based on the current actual energy storage capacity, electricity price probability prediction and photovoltaic probability prediction, so as to obtain the optimized scheduling curve. The variance of the electricity price prediction error is calculated in real time. When the variance exceeds a preset threshold, a pre-trained deep Q-network is activated. The deep Q-network outputs discrete charging and discharging action commands and maps these commands to real-time charging and discharging power commands. The real-time charging and discharging power commands are then weighted and summed with the charging and discharging power commands of the optimized scheduling curve at the current moment to obtain the target charging and discharging power command. The weight of the real-time charging and discharging power command increases linearly with the increase of the variance of the electricity price prediction error. When the variance of the electricity price prediction error is less than or equal to the preset threshold, the charging and discharging power command of the optimized scheduling curve at the current moment is used as the target charging and discharging power command.

10. A multi-source coordinated charging and discharging strategy optimization system for energy storage systems, characterized in that, include: The electricity price prediction unit is used to acquire multi-source historical data of the area where the energy storage system is located, and to predict the electricity price at multiple future times based on the multi-source historical data to obtain the predicted electricity price value. The load curve optimization unit is used to identify high-electricity-price periods and low-electricity-price periods based on the electricity price forecast, classify industrial loads into rigid loads and flexible loads according to their adjustability, and set a time window for allowing power shifting for flexible loads. Within the time window, a portion of the flexible load power in the high-electricity-price period is transferred to the low-electricity-price period for execution, with the goal of minimizing the weighted sum of the maximum demand power and electricity cost within the scheduling cycle, and outputting the optimized load curve. The dynamic operating range determination unit is used to obtain the photovoltaic power generation prediction value, and use the optimized load curve as the absorption benchmark. When the photovoltaic power generation prediction value exceeds the sum of the absorption benchmark and the current rechargeable capacity of the energy storage system, it is determined that there is a curtailment risk, and a curtailment risk index is determined. Based on the curtailment risk index and the current state of charge of the energy storage system, the dynamic operating range is determined from multiple preset state of charge operating ranges. The charging and discharging scheduling curve determination unit is used to construct a scheduling curve model with the goal of maximizing net revenue based on the electricity price forecast, construct the constraints of the scheduling curve model based on the optimized load curve and the dynamic operating range, and solve the scheduling curve model based on the constraints to obtain the charging and discharging scheduling curve. A control unit is used to control the operation of the energy storage system based on the charge-discharge scheduling curve.

Images