Method and device for determining electric energy charging and discharging strategy, energy storage scheduling system, storage medium and product

Abstract

The invention discloses an electric energy charging and discharging strategy determination method and device, an energy storage scheduling system, a storage medium and a product. The method comprises the steps of obtaining first information representing a current weather condition; determining a first power prediction value of the photovoltaic inverter based on the photovoltaic prediction model, and determining a first load prediction value of the energy storage scheduling system based on the load prediction model; determining a second power prediction value based on the first information, the current photovoltaic power value, the first power prediction value and a photovoltaic residual prediction model; determining a second load prediction value based on the first information, the current load value, the first load prediction value and a load residual prediction model; and based on the second power prediction value, the second load prediction value and a linear programming model, determining an electric energy charging and discharging strategy used for reducing the operation cost of the energy storage scheduling system. Therefore, the electric energy charging and discharging strategy with the optimal economical efficiency is determined based on the more accurate prediction result, and the problem that the charging and discharging strategy of the energy storage dispatching system is not economical enough is solved.

Description

Technical Field

[0001] This application relates to the field of photovoltaic technology, and in particular to a method, apparatus, energy storage scheduling system, storage medium and product for determining an energy charging and discharging strategy. Background Technology

[0002] In the photovoltaic-storage-charging scenario, the energy storage dispatch system is a key component of the new energy power system, mainly used to coordinate energy fluctuations between photovoltaic power generation, energy storage devices, and loads. Related technologies include rule-based control strategies, such as energy charging and discharging strategies adjusted according to electricity price time periods (e.g., peak shaving and valley filling plans), or adjustments based on SOC (State of Charge) ranges (e.g., charging below a set SOC threshold and discharging above a set SOC threshold). Energy storage dispatch schemes also include static optimization dispatch, which involves pre-planning day-ahead schedules based on historical photovoltaic power generation and load data (discharge plans pre-planned based on forecast data for the next day). However, these technologies struggle to address the intermittent fluctuations in photovoltaic power generation. These fluctuations lead to low energy utilization, making charging and discharging strategies uneconomical under peak-valley time-of-use pricing mechanisms. Summary of the Invention

[0003] In view of this, embodiments of this application provide a method, apparatus, energy storage scheduling system, storage medium, and product for determining an energy charging and discharging strategy, aiming to optimize the charging and discharging strategy of the energy storage scheduling system under the peak-valley time-of-use pricing mechanism.

[0004] The technical solution of this application embodiment is implemented as follows:

[0005] In a first aspect, embodiments of this application provide a method for determining an energy charging and discharging strategy, applied to an energy storage dispatching system. The energy storage dispatching system is connected to the power grid and includes photovoltaic panels, a photovoltaic inverter, an energy storage converter, and an energy storage battery. The method includes:

[0006] Obtain primary information representing the current weather conditions;

[0007] The first power prediction value of the photovoltaic inverter is determined based on the photovoltaic prediction model, and the first load prediction value of the energy storage dispatch system is determined based on the load prediction model.

[0008] Based on the first information, the current power value of the photovoltaic inverter, the first power prediction value and the photovoltaic residual prediction model, a second power prediction value is determined; and based on the first information, the current load value of the energy storage dispatch system, the first load prediction value and the load residual prediction model, a second load prediction value is determined.

[0009] Based on the second power forecast, the second load forecast, and the linear programming model, an energy charging and discharging strategy is determined; the energy charging and discharging strategy is used to reduce the operating cost of the energy storage dispatch system.

[0010] In some implementations, both the photovoltaic residual prediction model and the load residual prediction model include an XGBoost regression model.

[0011] In some implementations, determining the first power forecast value of the photovoltaic inverter based on a photovoltaic forecasting model and determining the first load forecast value of the energy storage dispatching system based on a load forecasting model includes:

[0012] Obtain secondary information that characterizes the weather conditions in the first time period;

[0013] Based on the second information and the photovoltaic prediction model, the first power prediction value of the photovoltaic inverter in the target time period is determined; and based on the second information and the load prediction model, the first load prediction value of the energy storage dispatch system in the target time period is determined.

[0014] The first time period and the target time period are updated based on the length of the target time period to obtain at least one first power prediction value and one first load prediction value;

[0015] The first time period is earlier than the target time period.

[0016] In some implementations, the photovoltaic prediction model includes a first linear layer, a Long Short-Time Memory (LSTM) network model, a second linear layer, and an activation function module. The step of determining the first power prediction value of the photovoltaic inverter within a target time period based on the second information and the photovoltaic prediction model includes:

[0017] Based on the first linear layer, the dimension of the second information is increased to a preset dimension;

[0018] The first prediction result is determined based on the second information of a preset dimension and the LSTM model.

[0019] The second linear layer reduces the dimension of the first prediction result to one dimension.

[0020] Based on the one-dimensional first prediction result and the activation function module, a first probability is determined;

[0021] The first power prediction value is determined based on the first probability and the rated power of the photovoltaic inverter.

[0022] In some implementations, determining the second power prediction value based on the first information, the current power value of the photovoltaic inverter, the first power prediction value, and the photovoltaic residual prediction model includes:

[0023] The first information, the current power value of the photovoltaic inverter, the first power prediction value, and the current time information are input into the photovoltaic residual prediction model to obtain the first difference;

[0024] The second power prediction value is determined based on the sum of the first difference and the first power prediction value.

[0025] In some implementations, the load forecasting model includes an XGBoost regression model.

[0026] In some implementations, determining the second load forecast value based on the first information, the current load value of the energy storage scheduling system, the first load forecast value, and the load residual forecast model includes:

[0027] The first information, the current load value of the energy storage scheduling system, the first load prediction value, and the current time information are input into the load residual prediction model to obtain the second difference;

[0028] The second load forecast value is determined based on the sum of the second difference and the first load forecast value.

[0029] In some implementations, the linear programming model includes an objective function and constraints;

[0030] The objective function is used to minimize the difference between the first amount and the second amount within a first preset time period;

[0031] The first amount is determined by the power and electricity price when the energy storage dispatch system obtains electrical energy from the grid, and the second amount is determined by the power and electricity price when the energy storage dispatch system outputs electrical energy to the grid;

[0032] The constraint condition is used to ensure that the energy consumption and energy acquisition of the energy storage scheduling system are the same during the first preset time period.

[0033] Secondly, embodiments of this application provide a device for determining an energy charging and discharging strategy, applied to an energy storage dispatching system. The energy storage dispatching system is connected to the power grid and includes photovoltaic panels, a photovoltaic inverter, an energy storage converter, and an energy storage battery. The device includes:

[0034] The acquisition module is used to acquire the first information representing the current weather conditions;

[0035] The first determining module is used to determine the first power prediction value of the photovoltaic inverter based on the photovoltaic prediction model, and to determine the first load prediction value of the energy storage dispatch system based on the load prediction model.

[0036] The second determining module is used to determine a second power prediction value based on the first information, the current power value of the photovoltaic inverter, the first power prediction value and the photovoltaic residual prediction model; and to determine a second load prediction value based on the first information, the current load value of the energy storage scheduling system, the first load prediction value and the load residual prediction model.

[0037] The third determining module is used to determine the energy charging and discharging strategy based on the second power prediction value, the second load prediction value and the linear programming model; the energy charging and discharging strategy is used to reduce the operating cost of the energy storage scheduling system.

[0038] Thirdly, embodiments of this application provide an energy storage dispatching system, which is connected to the power grid. The energy storage dispatching system includes photovoltaic panels, photovoltaic inverters, energy storage converters, and energy storage batteries. The energy storage dispatching system also includes a processor and a memory for storing a computer program that can run on the processor. When the processor runs the computer program, it performs the steps of the method described in the first aspect.

[0039] In some implementations, the energy storage dispatch system further includes a server on which photovoltaic forecasting models and load forecasting models are deployed;

[0040] The processor deploys a photovoltaic residual prediction model and a load residual prediction model.

[0041] Fourthly, embodiments of this application provide a computer storage medium storing a computer program, which, when executed by a processor, implements the steps of the method described in the first aspect.

[0042] Fifthly, embodiments of this application provide a computer program product, including a computer program that, when executed by a processor, implements the steps of the method described in the first aspect.

[0043] This application provides a method for determining an energy charging and discharging strategy, applied to an energy storage dispatching system. The energy storage dispatching system is connected to the power grid and includes photovoltaic panels, photovoltaic inverters, energy storage converters, and energy storage batteries. The method includes: acquiring first information representing current weather conditions; determining a first power prediction value for the photovoltaic inverter based on a photovoltaic prediction model, and determining a first load prediction value for the energy storage dispatching system based on a load prediction model; determining a second power prediction value based on the first information, the current power value of the photovoltaic inverter, the first power prediction value, and a photovoltaic residual prediction model; determining a second load prediction value based on the first information, the current load value of the energy storage dispatching system, the first load prediction value, and a load residual prediction model; and determining an energy charging and discharging strategy based on the second power prediction value, the second load prediction value, and a linear programming model. The energy charging and discharging strategy is used to reduce the operating cost of the energy storage dispatching system. Therefore, the solution in this application introduces the rolling optimization approach of MPC (Model Predictive Control). First, it predicts the power of the photovoltaic inverter and the load of the energy storage dispatch system in the future. Then, it corrects the predicted photovoltaic inverter power and the predicted energy storage dispatch system load based on the current weather conditions and the current photovoltaic inverter power. This improves the accuracy of the prediction of photovoltaic inverter power and energy storage dispatch system load. Furthermore, it uses a linear programming model to transform the prediction results into an economically optimal dispatch strategy, solving the problem that inaccurate predictions caused by abnormal weather in the energy storage dispatch system lead to uneconomical charging and discharging strategies and low energy utilization. Attached Figure Description

[0044] Figure 1 This is a schematic diagram of the structure of a static optimization scheduling system in related technologies;

[0045] Figure 2 This is a flowchart illustrating the method for determining the energy charging and discharging strategy according to an embodiment of this application.

[0046] Figure 3 This is a schematic diagram of the structure of the photovoltaic prediction model used in this application.

[0047] Figure 4 A flowchart illustrating the process of determining the energy charging and discharging strategy for an application example of this application;

[0048] Figure 5 This is a schematic diagram of the structure of an energy storage dispatch system, which is an application example of this application.

[0049] Figure 6 This is a schematic diagram of the structure of the device for determining the energy charging and discharging strategy according to an embodiment of this application;

[0050] Figure 7 This is a schematic diagram of the circuit structure of the energy storage scheduling system according to an embodiment of this application. Detailed Implementation

[0051] The present application will now be described in further detail with reference to the accompanying drawings and embodiments.

[0052] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein in the specification of this application is for the purpose of describing particular embodiments only and is not intended to be limiting of this application.

[0053] Before describing the embodiments of this application, the relevant terms will be explained.

[0054] An energy storage dispatch system is a comprehensive energy management system that integrates photovoltaic power generation, energy storage, and grid interaction. It achieves intelligent control over the entire process of energy production, storage, and consumption by coordinating the operating status of core equipment such as photovoltaic inverters (PVIs) and power conversion systems (PCSs). The main function of a photovoltaic inverter is to convert the direct current (DC) generated by photovoltaic panels (solar panels) into alternating current (AC) that can be used by loads or fed into the grid. The main function of a power conversion system is to realize the conversion of electrical energy between energy storage batteries and the grid or loads. For example, it can convert AC power from the grid into DC power for storage in the battery, or convert DC power from the battery into AC power to feed into the grid or supply the load.

[0055] In photovoltaic-storage-charging (PV-SGC) scenarios, the energy storage dispatch system is a key component of the new energy power system, primarily used to coordinate energy fluctuations between photovoltaic power generation, energy storage devices, and loads. Related technologies include rule-based control strategies, such as energy charging and discharging strategies adjusted according to electricity price time periods (e.g., peak shaving and valley filling plans), or adjustments based on SOC (State of Charge) ranges (e.g., charging below a set SOC threshold and discharging above a set SOC threshold). Energy storage dispatch schemes also include static optimization dispatch, which involves pre-planning day-ahead schedules based on historical PV power generation and load data (dispatch plans pre-planned based on forecast data for the next day). Figure 1As shown, the system equipment in this scenario mainly includes a cloud server 101 of the Energy Management System (EMS), an embedded controller 102, a PCS 103, a BMS (Battery Management System) 104, a PVI 105, a load 106, a photovoltaic panel (not shown in the figure), and an electricity meter (not shown in the figure). The system contains two information links: status uplink and control downlink. In the uplink, the embedded controller collects information from the underlying devices (such as the PCS and PVI) and then transmits it to the cloud server 101. In the downlink control, the cloud server 101 calculates the energy charging and discharging strategy or a static strategy (i.e., a pre-defined strategy that will not change regardless of changes in the external environment) and sends it to the embedded controller 102 for execution. The embedded controller 102 controls the various underlying devices according to the strategy.

[0056] The aforementioned technologies are all inadequate to address the intermittent fluctuations in photovoltaic power generation. These intermittent fluctuations lead to low energy utilization rates, and the charging and discharging strategies are not economical under the peak-valley time-of-use pricing mechanism.

[0057] In various embodiments of this application, the rolling optimization approach of MPC is introduced. First, the power of the photovoltaic inverter and the load of the energy storage dispatch system are predicted for a period of time in the future. Then, the predicted power of the photovoltaic inverter is corrected according to the current weather conditions and the current power of the photovoltaic inverter, and the predicted load of the energy storage dispatch system is corrected according to the current weather conditions and the current load. This improves the accuracy of the prediction of the power of the photovoltaic inverter and the load of the energy storage dispatch system. Furthermore, the prediction results are transformed into the most economically optimal dispatch strategy using a linear programming model. This solves the problem that the prediction values ​​of the energy storage dispatch system are inaccurate due to abnormal weather, which leads to uneconomical charging and discharging strategies and low energy utilization.

[0058] This application provides a method for determining an energy charging and discharging strategy, applied to an energy storage dispatching system. The energy storage dispatching system is connected to the power grid and includes photovoltaic panels, photovoltaic inverters, energy storage converters, and energy storage batteries. Figure 2 As shown, the method includes:

[0059] Step 201: Obtain the first information representing the current weather conditions;

[0060] In practical applications, the energy storage dispatch system can be equipped with a temperature, humidity and light sensor whose probe direction is the same as that of the photovoltaic panel's light-receiving surface. The temperature, humidity and light sensor can be used to obtain the first information representing the current weather conditions. The first information can be parameters such as temperature, humidity and light intensity at the current moment.

[0061] Step 202: Determine the first power prediction value of the photovoltaic inverter based on the photovoltaic prediction model, and determine the first load prediction value of the energy storage dispatch system based on the load prediction model;

[0062] Here, a photovoltaic (PV) prediction model is used to predict the average power of the PV inverters and the average load of the energy storage dispatch system (i.e., the average power consumed by the load) within a future time period (i.e., the target time period). The predicted average power of the PV inverters and the average load of the energy storage dispatch system for the target time period are used as the first power prediction value and the first load prediction value, respectively. It should be noted that the first power prediction value and the first load prediction value can be the average value within the target time period, or they can be multiple values ​​corresponding to multiple time points within the target time period. This embodiment of the application does not limit this.

[0063] Step 203: Based on the first information, the current power value of the photovoltaic inverter, the first power prediction value, and the photovoltaic residual prediction model, determine the second power prediction value; and based on the first information, the current load value of the energy storage dispatch system, the first load prediction value, and the load residual prediction model, determine the second load prediction value.

[0064] Here, the current power value of the photovoltaic inverter can be simply referred to as the current photovoltaic power value. The power value of the photovoltaic inverter can also be called the photovoltaic power value or the real-time output power value of the photovoltaic system.

[0065] In practical applications, when training the photovoltaic residual prediction model, a training set is constructed using historical temperature, humidity, and light sensor data, historical photovoltaic power values, historical first power prediction values, and corresponding actual photovoltaic power values. This training set is used to train the photovoltaic residual prediction model so that the model can fit the residuals based on the first information, the current photovoltaic power value, and the first power prediction value, thereby obtaining a corrected second power prediction value that is closer to the actual photovoltaic power value.

[0066] Similarly, when training the load residual prediction model, a training set is constructed using historical temperature, humidity, and light sensor data, historical load data, historical first load prediction values, and corresponding actual load values. This set is used to train the load residual prediction model so that the load residual prediction model can fit the residuals based on the first information, the current load value, and the first load prediction value, making the obtained second load prediction value closer to the actual load value.

[0067] Step 204: Based on the second power forecast value, the second load forecast value and the linear programming model, determine the energy charging and discharging strategy; the energy charging and discharging strategy is used to reduce the operating cost of the energy storage dispatch system.

[0068] Here, the second power prediction value and the second load prediction value, which are closer to the actual situation, are input into the linear programming model. The linear programming model is solved to obtain the energy charging and discharging strategy, which is used to reduce the operating cost of the energy storage dispatch system.

[0069] Understandably, the introduction of the MPC rolling optimization approach first predicts the photovoltaic inverter power and energy storage dispatch system load over a future period. Then, it corrects the predicted photovoltaic inverter power and the predicted energy storage dispatch system load based on the current weather conditions and current photovoltaic inverter power. This improves the accuracy of the photovoltaic inverter power and energy storage dispatch system load predictions. Furthermore, it uses a linear programming model to transform the prediction results into an economically optimal dispatch strategy, solving the problem that inaccurate predictions caused by abnormal weather in the energy storage dispatch system lead to uneconomical charging and discharging strategies and low energy utilization.

[0070] In some embodiments, both the photovoltaic residual prediction model and the load residual prediction model include an XGBoost regression model.

[0071] The residual between photovoltaic power and load (the difference between the actual value and the initial predicted value) often exhibits complex characteristics such as nonlinearity, strong noise, and multi-feature correlation. For example, the residual is affected by factors such as sudden weather changes (e.g., cloud cover) and random load fluctuations (e.g., sudden start-up of electrical equipment), and its variation is not a simple linear relationship; the real-time monitored data may contain noise such as sensor errors and communication interference, resulting in large fluctuations in the residual data; the residual is correlated with multiple dimensions of characteristics such as current weather (solar intensity, temperature), time (time period characteristics), and historical power / load values.

[0072] XGBoost (Extreme Gradient Boosting) is an ensemble learning model based on Gradient Boosting Trees (GBDT), which has the following beneficial effects:

[0073] 1. Strong nonlinear fitting ability, using an additive model of multiple decision trees to progressively optimize prediction results;

[0074] 2. It has a regularization mechanism that can suppress noise interference and prevent the model from overfitting noisy data;

[0075] 3. It supports automatic handling of missing values ​​and categorical feature encoding, and is insensitive to feature scale. It is suitable for fusing multiple types of inputs such as weather, time, and historical power / load data, which can improve the robustness of residual prediction.

[0076] 4. High training efficiency: It adopts parallel computing optimization and approximation algorithms to shorten the training time while ensuring accuracy, making it suitable for energy storage scheduling scenarios with high real-time requirements.

[0077] Therefore, the use of the XGBoost model can adapt to the residual prediction requirements of photovoltaic power and load, and can obtain more accurate prediction values ​​of photovoltaic power and load, thus ensuring the economy and stability of the energy storage dispatch system.

[0078] In related technologies, the photovoltaic power or load situation for the next day or several days is usually predicted directly through weather forecasts. However, this does not take into account the characteristics of weather forecast updates and the fact that photovoltaic power is easily affected by short-term weather conditions. This leads to the accumulation of errors in the prediction results, which affects the accuracy of power charging and discharging strategies.

[0079] Based on this, in some embodiments, determining the first power prediction value of the photovoltaic inverter based on the photovoltaic prediction model and determining the first load prediction value of the energy storage dispatch system based on the load prediction model includes:

[0080] Obtain second information representing the weather conditions in the first time period of the forecast;

[0081] Based on the second information and the photovoltaic prediction model, the first power prediction value of the photovoltaic inverter in the target time period is determined; and based on the second information and the load prediction model, the first load prediction value of the energy storage dispatch system in the target time period is determined.

[0082] The first time period and the target time period are updated based on the length of the target time period to obtain at least one first power prediction value and one first load prediction value;

[0083] The first time period is earlier than the target time period.

[0084] To predict the photovoltaic power and load values ​​for a target period, it is necessary to input weather forecast information (secondary information) prior to the target period into the photovoltaic prediction model. This is because photovoltaic power is easily affected by weather changes over a past period. For example, when clouds block the sun, photovoltaic power will drop rapidly, and the movement trajectory and speed of clouds need to be obtained using weather forecast information from an earlier period. Similarly, load is easily affected by weather conditions from an earlier period. For example, the weather may begin to change several hours before high temperatures or rainfall arrive, and when high temperatures or rainfall occur, the electricity consumption of equipment such as air conditioners and lights will surge.

[0085] For example, to predict the photovoltaic power and load values ​​for a target period from 8:00 to 8:15 tomorrow, the weather forecast information for the first period (6:00 to 8:00 tomorrow) two hours prior to the target period needs to be input into the photovoltaic prediction model and the load prediction model. After obtaining the photovoltaic power and load values ​​for 8:00 to 8:15 tomorrow, the first period and the target period are updated by the length of the target period (15 minutes), resulting in a new first period of 6:15 to 8:15 tomorrow and a target period of 8:15 to 8:30 tomorrow. This process is repeated, arranging the prediction results for multiple target periods in chronological order to obtain the predicted values ​​for photovoltaic power and load over a longer period. Because this rolling prediction only inputs short-term data prior to each target period, it can utilize the latest weather forecasts in a timely manner, reducing accumulated errors and improving prediction accuracy. Here, the duration of the target period can be a time frame (a time frame can be 5 minutes or 15 minutes), and the specific duration of the target period and the first period can be determined according to actual conditions; this embodiment does not limit this.

[0086] In practical applications, the training set can be preprocessed before training the photovoltaic prediction model. This involves aligning historical weather forecast information and photovoltaic power values ​​for the time period to be predicted according to the forecast, dividing the data into time frames, calculating the mean for each frame, and performing linear interpolation to fill in missing data in a time frame if any is missing. Finally, the means are normalized to unify the dimensions of each feature and balance the feature weights. A training set is then constructed based on the processed weather forecast information and the corresponding photovoltaic power values ​​to train the photovoltaic prediction model.

[0087] Before training the load prediction model, historical weather forecast information and the load values ​​for the time period to be predicted can be categorized by holidays and weekdays. The categorized data should be aligned chronologically, divided into time frames, and the mean value within each frame should be calculated. Linear interpolation should then be performed, and finally, the means should be normalized. A training set is constructed based on the processed weather forecast information and corresponding load values ​​to train the load prediction model. Stochastic gradient descent can be used to train both the photovoltaic prediction model and the load prediction model; the stochastic gradient descent method can be understood by referring to relevant techniques. The specific methods for linear interpolation and normalization can be found in relevant techniques; therefore, the specific methods for linear interpolation and normalization will not be elaborated here.

[0088] In related technologies, neural network models have difficulty accurately capturing the dynamic changes in photovoltaic power, resulting in inaccurate first power prediction values.

[0089] Based on this, in some embodiments, the photovoltaic prediction model includes a first linear layer, an LSTM model, a second linear layer, and an activation function module. The step of determining the first power prediction value of the photovoltaic inverter within a target time period based on the second information and the photovoltaic prediction model includes:

[0090] Based on the first linear layer, the dimension of the second information is increased to a preset dimension;

[0091] The first prediction result is determined based on the second information of a preset dimension and the LSTM model.

[0092] The second linear layer reduces the dimension of the first prediction result to one dimension.

[0093] Based on the one-dimensional first prediction result and the activation function module, a first probability is determined;

[0094] The first power prediction value is determined based on the first probability and the rated power of the photovoltaic inverter.

[0095] For example, the second information may include 22 weather forecast elements, such as temperature, air pressure, humidity, wind speed, wind direction, cloud cover, precipitation, visibility, fog, haze, sandstorm, short-term heavy precipitation, thunderstorm, hail, strong wind, cold wave, etc. This application embodiment does not limit the specific number of weather forecast elements included in the second information.

[0096] Based on the above example, to predict the average photovoltaic power and average load for the target period from 8:00 to 8:15 tomorrow, we need to input the weather forecast information (second information) for the first period (6:00 to 8:00) two hours before the target period into the photovoltaic prediction model and the load prediction model. The weather forecast information for the first period can be divided into 8 parts according to time frames (15 minutes), forming an 8×22 shape of second information. This 1×22 shape of second information is then used as an input vector and sequentially input into the photovoltaic prediction model. The second information in a 1×22 shape first enters the first linear layer, which contains three sub-linear layers with dimensional changes of [22, 64], [64, 128], and [128, 64], respectively. That is, the first linear layer can increase the dimensionality of the 1×22 shape second information to 64 dimensions, then increase it from 64 dimensions to 128 dimensions, and finally reduce it from 128 dimensions to 64 dimensions. Subsequently, the 64-dimensional second information is input into a bidirectional LSTM model with an intermediate variable dimension of 64 dimensions. Since the output dimension of the bidirectional LSTM model is twice the dimension of the intermediate variable, the bidirectional LSTM model outputs a 128-dimensional first prediction result. The second linear layer reduces the 128-dimensional first prediction result to 1 dimension and inputs the 1-dimensional first prediction result into the sigmoid activation function (activation function module). The sigmoid activation function can map the 1-dimensional first prediction result to the 0-1 space to obtain a number between 0 and 1, i.e., the first probability. Multiplying the first probability by the rated power of the photovoltaic inverter yields the first power prediction value.

[0097] Here, transforming the second information from low-dimensional to high-dimensional allows for the representation of more complex feature combinations, helping the neural network model learn nonlinear relationships in the input vector. The high-dimensional intermediate layer increases the number of parameters, improving the model's fitting ability. Conversely, reducing the dimensionality of the high-dimensional second information removes redundant information, retains key features, and prevents overfitting. Therefore, the first linear layer is used to first increase the dimensionality of the input vector and then reduce it. Photovoltaic power is easily affected by weather conditions (such as irradiance and cloud cover), and weather often exhibits correlation and dynamic changes over a long period. The gating mechanism of the LSTM model (input gate, forget gate, output gate) can dynamically and selectively retain or forget historical information, thereby preserving long-term trends (such as seasonal irradiance changes), filtering short-term noise (such as sensor errors or sudden weather changes), and adapting to non-stationary data (such as sudden changes in power generation caused by day-night cycles). Furthermore, the second set of information includes 22 weather forecast elements. The memory cells of the LSTM model can help establish the correlations between features, such as the negative correlation between irradiance and cloud cover, and the positive correlation between wind speed and module temperature. The LSTM model can also suppress irrelevant features, such as the impact of wind direction data during non-critical periods, on the prediction through forget gates. Compared to the LSTM model, the bidirectional LSTM model can capture the impact of future trends on the present through the inverse LSTM model, further improving the prediction accuracy based on contextual information. Therefore, the bidirectional LSTM model is used to obtain the predicted value of photovoltaic power.

[0098] In some embodiments, determining the second power prediction value based on the first information, the current power value of the photovoltaic inverter, the first power prediction value, and the photovoltaic residual prediction model includes:

[0099] The first information, the current power value of the photovoltaic inverter, the first power prediction value, and the current time information are input into the photovoltaic residual prediction model to obtain the first difference;

[0100] The second power prediction value is determined based on the sum of the first difference and the first power prediction value.

[0101] In practical applications, the photovoltaic residual prediction model can be trained to fit the difference between the first power prediction value and the actual photovoltaic power value for the next time frame (e.g., 15 minutes after the current time). Based on the first information (including current temperature, humidity, and irradiance parameters), the current photovoltaic power value, and the first power prediction value, a 6-dimensional input vector is constructed from the current time information. This input vector is fed into the photovoltaic residual prediction model, which fits the difference between the first power prediction value and the actual photovoltaic power value for the next time frame. The first power prediction value is then added to this difference to obtain the second power prediction value. Because the residual is compensated for, the second power prediction value for the next time frame more closely approximates the actual load value. Here, the current time information can be an encoding corresponding to the current time, as long as the encoding can represent the current time. This application embodiment does not limit the specific form of the current time information. Furthermore, the photovoltaic residual prediction model can be used to obtain the second power prediction value for the next time frame, or it can be used to obtain the second power prediction value after a relatively long period of time from the current time. This application embodiment does not limit this either.

[0102] In some embodiments, the load prediction model includes an XGBoost regression model.

[0103] Here, because load is affected by multiple factors such as weather, user behavior, and time, the load forecasting models in related technologies struggle to handle the complex interactions between multiple features, resulting in low load forecasting accuracy and affecting the rationality of energy storage scheduling. Therefore, the XGBoost regression model is adopted as the load forecasting model, which can efficiently handle multi-dimensional features, learn nonlinear relationships, and resist outlier interference. It can significantly improve the accuracy of the first load forecast value, providing a high-quality initial load forecast basis for subsequent residual correction and charging / discharging strategies.

[0104] Based on the above example, the weather forecast information (second information, including 22 weather forecast elements) for the two hours before the target period (6:00 to 8:00) and the codes corresponding to each time frame in the second information (such as the eight 15-minute intervals between 6:00 and 8:00) are combined to form an 8×23 input vector, which is then input into the XGBoost regression model. The XGBoost regression model can determine whether the target period is a holiday or a weekday based on the codes corresponding to each time frame in the second information, thereby fitting a more accurate load (electrical load power) within the target period according to the different characteristics of holidays and weekdays, thus improving the accuracy of the first load prediction value.

[0105] In some embodiments, determining the second load forecast value based on the first information, the current load value of the energy storage scheduling system, the first load forecast value, and the load residual forecast model includes:

[0106] The first information, the current load value of the energy storage scheduling system, the first load prediction value, and the current time information are input into the load residual prediction model to obtain the second difference;

[0107] The second load forecast value is determined based on the sum of the second difference and the first load forecast value.

[0108] Here, the load residual prediction model is trained to fit the difference between the first load prediction value and the actual load value in the next time frame (e.g., 15 minutes after the current time). Based on the first information (including the current temperature, humidity, and light parameters), the current load value, the first load prediction value, and the current time information, a 6-dimensional input vector is constructed and input into the load residual prediction model. The load residual prediction model can fit the difference between the first load prediction value and the actual load value in the next time frame. The first load prediction value and this difference are summed to obtain the second load prediction value, which can more closely approximate the actual load value.

[0109] In related technologies, the formulation of power charging and discharging strategies often neglects the economic balance between purchasing and selling electricity, resulting in excessively high operating costs and affecting the economic efficiency of the system.

[0110] Accordingly, in some embodiments, the linear programming model includes an objective function and constraints;

[0111] The objective function is used to minimize the difference between the first amount and the second amount within a first preset time period;

[0112] The first amount is determined by the power and electricity price when the energy storage dispatch system obtains electrical energy from the grid, and the second amount is determined by the power and electricity price when the energy storage dispatch system outputs electrical energy to the grid;

[0113] The constraint condition is used to ensure that the energy consumption and energy acquisition of the energy storage scheduling system are the same during the first preset time period.

[0114] In practical applications, the objective function of a linear programming model can be expressed as shown in Equation 1:

[0115]

[0116] in, This represents the average power input from the power grid to the energy storage dispatch system during time period t. This represents the average power output of the energy storage dispatch system to the grid during time period t. Let represent the price at which electricity is purchased from the grid during time period t, and Let represent the price at which electricity is sold to the grid during time period t. It is evident that the objective of solving the linear programming model is to minimize the difference between the funds spent by the energy storage dispatch system on buying electricity and the funds earned from selling electricity, thus minimizing the operating cost of the energy storage dispatch system.

[0117] Constraints may include:

[0118]

[0119]

[0120]

[0121]

[0122]

[0123]

[0124]

[0125] Among them, load t pvi represents the average load during the t-th time period. t Let represent the average power of the photovoltaic inverter during time period t, and let initCap, maxCap, and minCap represent the initial, maximum, and minimum capacities of the energy storage battery, respectively. This represents the average charging power of the energy storage converter (PCS) during time period t. This represents the average discharge power of the energy storage converter in time period t. This indicates the maximum charging power of the PCS. This indicates the maximum discharge power of the PCS. This represents the maximum reverse power, that is, the maximum power value of electrical energy transmitted to the grid. This represents the demand limit, i.e., the maximum amount of electrical energy that can be obtained from the grid. pcs b is a binary variable, equal to 1 when the energy storage converter controls the energy storage battery to charge, and equal to 0 when discharging; grid This is a binary variable, with a value of 1 when outputting electrical energy to the grid and a value of 0 when receiving electrical energy from the grid. In formulas 3-4, the upper limit of the number of time periods t is m, where m is a value set by the user, i.e., the user's desired time constraint range. It should be noted that t here can represent a time frame, an hour, or a multiple of the length of a time frame; this application does not limit this.

[0126] The constraint in Formula 2 is a power balance constraint, which means that within a total of 24 t time periods, the output power and input power need to be balanced, that is, the power consumed by the load + the power charged to the energy storage battery + the power output to the grid = the power input from the grid + the power discharged by the energy storage battery + the power provided by the photovoltaic.

[0127] The constraints in Formulas 3 and 4 are the maximum / minimum battery capacity constraints, meaning that the battery capacity cannot exceed the maximum capacity or be less than the minimum capacity within m time periods.

[0128] The constraints in Formulas 5 and 6 are power constraints for the energy storage converter, namely, the power of the energy storage converter when controlling the energy storage battery to charge cannot exceed the maximum charging power; the power of the energy storage converter when controlling the energy storage battery to discharge cannot exceed the maximum discharging power.

[0129] The constraints in Formulas 7 and 8 are anti-reverse flow and demand control constraints, that is, the power output to the grid cannot exceed the maximum reverse flow power, and the power obtained from the grid cannot exceed the demand limit value.

[0130] For example, the second power prediction value is pvi. t The second load forecast value is the load. t Based on these two known values, the linear programming model is solved. The specific method for solving the linear programming model can refer to relevant technologies, such as graphical method, simplex method, or using existing computer software (such as MATLAB). This application does not limit the specific methods used.

[0131] Based on the above example, the target time period can be the next time frame of the current moment. After the second power prediction value and the second load prediction value of the target time period are solved, they are input into the linear programming model to solve the linear programming model, thereby obtaining the power charging and discharging strategy for the next time frame. After the power charging and discharging strategy is implemented, the target time period is updated to the next time frame of the current moment, and the second power prediction value and the second load prediction value of the new target time period are solved. Therefore, the linear programming model can generate power charging and discharging strategies for multiple time frames in a rolling manner, which can solve the problems of insufficient economic optimization and low energy utilization rate of energy storage scheduling system.

[0132] The method for determining an energy charging and discharging strategy provided in this application embodiment is applied to an energy storage dispatching system. The energy storage dispatching system is connected to the power grid and includes photovoltaic panels, photovoltaic inverters, energy storage converters, and energy storage batteries. The method includes: acquiring first information representing the current weather conditions; determining a first power prediction value for the photovoltaic inverter based on a photovoltaic prediction model, and determining a first load prediction value for the energy storage dispatching system based on a load prediction model; determining a second power prediction value based on the first information, the current power value of the photovoltaic inverter, the first power prediction value, and a photovoltaic residual prediction model; determining a second load prediction value based on the first information, the current load value of the energy storage dispatching system, the first load prediction value, and a load residual prediction model; and determining an energy charging and discharging strategy based on the second power prediction value, the second load prediction value, and a linear programming model. The energy charging and discharging strategy is used to reduce the operating cost of the energy storage dispatching system. Thus, the proposed solution introduces the rolling optimization approach of MPC. First, it predicts the photovoltaic inverter power and energy storage dispatch system load over a future period. Then, it corrects the predicted photovoltaic inverter power and the predicted energy storage dispatch system load based on the current weather conditions and the current photovoltaic inverter power. This improves the accuracy of the photovoltaic inverter power and energy storage dispatch system load predictions. Furthermore, it uses a linear programming model to transform the prediction results into an economically optimal dispatch strategy, solving the problem that inaccurate predictions caused by abnormal weather in the energy storage dispatch system lead to uneconomical charging and discharging strategies and low energy utilization.

[0133] The following section provides a more detailed description of this application with reference to application examples.

[0134] The photovoltaic prediction model provided in this application example is as follows: Figure 3 As shown, the model includes a first linear layer 301, a long short-term memory network 302 (i.e., the LSTM model mentioned above), and a second linear layer 303. Weather forecast data (i.e., the second information mentioned above) is input into the photovoltaic prediction model, and the photovoltaic predicted power (i.e., the first power prediction value mentioned above) is output.

[0135] The flowchart illustrating the determination of the energy charging and discharging strategy provided in this application example is as follows: Figure 4 As shown, it includes:

[0136] Step 401: The server obtains weather forecast data (i.e., the second information mentioned above), and then executes step 402;

[0137] Step 402: Predict the photovoltaic / load power generation for the next day or more and send it to the embedded controller, then execute step 403;

[0138] Here, the second information is input into the photovoltaic prediction model and the load prediction model to obtain the photovoltaic / load power generation for one or more days in the future, and the predicted photovoltaic / load power generation is sent to the processor of the energy storage scheduling system.

[0139] Because weather forecasts are updated relatively slowly, and the prediction models require significant computing resources, the photovoltaic (PV) forecasting model and the load forecasting model need to be deployed on a cloud server. Therefore, steps 401 and 402 occur on the cloud server, meaning the PV forecasting model and the load forecasting model are deployed on the energy storage dispatching system's server. Since weather forecasts are typically updated every 8 hours, steps 401 and 402 are executed every 8 hours.

[0140] Step 403: The controller collects data from the implementation device, performs residual correction, and then executes step 404;

[0141] Here, the first power prediction value of the next time frame at the current moment is input into the photovoltaic residual prediction model to obtain the second power prediction value, and the first load prediction value of the next time frame at the current moment is input into the load residual prediction model to obtain the second load prediction value.

[0142] Step 404: Using the corrected photovoltaic / load data, calculate the future strategy from the planning model, and then proceed to step 405;

[0143] Here, the residual-corrected second power prediction value and the second load prediction value are input into the linear programming model to obtain the power charging and discharging strategy for the next time frame at the current moment.

[0144] Step 405: The controller executes the strategy for the next time frame.

[0145] Here, after waiting for one time frame, the controller begins calculating the second power prediction value and the second load prediction value for the next time frame to obtain the energy charging and discharging strategy for the next time frame. To reduce communication burden, the photovoltaic residual prediction model and the load residual prediction model are deployed in the processor of the energy storage scheduling system. In order to continuously predict the second power prediction value and the second load prediction value for the next time frame, and thus obtain the latest energy charging and discharging strategy, steps 403-405 are executed once per time frame.

[0146] The structural diagram of the energy storage dispatch system provided in the application example of this application is as follows: Figure 5As shown, it includes a cloud-based server 501, an embedded controller 502, a temperature, humidity and light sensor 503, a PCS 504, a BMS 505, a PVI 506, and a load 507; among them, the prediction models (photovoltaic prediction model and load prediction model) are deployed in the server 501, and the residual models (photovoltaic residual prediction model and load residual prediction model) are deployed in the embedded controller.

[0147] Based on the above analysis, this application uses the trained photovoltaic prediction model, load prediction model and corresponding residual model to predict the photovoltaic power generation and load for the next day, respectively. Then, the photovoltaic power prediction value and load prediction value for the next time frame are corrected using the residual model, and other time frames are not processed. Thus, the corrected photovoltaic power prediction value sequence and load prediction value sequence are obtained.

[0148] The solution proposed in this application has at least the following advantages:

[0149] 1. By combining real-time data from temperature, humidity and illumination sensors with the LSTN-XGBoost residual correction model, the accuracy of photovoltaic forecasting is significantly improved compared to forecasting methods using a single weather forecast element.

[0150] 2. By utilizing the rolling optimization MPC strategy, this solution significantly improves peak-valley arbitrage returns compared to the traditional static strategy, while also having a stronger ability to suppress microgrid power fluctuations.

[0151] 3. Low-cost temperature, humidity and light sensors replace high-precision meteorological services, resulting in lower overall system costs.

[0152] 4. Due to optimizations in model deployment, the real-time response performance of the strategy has been significantly improved.

[0153] In order to implement the method of the embodiments of this application, the embodiments of this application also provide a device for determining the energy charging and discharging strategy. This device corresponds to the above-described method for determining the energy charging and discharging strategy, and each step in the embodiments of the above-described method for determining the energy charging and discharging strategy is also fully applicable to the embodiments of this device.

[0154] like Figure 6 As shown in the figure, this application provides a device for determining an energy charging and discharging strategy, which is applied to an energy storage dispatching system. The energy storage dispatching system is connected to the power grid and includes photovoltaic panels, photovoltaic inverters, energy storage converters and energy storage batteries. The device includes: an acquisition module 601, a first determination module 602, a second determination module 603 and a third determination module 604.

[0155] The acquisition module 601 is used to acquire the first information representing the current weather conditions;

[0156] The first determining module 602 is used to determine the first power prediction value of the photovoltaic inverter based on the photovoltaic prediction model, and to determine the first load prediction value of the energy storage dispatch system based on the load prediction model.

[0157] The second determining module 603 is used to determine a second power prediction value based on the first information, the current power value of the photovoltaic inverter, the first power prediction value and the photovoltaic residual prediction model; and to determine a second load prediction value based on the first information, the current load value of the energy storage scheduling system, the first load prediction value and the load residual prediction model.

[0158] The third determining module 604 is used to determine the energy charging and discharging strategy based on the second power prediction value, the second load prediction value and the linear programming model; the energy charging and discharging strategy is used to reduce the operating cost of the energy storage scheduling system.

[0159] In some embodiments, the first determining module 602 is specifically used for:

[0160] Obtain second information representing the weather conditions in the first time period of the forecast;

[0161] Based on the second information and the photovoltaic prediction model, the first power prediction value of the photovoltaic inverter in the target time period is determined; and based on the second information and the load prediction model, the first load prediction value of the energy storage dispatch system in the target time period is determined.

[0162] The first time period and the target time period are updated based on the length of the target time period to obtain at least one first power prediction value and one first load prediction value;

[0163] The first time period is earlier than the target time period.

[0164] In some embodiments, the photovoltaic prediction model includes a first linear layer, a Long Short-Term Memory (LSTM) network model, a second linear layer and an activation function module, and a first determination module 602, specifically used for:

[0165] Based on the first linear layer, the dimension of the second information is increased to a preset dimension;

[0166] The first prediction result is determined based on the second information of a preset dimension and the LSTM model.

[0167] The second linear layer reduces the dimension of the first prediction result to one dimension.

[0168] Based on the one-dimensional first prediction result and the activation function module, a first probability is determined;

[0169] The first power prediction value is determined based on the first probability and the rated power of the photovoltaic inverter.

[0170] In some embodiments, the second determining module 603 is specifically used for:

[0171] The first information, the current power value of the photovoltaic inverter, the first power prediction value, and the current time information are input into the photovoltaic residual prediction model to obtain the first difference;

[0172] The second power prediction value is determined based on the sum of the first difference and the first power prediction value.

[0173] In some embodiments, the second determining module 603 is specifically used for:

[0174] The first information, the current load value of the energy storage scheduling system, the first load prediction value, and the current time information are input into the load residual prediction model to obtain the second difference;

[0175] The second load forecast value is determined based on the sum of the second difference and the first load forecast value.

[0176] It should be noted that the device provided in the above embodiments is only illustrated by the division of the above-described program modules when determining the energy charging and discharging strategy. In practical applications, the above processing can be assigned to different program modules as needed, that is, the internal structure of the device can be divided into different program modules to complete all or part of the processing described above. In addition, the device for determining the energy charging and discharging strategy provided in the above embodiments belongs to the same concept as the embodiments, and its specific implementation process can be found in the method embodiments, which will not be repeated here.

[0177] Based on the hardware implementation of the above program modules, and in order to implement the method for determining the energy charging and discharging strategy in the embodiments of this application, the embodiments of this application also provide an energy storage scheduling system, such as... Figure 7 As shown, the energy storage dispatch system 700 includes at least one processor 701, a memory 702, a user interface 703, and at least one network interface 704. The various components in the energy storage dispatch system 700 are coupled together via a bus system 705. It can be understood that the bus system 705 is used to implement communication between these components. In addition to a data bus, the bus system 705 also includes a power bus, a control bus, and a status signal bus. However, for clarity, in... Figure 7 The general labeled all buses as Bus System 705.

[0178] The processor 701 can deploy photovoltaic residual prediction models and load residual prediction models. The energy storage dispatch system 700 also includes a server, on which photovoltaic prediction models and load prediction models can be deployed.

[0179] The user interface 703 may include a monitor, keyboard, mouse, trackball, click wheel, buttons, touchpad, or touch screen.

[0180] The memory 702 in this embodiment is used to store various types of data to support the operation of the energy storage dispatch system 700. Examples of such data include any computer program used to operate on the energy storage dispatch system 700.

[0181] The embodiments of this application disclose methods applicable to or implemented by processor 701. Processor 701 may be an integrated circuit chip with signal processing capabilities. During implementation, each step can be completed through integrated logic circuits in the hardware of processor 701 or instructions in software form. The processor 701 described above may be a general-purpose processor, a digital signal processor (DSP), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Processor 701 can implement or execute the methods, steps, and logic block diagrams disclosed in the embodiments of this application. A general-purpose processor may be a microprocessor or any conventional processor, etc. The steps of the methods disclosed in the embodiments of this application can be directly manifested as execution by a hardware decoding processor, or execution by a combination of hardware and software modules in the decoding processor. The software modules may be located in a storage medium, specifically memory 702. Processor 701 reads information from memory 702 and, in conjunction with its hardware, completes the steps provided in the embodiments of this application.

[0182] In an exemplary embodiment, the energy storage dispatch system 700 may be implemented by one or more application-specific integrated circuits (ASICs), DSPs, programmable logic devices (PLDs), complex programmable logic devices (CPLDs), FPGAs, general-purpose processors, controllers, microcontrollers (MCUs), microprocessors, or other electronic components to perform the aforementioned methods.

[0183] It is understood that memory 702 can be volatile memory or non-volatile memory, or both. Non-volatile memory can be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), EEPROM, ferromagnetic random access memory (FRAM), flash memory, magnetic surface memory, optical disc, or compact disc read-only memory (CD-ROM); magnetic surface memory can be disk storage or magnetic tape storage. Volatile memory can be random access memory (RAM), which is used as an external cache. By way of example, but not limitation, many forms of RAM are available, such as Static Random Access Memory (SRAM), Synchronous Static Random Access Memory (SSRAM), Dynamic Random Access Memory (DRAM), Synchronous Dynamic Random Access Memory (SDRAM), Double Data Rate Synchronous Dynamic Random Access Memory (DDRSDRAM), Enhanced Synchronous Dynamic Random Access Memory (ESDRAM), Sync Link Dynamic Random Access Memory (SLDRAM), and Direct Rambus Random Access Memory (DRRAM). The memory 701 described in this application embodiment is intended to include, but is not limited to, these and any other suitable types of memory.

[0184] In an exemplary embodiment, this application also provides a storage medium, namely a computer storage medium, specifically a computer-readable storage medium, such as a memory 702 storing a computer program. This computer program can be executed by the processor 701 of the energy storage scheduling system 700 to complete the steps described in the method for determining the energy charging and discharging strategy of this application embodiment. The computer-readable storage medium can be a ROM, PROM, EPROM, EEPROM, Flash Memory, magnetic surface memory, optical disc, or CD-ROM, etc.

[0185] In an exemplary embodiment, this application also provides a computer program product, including a computer program that can be executed by a processor 701 of an energy storage scheduling system 700 to complete the steps described in the method of this application embodiment.

[0186] It should be noted that terms such as "first" and "second" are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. In this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0187] Furthermore, the technical solutions described in the embodiments of this application can be combined arbitrarily without conflict.

[0188] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

Claims

1. A method for determining an electric energy charging and discharging strategy, characterized in that, The method is applied to an energy storage scheduling system connected with a power grid, and the energy storage scheduling system comprises a photovoltaic panel, a photovoltaic inverter, an energy storage converter and an energy storage battery. obtaining first information representing a current weather condition; determining a first power prediction value of the photovoltaic inverter based on a photovoltaic prediction model and determining a first load prediction value of the energy storage scheduling system based on a load prediction model; determining a second power prediction value based on the first information, a current power value of the photovoltaic inverter, the first power prediction value and a photovoltaic residual prediction model and determining a second load prediction value based on the first information, a current load value of the energy storage scheduling system, the first load prediction value and a load residual prediction model; determining an energy charging and discharging strategy based on the second power prediction value, the second load prediction value and a linear programming model, wherein the energy charging and discharging strategy is used to reduce an operation cost of the energy storage scheduling system.

2. The method of claim 1, wherein, The photovoltaic residual prediction model and the load residual prediction model each comprise an XGBoost regression model.

3. The method of claim 1, wherein, The method comprises: obtaining second information representing a weather condition in a predicted first time period; determining the first power prediction value of the photovoltaic inverter in a target time period based on the second information and the photovoltaic prediction model and determining the first load prediction value of the energy storage scheduling system in the target time period based on the second information and the load prediction model; updating the first time period and the target time period based on a time length of the target time period to obtain at least one of the first power prediction value and the first load prediction value; wherein the first time period is earlier than the target time period.

4. The method of claim 3, wherein, The photovoltaic prediction model comprises a first linear layer, a long short-term memory (LSTM) model, a second linear layer and an activation function module, and the method comprises: increasing a dimension of the second information to a preset dimension based on the first linear layer; determining a first prediction result based on the second information of the preset dimension and the LSTM model; reducing a dimension of the first prediction result to one dimension based on the second linear layer; determining a first probability based on the first prediction result of one dimension and the activation function module; determining the first power prediction value based on the first probability and a rated power of the photovoltaic inverter.

5. The method of claim 1, wherein, The method comprises: inputting the first information, the current power value of the photovoltaic inverter, the first power prediction value and current time information into the photovoltaic residual prediction model to obtain a first difference value; determining the second power prediction value based on a sum of the first difference value and the first power prediction value.

6. The method of claim 1, wherein, The load prediction model comprises an XGBoost regression model.

7. The method of claim 1, wherein, The second load prediction value is determined based on the first information, the current load value of the energy storage scheduling system, the first load prediction value, and a load residual prediction model. The first information, the current load value of the energy storage scheduling system, the first load prediction value, and current time information are input into the load residual prediction model to obtain a second difference value. The second load prediction value is determined based on the sum of the second difference value and the first load prediction value.

8. The method of claim 1, wherein, The linear programming model includes an objective function and a constraint condition. The objective function is used to minimize the difference between the first amount and the second amount within a first preset time period. The first amount is determined by the power and the electricity price when the energy storage scheduling system obtains power from the power grid, and the second amount is determined by the power and the electricity price when the energy storage scheduling system outputs power to the power grid. The constraint condition is used to make the energy consumption and the energy acquisition of the energy storage scheduling system the same within the first preset time period.

9. A device for determining an electric energy charging and discharging strategy, characterized in that The device is applied to an energy storage scheduling system connected with a power grid, and includes a photovoltaic panel, a photovoltaic inverter, an energy storage converter, and an energy storage battery. An acquisition module is configured to acquire first information representing a current weather condition. A first determination module is configured to determine a first power prediction value of the photovoltaic inverter based on a photovoltaic prediction model, and determine a first load prediction value of the energy storage scheduling system based on a load prediction model. A second determination module is configured to determine a second power prediction value based on the first information, a current power value of the photovoltaic inverter, the first power prediction value, and a photovoltaic residual prediction model, and determine a second load prediction value based on the first information, a current load value of the energy storage scheduling system, the first load prediction value, and a load residual prediction model. A third determination module is configured to determine an energy charging and discharging strategy based on the second power prediction value, the second load prediction value, and a linear programming model, and the energy charging and discharging strategy is used to reduce the operation cost of the energy storage scheduling system.

10. An energy storage dispatch system, characterized by, The energy storage scheduling system is connected with a power grid, and includes a photovoltaic panel, a photovoltaic inverter, an energy storage converter, and an energy storage battery. The energy storage scheduling system further includes a processor and a memory for storing a computer program capable of running on the processor. When the processor runs the computer program, the processor executes the steps of the method according to any one of claims 1 to 8.

11. The system of claim 10, wherein, The energy storage scheduling system further includes a server on which a photovoltaic prediction model and a load prediction model are deployed. The photovoltaic residual prediction model and the load residual prediction model are deployed on the processor.

12. A computer storage medium having stored thereon a computer program, characterized in that The computer program is executed by the processor to implement the steps of the method according to any one of claims 1 to 8.

13. A computer program product comprising a computer program, characterized in that, The computer program is executed by the processor to implement the steps of the method according to any one of claims 1 to 8.

Images