Virtual power plant scheduling strategy generation method, device and equipment based on digital simulation

Abstract

The invention discloses a virtual power plant scheduling strategy generation method, device and equipment based on digital simulation. The method comprises the following steps: acquiring power data of a power plant in a first time period in a target area; inputting feature data in the power data into a target prediction model, and outputting predicted power data of a second time period; and inputting the output predicted power data into a pre-constructed target function, so that when the target function is solved based on a set constraint condition, an output solving result represents a scheduling strategy of the power equipment of the virtual power plant in the target area in the second time period. The virtual simulation model capable of reflecting the real-time dynamic change of each power device in the virtual power plant is introduced into the constructed prediction model, so that the prediction precision of the prediction model is improved, the scheduling strategy obtained by solving the objective function through the accurate and reliable predicted power data is more reasonable and scientific, and the scheduling efficiency is improved. And finally, reasonable scheduling of each power device of the virtual power plant can be realized, and the purpose of optimizing economical efficiency and system stability is achieved.

Description

Technical Field

[0001] The present application relates to the technical field of power system dispatching, and particularly to a method, device and equipment for generating a virtual power plant dispatching strategy based on digital simulation. Background Art

[0002] With the acceleration of the global energy transformation, the virtual power plant, as a special power plant participating in the power market and the power grid operation power coordination management system, has received extensive attention. Through means such as advanced information and communication technology and control metering technology, it aggregates distributed energy resources (DERs) such as distributed generation, energy storage systems, controllable loads, electric vehicles, etc., and can generate corresponding dispatching strategies through the prediction and analysis of the electricity consumption of each power equipment in the corresponding area.

[0003] Currently, by collecting the electricity consumption data in the target area, a virtual power plant model is constructed, and then the electricity consumption in the future period is predicted to determine the corresponding dispatching strategy.

[0004] However, for the static virtual power plant model constructed above, the generated dispatching strategy is difficult to adapt to the real-time fluctuations of distributed energy, and it is also unable to coordinate the operation efficiency of multiple types of energy equipment at the same time, ultimately resulting in low resource utilization and energy waste. Summary of the Invention

[0005] The purpose of the present application is to provide a method, device and equipment for generating a virtual power plant dispatching strategy based on digital simulation, so as to achieve efficient and accurate dispatching of each power equipment in the virtual power plant.

[0006] To achieve the above purpose, the present application provides the following solutions:

[0007] In the first aspect, the present application provides a method for generating a virtual power plant dispatching strategy based on digital simulation, and the method includes:

[0008] Obtain the power data of the first period in the target area, where the power data includes the charge and discharge amounts and energy storage amounts of each power equipment in the target area;

[0009] Extract features from the power data;

[0010] Input the extracted feature data into a pre-constructed target prediction model, and output the predicted power data of the second period. The target prediction model is constructed based on the LSTM neural network and the virtual simulation model of the power equipment in the target area. The virtual simulation model is used to represent the operating state of the power equipment in the target area, and the first period is earlier than the second period;

[0011] Input the predicted power data output into a pre-constructed objective function, such that when solving the objective function based on set constraints, the solution result corresponding to the objective function is output, and the solution result represents the scheduling strategy of the power equipment in the target area during the second time period.

[0012] Optionally, for the virtual power plant scheduling strategy generation method based on digital simulation provided in this application, constructing the prediction model based on the LSTM neural network and the virtual simulation model of the power equipment in the target area includes:

[0013] Obtain training data, where the training data includes historical power data in the target area;

[0014] Input the training data into an initial prediction model built based on the LSTM neural network, and train the initial prediction model to obtain an intermediate prediction model;

[0015] Input the intermediate prediction result of the intermediate prediction model into the virtual simulation model, such that the virtual simulation model outputs a simulation result, and the simulation result characterizes the simulated power data in the target area;

[0016] Based on the simulation result, adjust the parameters of the intermediate prediction model to obtain a target prediction model.

[0017] Optionally, for the virtual power plant scheduling strategy generation method based on digital simulation provided in this application, the step of inputting the training data into an initial prediction model built based on the LSTM neural network and training the initial prediction model to obtain an intermediate prediction model includes:

[0018] Preprocess the training data;

[0019] Perform serialization processing on the preprocessed training data to obtain corresponding time series data;

[0020] Input the time series data into a pre-constructed LSTM neural network unit to output an initial prediction result;

[0021] Determine a loss function according to the initial prediction result;

[0022] Update the parameters of the LSTM neural network unit according to the loss function, and repeat the above training process until the loss function reaches a set threshold, and the corresponding LSTM neural network unit is the intermediate prediction model.

[0023] Optionally, for the virtual power plant scheduling strategy generation method based on digital simulation provided in this application, the step of inputting the time series data into a pre-constructed LSTM neural network unit to output an initial prediction result includes:

[0024] Based on the forward and backward propagation processing of the forget gate, input gate, output gate and fully connected layer in the LSTM neural network unit, feature extraction is performed on the time series data to obtain the initial prediction result.

[0025] Optionally, for the virtual power plant dispatching strategy generation method based on digital simulation provided by this application, inputting the intermediate prediction result of the intermediate prediction model into the virtual simulation model includes:

[0026] Input the intermediate prediction result into a virtual simulation model constructed based on the following expression, and output the corresponding simulation result:

[0027]

[0028] Among them, SOC(t) is the energy storage state in the virtual power plant system at time t, SOC(t + 1) is the updated energy storage state in the virtual power plant system at time t + 1, Pcharge(t) is the charging power of the energy storage system at time t, Pdischarge(t) is the discharging power of the energy storage system at time t, ηcharge is the charging efficiency, ηdischarge is the discharging efficiency, 1 / ηdischarge is used to characterize the compensation for energy loss during the discharging process, the time t is the first time period, and the time t + 1 is the second time period.

[0029] Optionally, for the virtual power plant dispatching strategy generation method based on digital simulation provided by this application, the objective function is constructed based on minimizing the electricity consumption cost of the target area, and is expressed as:

[0030]

[0031] Among them, Cgrid(t) is the grid power purchase cost at time t, Egrid(t) is the electric energy purchased from the grid at time t, and Closs(t) is the virtual power plant system loss cost at time t.

[0032] Optionally, for the virtual power plant dispatching strategy generation method based on digital simulation provided by this application, the constraint conditions include:

[0033] The sum of the total power generation and charging amount of the virtual power plant system in the target area is equal to the sum of the load power consumption, discharging amount and stored power amount, and the charging amount includes photovoltaic charging and wind charging;

[0034] The charging amount is within the set charging interval;

[0035] The discharging amount is within the set discharging interval;

[0036] The stored power amount is within the set storage interval.

[0037] Optionally, when solving the objective function based on the set constraint conditions in the virtual power plant scheduling strategy generation method provided by this application, the solution result corresponding to the objective function output includes:

[0038] Solve the objective function based on the genetic algorithm, so that when the objective function satisfies the set constraint conditions, output the solution result, and the solution result includes the start and stop times of the power generation equipment and charging equipment in the target area.

[0039] In a second aspect, this application provides a virtual power plant scheduling strategy generation device based on digital simulation, characterized in that the device includes:

[0040] An acquisition module, configured to acquire power data for a first time period in a target area, where the power data includes the charge and discharge amounts and energy storage amounts of each power equipment in the target area;

[0041] A preprocessing module, configured to preprocess the power data;

[0042] A prediction module, configured to input the preprocessed power data into a pre-constructed target prediction model, and output predicted power data for a second time period. The target prediction model is constructed based on an LSTM neural network and a virtual simulation model of the power equipment in the target area, and the virtual simulation model is used to characterize the operating state of the power equipment in the target area. The first time period is earlier than the second time period;

[0043] A solution module, configured to input the output predicted power data into a pre-constructed objective function, so that when solving the objective function based on the set constraint conditions, output the solution result corresponding to the objective function, and the solution result represents the scheduling strategy of the power equipment in the target area in the second time period.

[0044] In a third aspect, this application provides a computer device, which includes: a memory, a processor, and a computer program stored on the memory and executable on the processor. The processor executes the computer program to implement the virtual power plant scheduling strategy generation method based on digital simulation as described in the first aspect.

[0045] According to the specific embodiments provided by this application, the following technical effects are disclosed in this application:

[0046] The method, device, and equipment for generating a virtual power plant scheduling strategy based on digital simulation provided by this application, when formulating the scheduling strategy for the virtual power plant corresponding to the target area, first obtain the power data of the first time period as input data and input it into the prediction model pre-constructed based on the LSTM neural network and the virtual simulation model to predict the power data of the next time period, that is, the second time period. Then, after obtaining the predicted power data, use it as the input data of the objective function and input it into the established objective function to solve the objective function based on the set constraints and obtain the optimal solution that meets the constraints as the scheduling strategy of the virtual power plant in the next time period of the target area, thus realizing the automatic generation of the virtual power plant strategy. Among them, since the constructed prediction model introduces a virtual simulation model that can reflect the real-time dynamic changes of each power equipment in the virtual power plant, it overcomes the limitations of the traditional static model, improves the prediction accuracy of the prediction model, makes the prediction results output by the prediction model more in line with the actual power data, and finally provides a reliable basis for the solution of the objective function. As a result, the scheduling strategy obtained by solving the objective function through accurate and reliable predicted power data is more reasonable and scientific, and finally can realize the reasonable scheduling of each power equipment in the virtual power plant, achieving the purpose of optimizing economy and system stability, and significantly improving the resource utilization efficiency. This method can be widely applied to various scenarios such as residential areas and industrial parks, supporting cross-regional energy scheduling and market transactions. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] In order to more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the following will briefly introduce the drawings required to be used in the embodiments. Obviously, the drawings in the following description are only some embodiments of this application. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0048] Figure 1 Schematic flowchart of the method for generating a virtual power plant scheduling strategy based on digital simulation in some embodiments of this application;

[0049] Figure 2 Schematic structural diagram of data acquisition in some embodiments of this application;

[0050] Figure 3 Schematic flowchart of the method for generating a virtual power plant scheduling strategy based on digital simulation in some other embodiments of this application;

[0051] Figure 4 Schematic flowchart of the construction of the prediction model in some embodiments of this application;

[0052] Figure 5 Schematic structural diagram of the device for generating a virtual power plant scheduling strategy in some embodiments of this application;

[0053] Figure 6 A schematic structural diagram of a computer device provided by some embodiments of the present application. Detailed implementation manners

[0054] Next, the technical solutions in the embodiments of the present application will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present application.

[0055] To make the above objects, features, and advantages of the present application more obvious and understandable, the present application will be further described in detail below in conjunction with the accompanying drawings and specific implementation manners.

[0056] It can be understood that for the electricity consumption situation in a certain area, such as a residential area in a certain city or an industrial park in a certain city, in order to save energy, a corresponding prediction model can be constructed and a virtual power plant can be built to predict the power consumption in a future time period in a certain area. Then, according to the prediction results, each power device involved in the corresponding virtual power plant can be reasonably scheduled to achieve the purpose of reducing costs and saving energy.

[0057] It can also be understood that a virtual power plant integrates and optimizes distributed energy resources through software systems and communication technologies, enabling it to participate in the power market and grid operation like a traditional large-scale power plant. Its core lies in "communication" and "aggregation", and through the coordinated control of distributed energy, the reasonable allocation and utilization of resources are realized.

[0058] That is, it can act as an independent market entity, participate in the electricity spot market, ancillary service market, etc., and provide electricity supply and ancillary services.

[0059] In the related art, by constructing a static virtual power plant to predict each electric energy parameter in the future time period, it is difficult to adapt to the real-time fluctuations of distributed energy and difficult to coordinate the operation efficiency of multiple types of energy devices at the same time, resulting in low resource utilization rate.

[0060] In the present application, in order to improve the prediction accuracy and thus ensure the rationality and scientificity of the generated scheduling strategy, the real-time operation data is dynamically coupled with the simulation model, and in combination with the multi-energy synergy effect and the complexity of the dynamic change of market rules, a dynamic virtual model is constructed to perform real-time simulation on the energy consumption situation in the area. Then, the real-time simulation results are used to optimize and improve the prediction model, so that the prediction model can accurately predict the power data in the future time period. Finally, using the accurate predicted power data, an effective and reasonable scheduling strategy is generated by solving the objective function.

[0061] To better understand the method for generating a virtual power plant scheduling strategy based on digital simulation provided in this application, it will be elaborated in detail below with reference to the accompanying drawings.

[0062] As Figure 1 shown in the schematic flowchart of the method for generating a virtual power plant scheduling strategy based on digital simulation provided in the embodiment of this application, as Figure 1 shown, the method includes:

[0063] S110. Obtain the power data of the first time period in the target area, where the power data includes the charge and discharge amounts and energy storage amounts of each power device in the target area.

[0064] S120. Extract features from the power data.

[0065] S130. Input the preprocessed power data into a pre-constructed target prediction model, and output the predicted power data of the second time period. The target prediction model is constructed based on an LSTM neural network and a virtual simulation model of the power devices in the target area. The virtual simulation model is used to characterize the operating state of the power devices in the target area. The first time period is earlier than the second time period.

[0066] S140. Input the output predicted power data into a pre-constructed target function. When solving the target function based on the set constraint conditions, output the solution result corresponding to the target function. The solution result represents the scheduling strategy of the power devices in the target area in the second time period.

[0067] Specifically, as combined with Figure 3 shown, first, the power data of each power device, that is, distributed energy, in the target area in the first time period, such as the current power data, can be collected through IoT devices. For example, the power generation amounts, energy storage states, etc. of each power generation device in the current time period in the target area can be collected.

[0068] For example, if the target area is a certain city or a certain industrial park, then in the distributed energy scenario of a residential area in a certain city, the power data such as the current photovoltaic power generation amount, energy storage amount, and electricity consumption load of each household can be collected.

[0069] Or, for the distributed energy scenario of a certain industrial park, which is connected to multiple energy devices such as wind power, photovoltaic, and gas generators. Then, the operation parameters and other power data of each device in the park can be collected.

[0070] Furthermore, the collected data can be preprocessed to extract the feature data in the power data.

[0071] For example, the collected power data can be first cleaned and standardized, and then features can be extracted from the preprocessed power data to obtain the feature data.

[0072] That is, outliers can be removed first, and missing data can be filled (for example, using the moving average method) to complete data cleaning. Furthermore, by normalizing the power data of different devices, the consistency of the input to the model is ensured to complete the standardization process. Finally, key features (such as daily average power generation, historical SOC change trend, relationship between temperature and light, etc.) are extracted from the power data that has completed the above processing, that is, the original data.

[0073] Furthermore, through the above steps, after extracting the features in the collected data, the feature data can be input into a pre-constructed target prediction model to predict the energy operation data in the second period, such as the power data in the next period, and output the predicted power data.

[0074] Among them, the target prediction model can be constructed based on the LSTM neural network and the virtual simulation model of the power equipment in the target area.

[0075] The virtual simulation model is used to represent the operating state of the power equipment in the target area, that is, it can visually represent the dynamic operation of each power equipment in the target area to simulate the real-time dynamic behavior of distributed energy.

[0076] In practice, during the construction of the prediction model, the virtual simulation model is used to adjust and optimize the parameters of the prediction model so that the finally generated target prediction model can output accurate prediction data.

[0077] Finally, the output result of the target prediction model, that is, the predicted power data in the second period, can be input into a pre-constructed target strategy model, that is, the established objective function. Then, by solving the objective function, on the basis of satisfying each set constraint condition, the corresponding solution result is output, and this solution result can be used as the scheduling strategy of each power equipment in the target area in the second period.

[0078] The objective function can be constructed based on cost minimization or revenue maximization. Correspondingly, the solution result output, that is, the variable corresponding to the objective function, can specifically include the electricity consumption required in the next period.

[0079] Then, this scheduling strategy, as the scheduling strategy of the virtual power plant, can be used to represent the operating state of each device in the target area in the next period. That is, on the basis of satisfying this electricity consumption, the start-stop time and operation duration of each power equipment.

[0080] For example, in the above urban residential area scenario, the output result of the solution is the optimal power scheduling plan for charging and discharging of each user in the next period.

[0081] Specifically, the output strategy can include a set of 24-hour time series scheduling instructions, such as the concentrated charging duration during the low electricity price period; the discharging duration during the high electricity price and peak demand periods.

[0082] Finally, after the multi-household strategies are aggregated, a trading order is submitted to the power market in the form of a virtual power plant, realizing collaborative participation in market transactions and ensuring the optimal overall economic benefits.

[0083] That is, the scheduling strategy of the virtual power plant generated in the embodiments of the present application can be used as a participation basis when the virtual power plant is an independent market entity. Based on the scheduling strategy, it participates in the power spot market, ancillary service market, etc., and finally provides power supply and ancillary services, enabling the optimization of the scheduling of power generation, energy storage, and load according to the grid demand and market price signals.

[0084] For example, during peak electricity consumption, the virtual power plant can release the electricity stored in the energy storage or reduce the electricity consumption of controllable loads; during low electricity consumption, it can store the excess electricity or adjust the power generation plan.

[0085] It can be understood that in the embodiments of the present application, when formulating the scheduling strategy of the virtual power plant in the target area, first, the power data of the first period is obtained as input data and input into a prediction model pre-constructed based on the LSTM neural network and the virtual simulation model to predict the power data of the next period, that is, the second period. Then, after obtaining the predicted power data, it is used as the input data of the objective function and input into the established objective function to solve the objective function based on the set constraint conditions to obtain the optimal solution that meets the constraint conditions, which is used as the scheduling strategy of the virtual power plant in the next period of the target area, thus realizing the automatic generation of the virtual power plant strategy. Among them, since the constructed prediction model introduces a virtual simulation model that can reflect the real-time dynamic changes of each power device in the virtual power plant, it overcomes the limitations of traditional static models, improves the prediction accuracy of the prediction model, makes the prediction results output by the prediction model more in line with the actual power data, and finally provides a reliable basis for the solution of the objective function. Therefore, the scheduling strategy obtained by solving the objective function through accurate and reliable predicted power data is more reasonable and scientific, and finally can realize the reasonable scheduling of each power device in the virtual power plant, achieving the purpose of optimizing economy and system stability, and significantly improving the resource utilization efficiency. This method can be widely applied to various scenarios such as residential buildings and industrial parks, and supports cross-regional energy scheduling and market transactions.

[0086] It can be understood that in practice, after obtaining the scheduling strategy by solving the objective function, the obtained scheduling strategy can be output to the user side or the power market interface so that the corresponding execution entity can execute the scheduling strategy.

[0087] Optionally, in some embodiments, such as Figure 2As shown, the real-time power data of each power device in the target area, that is, the operation status data, can be collected through the built data collection module.

[0088] That is, the power data of each front-end device can be collected by building an equipment access layer (such as sessions, channels, etc.), a data collection layer (i.e., through operations such as encoding and decoding), and a conversion layer.

[0089] It can be understood that, as Figure 3 and Figure 4 shown, in the embodiments of the present application, for the prediction model of the above embodiments, it can be pre-constructed by collecting historical data and training the initial model based on the built initial model.

[0090] That is, based on the pre-constructed virtual simulation model and LSTM, the target prediction model is constructed.

[0091] Specifically, in the embodiments of the present application, for the construction of the prediction model, in order to improve the prediction accuracy and efficiency, the following steps are specifically included:

[0092] S01, Obtain training data, which includes historical power data in the target area.

[0093] S02, Input the training data into an initial prediction model built based on the LSTM neural network, and train the initial prediction model to obtain an intermediate prediction model.

[0094] S03, Input the intermediate prediction result of the intermediate prediction model into the virtual simulation model, so that the virtual simulation model outputs a simulation result, which represents the simulated power data of the virtual power plant corresponding to the target area.

[0095] S04, Based on the simulation result, adjust the parameters of the intermediate prediction model to obtain a target prediction model.

[0096] Specifically, in the embodiments of the present application, in order to improve the prediction result of the prediction model, the virtual simulation result of the built power equipment is used to verify and optimize the prediction model, so that the constructed target prediction model can better fit the dynamic changes of each power equipment.

[0097] Then, in the actual construction process, first, sample data can be obtained, that is, historical power data in the target area can be obtained as training data.

[0098] Furthermore, after obtaining the training data, model training and construction are carried out. Specifically, in S02, the following steps can be included:

[0099] S001, Preprocess the training data;

[0100] S002 Serialize the preprocessed training data to obtain corresponding time series data;

[0101] S003 Input the time series data into a pre-constructed LSTM neural network unit to output an initial prediction result;

[0102] S004 Determine a loss function based on the initial prediction result;

[0103] S005 Update the parameters of the LSTM neural network unit according to the loss function, and repeat the above training process until the loss function reaches a set threshold. The corresponding LSTM neural network unit is the intermediate prediction model.

[0104] Specifically, when constructing the model, first, preprocessing operations and feature extraction can be performed on the training data.

[0105] For example, it can include cleaning, standardization, and feature extraction, etc.

[0106] Specifically, the cleaning process, that is, removing outliers and filling in missing data (for example, using the moving average method). Standardization, that is, normalizing the data of different devices to ensure the consistency of the input to the model. Feature extraction, that is, extracting key features from the original data (such as daily average power generation, historical SOC change trend, relationship between temperature and light, etc.).

[0107] Furthermore, the extracted feature data is normalized by the following normalization formula to ensure that different features are on the same scale.

[0108] Data normalization formula:

[0109]

[0110] Where X is the original value of a certain feature, such as: the photovoltaic power generation power of a certain household during a certain period or the SOC of the energy storage system.

[0111] Xmin is the minimum value observed for this feature in the entire dataset.

[0112] Xmax is the maximum value observed for this feature in the entire dataset.

[0113] X′ is the normalized data value.

[0114] It can be understood that through the above normalization process, the original data X can be mapped to the interval [0,1], so that when subsequent models (such as LSTM) are trained, the feature values are on the same scale, which is convenient for the convergence of the gradient descent process.

[0115] Further, the normalized training data can be serialized to obtain corresponding sequence data.

[0116] In practice, the time step can be defined first to determine the length of each input sequence, and then input-output pairs can be created, that is, historical data is used as input and future data is used as output. For example, the data of the first 10 time steps are used to predict the value of the 11th time step.

[0117] That is, the sliding window method is used to construct time series data, such as:

[0118] X = [P t-3 , P t-2 , P t-1 ;

[0119] where X: the input feature vector at the current time step.

[0120] Pt-3, Pt-2, Pt-1: data points of the previous three time steps.

[0121] Sliding window: using past data as input, commonly used in time series prediction.

[0122] Target output: Y = Pt;

[0123] That is, using the sliding window method, the historical data of each household is constructed into a time series.

[0124] It can be understood that each sample data can include features of k consecutive time steps, such as photovoltaic power generation, SOC of the energy storage system, SOC of the load system, environmental temperature, light intensity, etc.

[0125] It can be understood that the data representation in practice can be a sequence structure.

[0126] For example, a sample with a window size of 3 may be represented as:

[0127] [PV t , SOC t , temp t , irradiance t , PV t+1 ,…, PV t+2 ,…];

[0128] Further, the serialized sequence data can be input into a pre-built initial model to train the initial model.

[0129] It can be understood that the initial training model is built based on the LSTM neural network, and the initial model can include an LSTM layer, a fully connected layer, and an output layer. Then, the training of the initial model can be an iterative process of processing the input data through each module and repeating it multiple times.

[0130] That is, in practice, in this application, the model structure can be predefined, that is, the initial architecture of the LSTM model can be defined using a deep learning framework (such as TensorFlow, PyTorch).

[0131] Moreover, after building the initial module, the parameters can also be initialized, that is, the parameters in the LSTM model are initialized, such as weight and bias parameters, etc.

[0132] Furthermore, a loss function can be selected. For example, an appropriate loss function can be selected according to the task type. For example, the mean squared error (MSE) is used for regression tasks, and the cross-entropy loss is used for classification tasks.

[0133] Finally, an optimizer can be selected, that is, an optimization algorithm can be selected, such as Adam, SGD, etc., for updating the model parameters.

[0134] To better understand the model training process, the processing process of each unit module for the serialized data is elaborated in detail below:

[0135] Processing step by step: In the LSTM layer, the input x t (including multiple features) at each time step t enters the LSTM cell.

[0136] Calculations within the cell are as follows:

[0137] Forget gate: Determines the proportion of information from the previous time step to be retained:

[0138] f t = σ(W f ·[h t-1 , x t + b f );

[0139] where, xt: The input vector at time period t (for example, including features such as the normalized photovoltaic power generation, the SOC of the energy storage system, the ambient temperature, and the light intensity).

[0140] h t-1 : The hidden state at the previous time period t-1, which contains the information of the previous time window.

[0141] [h t-1 , x t : The combined vector after concatenating h t-1 and x t .

[0142] W f : Weight matrix of the forget gate.

[0143] b f : Bias of the forget gate.

[0144] σ(·): Sigmoid function, whose output value ranges from 0 to 1, representing the retention ratio of each piece of information.

[0145] f t : Output of the forget gate at time step t, used to determine how much old information to forget.

[0146] It can be understood that the above processing process means using the hidden state h of the previous time step t-1 and the current input x t to update the internal state and generate the output of the current time step. Determine how much information of the cell state of the previous time step to retain.

[0147] Furthermore, the input gate and the candidate cell state: Determine the importance of the current input information.

[0148] i t = σ(W i ·[h t-1 , x t + b i );

[0149] C~t = tanh(WC·[h t-1 , x t + b C );

[0150] Among them, Wi, bi: Weight matrix and bias corresponding to the input gate.

[0151] W C , b C : Weight matrix and bias used to generate the candidate cell state.

[0152] tanh(·): Hyperbolic tangent function, mapping the candidate state to the range [-1, 1].

[0153] it: Output of the input gate at time step t, with a value between 0 and 1.

[0154] C~t: Candidate cell state vector at time step t.

[0155] It can be understood that the above processing process input gate i t can determine which information in the current input will be written into the cell state. The candidate cell state C~t can generate new candidate information using the current input and the previous hidden state.

[0156] State update: Update the cell state,

[0157] C t = f t × C t-1 + i t × C ~ t;

[0158] Where C t-1 : The cell state of the previous time period.

[0159] f t × C t-1 : The forgotten part, i.e., the old information retained after passing through the forget gate.

[0160] i t × C ~ t: The new input part, i.e., the new information introduced after passing through the input gate.

[0161] C t : The cell state updated at time period t.

[0162] It can be understood that through the above operations, the cell state can be updated, that is, by combining the retention of the old state by the forget gate and the introduction of the new candidate state by the input gate, to form the complete memory of the current time period.

[0163] Output gate: Generates the hidden state of the current time step,

[0164] o t = σ(Wo · [h t-1 , x t + b o );

[0165] h t = o t × tanh(C t );

[0166] Where W o , b o : The weight matrix and bias of the output gate.

[0167] o t : The output gate value at time period t, ranging from 0 to 1.

[0168] tanh(Ct): Compresses the cell state Ct to the interval [-1, 1] to generate a candidate output.

[0169] h t : The hidden state at time period t, which is the cell state information controlled by the output gate.

[0170] It can be understood that for the output gate o t : Determines which cell state information is transmitted to the final output.

[0171] Hidden state ht : The output of the current time period serves as one of the inputs for the next time period.

[0172] Output selection: Usually, the hidden state h of the last time step is adopted. t + k - 1 serves as the summary representation of the entire time window.

[0173] Tools and implementation: Implement the LSTM layer using the TensorFlow / Keras or PyTorch framework and accelerate it with GPU. During the training process, backpropagation through time (BPTT) is used to update the weights of each matrix.

[0174] Fully connected layer:

[0175] First, input transformation: Use the hidden state hfinal output by the LSTM layer as the input. This vector contains the key information of k consecutive time periods.

[0176] Furthermore, linear transformation and activation: Perform a linear transformation through the fully connected layer:

[0177] z = W fc ·h final + b fc ;

[0178] h final : The hidden state of the last time period within the sliding window, representing the comprehensive features of the entire sequence.

[0179] W fc : The weight matrix of the fully connected layer.

[0180] b fc : The bias term of the fully connected layer.

[0181] z: The feature vector output by the fully connected layer, used for subsequent prediction.

[0182] That is, through the above processing, a linear transformation can be performed on the hidden state hfinal of the last time step of the LSTM layer, mapping it to a new feature vector z.

[0183] Finally, this layer can integrate the temporal features extracted by the LSTM to form an intermediate feature vector suitable for processing by the output layer. Moreover, it is implemented using the same deep learning library, and the parameters of the fully connected layer are updated using the Adam optimizer during training.

[0184] Output layer:

[0185] First, prediction target: The output layer uses a single neuron (or multiple neurons for multi-target prediction) to directly output the predicted power value for the next time period (for example, predicting the photovoltaic power generation for the next hour).

[0186] Furthermore, linear activation: For continuous value prediction, the output layer usually adopts a linear activation function, that is, directly outputs the value after the transformation of the fully connected layer:

[0187] ŷ = W out ·z + b out

[0188] where, W out : The weight matrix of the output layer.

[0189] b out : The bias term of the output layer.

[0190] ŷ: The predicted output value, representing the power value in the next period.

[0191] It can be understood that through the above processing, the feature vector z integrated by the fully connected layer can be mapped to a continuous value, which is the predicted photovoltaic power generation (or other target quantity) in the next period.

[0192] Moreover, the integrated feature is mapped to a specific predicted value and compared with the actual measurement data.

[0193] It can be understood that the prediction result can include the initial prediction result. Then, the current loss function is calculated through the prediction result, and the loss function is compared with the set loss function threshold to return the parameters in the updated model. Then, the above operations are repeated to make the final output result satisfy the set loss function threshold.

[0194] It can also be understood that after the above processing, the constructed prediction model serves as an intermediate prediction model, and the corresponding output result serves as an intermediate prediction result.

[0195] Furthermore, the intermediate prediction result can be input into the constructed virtual simulation model to verify the intermediate prediction result.

[0196] That is, the intermediate prediction result is input into the pre-constructed virtual simulation model, so that each power device in the virtual power plant operates according to the intermediate prediction result to simulate the dynamic behavior of distributed energy.

[0197] In practice, the power generation data, power consumption data, etc. in the intermediate prediction result are input into the virtual simulation model, which can be specifically expressed by the following formula:

[0198] SOC(0) = S(0);

[0199]

[0200] where, SOC(0) represents the end value of the previous time as the initial value of this time, and SOC(t): the state of the energy storage system at time t, usually expressed as a percentage.

[0201] SOC(t + 1): The updated energy storage state at time period t + 1.

[0202] Pcharge(t): The charging power of the energy storage system at time period t (unit: kW).

[0203] Pdischarge(t): The discharging power of the energy storage system at time period t (unit: kW).

[0204] ηcharge: Charging efficiency (a value between 0 and 1, considering energy losses).

[0205] ηdischarge: Discharging efficiency (also between 0 and 1); in the formula, 1 / ηdischarge is used to reflect the compensation for energy losses during the discharging process.

[0206] It can be understood that the above expressions can represent the increased energy during charging at time period t (after multiplying by the charging efficiency) and the decreased energy during discharging (considering efficiency losses during discharging), which jointly determine the next time period. That is, as the core representation of the virtual power plant, this expression represents the power balance situation within the virtual power plant.

[0207] Then, in the embodiments of this application, the above intermediate prediction data is input into the virtual simulation model, enabling the virtual simulation model to dynamically simulate the dynamic operating states of various power equipment, that is, to simulate the dynamic behaviors of distributed energy sources, so as to determine whether the prediction data in the intermediate prediction results is reasonable, that is, whether each power equipment can meet the electricity consumption demands of each user during operation, or whether the storage capacity of the storage system reaches the set requirements, etc.

[0208] For example, the dynamic adjustment state of the device load can be output:

[0209] P net (t) = P generation (t) - P load (t);

[0210] Where, Pgeneration(t): The total power generation at time period t (from photovoltaic, wind power, etc.).

[0211] Pload(t): The total electricity load at time period t.

[0212] Pnet(t): Net power, a positive value indicates power generation surplus, and a negative value indicates electricity consumption exceeding power generation

[0213] It can be understood that this formula is used to determine whether the system needs additional scheduling (such as energy storage discharging or charging) during a certain time period.

[0214] Further, based on the virtual simulation results, the parameters of the intermediate prediction model can be adjusted to ultimately optimize the prediction model and construct a target prediction model with high output accuracy.

[0215] It can be understood that by comparing the above simulation results with the actual measurement data and using the error feedback mechanism to optimize the model, the prediction accuracy can be improved.

[0216] Optionally, in some embodiments of the present application, the construction of a virtual simulation model may further be included, that is, the energy relationship of the virtual power plant is represented by a function.

[0217] In practice, the collected training data or prediction data can be used as the input data for constructing the virtual simulation model.

[0218] That is, in some embodiments, the prediction model and the virtual simulation model can be constructed synchronously and optimized with each other.

[0219] Optionally, in some embodiments of the present application, the output prediction power data can be used to solve the objective function to generate the final scheduling strategy.

[0220] That is, in S140, the optimal scheduling module generates the charge and discharge strategy and coordinates multiple household devices to jointly participate in the electricity market transaction. The implementation steps are as follows: The prediction model and the virtual simulation model constructed through the above steps can accurately output the power data for the future time period.

[0221] In practice, problem modeling can be performed first, that is, the objective function, constraints, and variables are set.

[0222] Specifically, the objective function can be constructed based on cost minimization or revenue maximization first.

[0223] For example, in some embodiments, the objective function constructed based on cost minimization is expressed as follows:

[0224]

[0225] Among them, t represents the time period (for example, each hour in the next 24 hours).

[0226] Cgrid(t): The grid power purchase cost in time period t (unit cost, usually yuan / kWh). (2)

[0227] Egrid(t): The electric energy purchased from the grid in time period t (unit: kWh).

[0228] Closs(t): The system loss cost in time period t, including energy storage loss, equipment wear, etc.

[0229] It can be understood that for C in the above items loss(t), i.e., the virtual power plant system loss cost and C grid (t) the grid power purchase cost, which can usually be an empirical value.

[0230] For the first item E grid (t), i.e., the electric energy purchased by the grid, which can be related to the power data output by the above prediction model.

[0231] It can be understood that this objective function accumulates the costs for each time period, and the optimization algorithm will select the charge and discharge strategy to minimize this total cost.

[0232] Furthermore, for the set constraint conditions, specifically, they can include:

[0233] The sum of the total power generation and the charging amount of the virtual power plant system corresponding to the target area is equal to the sum of the load power consumption, the discharging amount, and the stored power amount. The charging amount includes photovoltaic charging and wind power charging;

[0234] The charging amount is within the set charging interval;

[0235] The discharging amount is within the set discharging interval;

[0236] The stored power amount is within the set storage interval.

[0237] Specifically, the charging / discharging power of each household's energy storage device and the power flow in and out of the grid side can be defined as variables.

[0238] Constraint conditions: including energy storage capacity, charge and discharge efficiency, upper and lower limits of device power, and power balance constraint.

[0239] Similarly, using the power balance formula, that is:

[0240]

[0241] It can be understood that by combining this power balance formula, the energy increased during charging (after multiplying by the charging efficiency) and the energy decreased during discharging (considering efficiency losses during discharging) in time period t together determine the remaining energy of the battery in the next time period. That is, the sum of the total power generation and the charging amount of the virtual power plant system corresponding to the target area is equal to the sum of the load power consumption, the discharging amount, and the stored power amount. The charging amount includes photovoltaic charging and wind power charging

[0242] In addition, at the same time, ensure that the charging amount is within the set charging interval; the discharging amount is within the set discharging interval; the stored power amount is within the set storage interval.

[0243] It is expressed as follows:

[0244] Pgeneration(t) + Pcharge(t) = Pload(t) + Pdischarge(t) + Pgrid(t);

[0245] 0 ≤ Pcharge(t) ≤ Pcharge_max, 0 ≤ Pdischarge(t) ≤ Pdischarge_max;

[0246] SOCmin ≤ SOC(t) ≤ SOCmax.

[0247] To ensure that the battery operates within a safe range, extend the device life and maintain system stability.

[0248] Where:

[0249] SOC(t): State of the energy storage device at time period t (usually in percentage).

[0250] SOCmin: The lowest allowable state of the energy storage device (e.g., 20%), to prevent over-discharge from damaging the battery.

[0251] SOCmax: The highest allowable state of the energy storage device (e.g., 95%), to prevent over-charging and potential safety hazards.

[0252] Furthermore, for the above constructed objective function, when solving, mixed-integer linear programming (MILP) or genetic algorithm can be used for solving.

[0253] Thus, after inputting the prediction data, device parameters, and market electricity price information, the optimal charging and discharging power scheduling scheme for each household in each time period is output after solving.

[0254] For example, the output strategy can be a set of 24-hour time series scheduling instructions, specifically including:

[0255] Charge concentratedly during low electricity price periods;

[0256] Discharge during high electricity price and peak demand periods.

[0257] After aggregating the multi-household strategies, trading orders are submitted to the power market in the form of a virtual power plant, realizing collaborative participation in market transactions to ensure the optimal overall economic benefits.

[0258] It can also be understood that in some embodiments of the present application, actual power data can also be used for model feedback and real-time adjustment, that is, during the process of generating scheduling strategies using the constructed prediction model, the prediction model and the virtual simulation model can be continuously optimized in real time by continuously collecting real-time data, and then the optimized virtual simulation model can be used to further optimize the prediction model.

[0259] That is, the execution situation is monitored in real time, the model parameters are corrected through the feedback data, and the closed-loop dynamic scheduling optimization is realized. The simulation module verifies the benefits and load balancing effects of the scheduling strategy during peak electricity price periods.

[0260] It can be understood that in the embodiments of the present application, when formulating the scheduling strategy of the virtual power plant in the target area, first, the power data of the first period is obtained as input data and input into the prediction model constructed in advance based on the LSTM neural network and the virtual simulation model to predict the power data of the next period, that is, the second period. Then, after obtaining the predicted power data, it is used as the input data of the objective function and input into the established objective function to solve the objective function based on the set constraint conditions, and the optimal solution that meets the constraint conditions is obtained as the scheduling strategy of the virtual power plant in the next period of the target area, thus realizing the automatic generation of the virtual power plant strategy. Among them, since the constructed prediction model introduces a virtual simulation model that can reflect the real-time dynamic changes of each power device in the virtual power plant, it overcomes the limitations of traditional static models, improves the prediction accuracy of the prediction model, makes the prediction results output by the prediction model more in line with the actual power data, and finally provides a reliable basis for the solution of the objective function. Therefore, the scheduling strategy obtained by solving the objective function through accurate and reliable predicted power data is more reasonable and scientific, and finally can realize the reasonable scheduling of each power device in the virtual power plant, achieving the purpose of optimizing economy and system stability, and significantly improving the resource utilization efficiency. This method can be widely applied to various scenarios such as residential areas and industrial parks, and supports cross-regional energy scheduling and market transactions.

[0261] On the other hand, as Figure 5 shown, the present application provides a device for generating a virtual power plant scheduling strategy, and the device includes:

[0262] An acquisition module 210, configured to acquire the power data of the first period in the target area, where the power data includes the charge and discharge amounts and energy storage amounts of each power device in the target area;

[0263] A preprocessing module 220, configured to preprocess the power data and extract the feature data in the power data;

[0264] A prediction module 230, configured to input the extracted feature data into a pre-constructed target prediction model and output the predicted power data of the second period. The target prediction model is constructed based on the LSTM neural network and the virtual simulation model of the power devices in the target area, and the virtual simulation model is used to characterize the operating state of the power devices in the target area. The first period is earlier than the second period;

[0265] A solution module 240, configured to input the outputted predicted power data into a pre-constructed objective function, such that when solving the objective function based on set constraint conditions, an obtained solution result corresponding to the objective function is outputted, and the solution result represents a scheduling strategy of power equipment in the target area during a second time period.

[0266] Optionally, the virtual power plant scheduling strategy generation device provided in the embodiment of the present application further includes a construction module 250, configured to:

[0267] Obtain training data, where the training data includes historical power data in the target area;

[0268] Input the training data into an initial prediction model built based on an LSTM neural network, and train the initial prediction model to obtain an intermediate prediction model;

[0269] Input the intermediate prediction result of the intermediate prediction model into a virtual simulation model, such that the virtual simulation model outputs a simulation result, and the simulation result represents the simulated power data in the target area;

[0270] Based on the simulation result, adjust parameters of the intermediate prediction model to obtain a target prediction model.

[0271] Optionally, for the virtual power plant scheduling strategy generation device provided in the embodiment of the present application, the construction module is specifically configured to:

[0272] Preprocess the training data;

[0273] Perform serialization processing on the preprocessed training data to obtain corresponding time series data;

[0274] Input the time series data into a pre-constructed LSTM neural network unit, and output an initial prediction result;

[0275] Determine a loss function according to the initial prediction result;

[0276] Update parameters of the LSTM neural network unit according to the loss function, and repeat the above training process until the loss function reaches a set threshold, and the corresponding LSTM neural network unit is the intermediate prediction model.

[0277] Optionally, for the virtual power plant scheduling strategy generation device provided in the embodiment of the present application, the construction module is specifically configured to:

[0278] Based on forward propagation and backward propagation processing of a forget gate, an input gate, an output gate, and a fully connected layer in an LSTM neural network unit, extract features from the time series data to obtain the initial prediction result.

[0279] Optionally, for the virtual power plant scheduling strategy generation device provided by the embodiments of the present application, the construction module is specifically configured to:

[0280] Input the intermediate prediction result into a virtual simulation model constructed based on the following expression, and output the corresponding simulation result:

[0281]

[0282] where SOC(t) is the energy storage state in the virtual power plant system at time t, SOC(t + 1) is the updated energy storage state in the virtual power plant system at time t + 1, Pcharge(t) is the charging power of the energy storage system at time t, Pdischarge(t) is the discharging power of the energy storage system at time t, ηcharge is the charging efficiency, ηdischarge is the discharging efficiency, 1 / ηdischarge is used to characterize the compensation for energy loss during the discharging process, the time t is the first time period, and the time t + 1 is the second time period.

[0283] Optionally, for the virtual power plant scheduling strategy generation device provided by the embodiments of the present application, the objective function is constructed based on the minimization of the electricity consumption cost of the target area, and is expressed as:

[0284]

[0285] where Cgrid(t) is the grid power purchase cost at time t, Egrid(t) is the amount of electric energy purchased from the grid at time t, and Closs(t) is the power plant system loss cost at time t.

[0286] Optionally, for the virtual power plant scheduling strategy generation device provided by the embodiments of the present application, the constraint conditions include:

[0287] The sum of the total power generation and the charging amount of the virtual power plant system corresponding to the target area is equal to the sum of the load power consumption, the discharging amount, and the stored electric energy, and the charging amount includes photovoltaic charging and wind power charging;

[0288] The charging amount is within the set charging interval;

[0289] The discharging amount is within the set discharging interval;

[0290] The stored electric energy is within the set storage interval.

[0291] Optionally, for the virtual power plant scheduling strategy generation device provided by the embodiments of the present application, the solving module is specifically configured to:

[0292] Solving the objective function based on a genetic algorithm, such that when the objective function satisfies the set constraint conditions, the solution result is output, where the solution result includes the start and stop times of power generation equipment and charging equipment within the target area.

[0293] In an exemplary embodiment, a computer device is provided. The computer device can be a server or a terminal, and its internal structure diagram can be as Figure 6 shown. The computer device includes a processor, a memory, an input / output interface (Input / Output, abbreviated as I / O), and a communication interface. Among them, the processor, the memory, and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store video tag processing data. The input / output interface of the computer device is used to exchange information between the processor and external devices. The communication interface of the computer device is used to communicate with external terminals through a network connection. When the computer program is executed by the processor, it implements a method for generating a virtual power plant scheduling strategy based on digital simulation.

[0294] Those skilled in the art can understand that Figure 6 the structure shown in

[0295] is only a block diagram of some structures related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0296] In an exemplary embodiment, a computer-readable storage medium is provided, storing a computer program, and when the computer program is executed by a processor, it implements the steps in the above method embodiments.

[0297] In an exemplary embodiment, a computer program product is provided, including a computer program, and when the computer program is executed by a processor, it implements the steps in the above method embodiments.

[0298] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use, and processing of relevant data need to comply with relevant regulations.

[0299] Those of ordinary skill in the art can understand that all or part of the processes in the methods of the above embodiments can be completed by instructing relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above methods. Among them, any reference to a memory, database, or other medium used in the embodiments provided in this application can include at least one of non-volatile and volatile memories. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc.

[0300] The databases involved in the embodiments provided in this application can include at least one of relational databases and non-relational databases. Non-relational databases can include distributed databases based on blockchain, etc., without limitation. The processors involved in the embodiments provided in this application can be general-purpose processors, computer devices, graphics processors, digital signal processors, programmable logics, data processing logics based on quantum computing, etc., without limitation.

[0301] The technical features of the above embodiments can be combined arbitrarily. For the sake of brevity of description, not all possible combinations of the technical features in the above embodiments are described. However, as long as the combinations of these technical features do not conflict, they should be considered to be within the scope described in this specification.

[0302] In this article, specific examples are used to elaborate on the principles and implementation modes of the present application. The description of the above embodiments is only used to help understand the method and its core idea of the present application. At the same time, for those of ordinary skill in the art, according to the idea of the present application, there will be changes in the specific implementation modes and application scopes. In summary, the content of this specification should not be construed as a limitation to the present application.

Claims

1. A method for generating a virtual power plant dispatching strategy based on digital simulation, characterized in that: The method comprises: Acquire power data of a first period of time in a target area, wherein the power data includes charge and discharge capacity and energy storage capacity of each power device in the target area; Preprocessing the power data to extract characteristic data from the power data; Input the characteristic data into a pre-built target prediction model, and output predicted power data for the second time period, wherein the target prediction model is built based on an LSTM neural network and a virtual simulation model of the power equipment in the target area, and the virtual simulation model is used to characterize the operating status of the power equipment in the target area, and the first time period is earlier than the second time period; The output predicted power data is input into a pre-constructed objective function, so that when the objective function is solved based on the set constraints, the solution result corresponding to the objective function is output, and the solution result represents the scheduling strategy of the power equipment in the target area in the second time period.

2. The method for generating a virtual power plant dispatching strategy based on digital simulation according to claim 1, characterized in that: Constructing the prediction model based on the LSTM neural network and the virtual simulation model of the target area power equipment includes: Acquiring training data, wherein the training data includes historical power data in the target area; Inputting the training data into an initial prediction model built based on an LSTM neural network, training the initial prediction model, and obtaining an intermediate prediction model; Inputting the intermediate prediction result of the intermediate prediction model into the virtual simulation model, so that the virtual simulation model outputs a simulation result, wherein the simulation result represents the simulated power data in the target area; Based on the simulation results, the parameters of the intermediate prediction model are adjusted to obtain a target prediction model.

3. The method for generating a virtual power plant dispatching strategy based on digital simulation according to claim 2 is characterized in that: The step of inputting the training data into an initial prediction model based on an LSTM neural network, training the initial prediction model, and obtaining an intermediate prediction model comprises: Preprocessing the training data; Performing serialization processing on the preprocessed training data to obtain corresponding time series data; Input the time series data into a pre-built LSTM neural network unit and output an initial prediction result; Determine a loss function according to the initial prediction result; The parameters of the LSTM neural network unit are updated according to the loss function, and the above training process is repeated until the loss function reaches a set threshold, and the corresponding LSTM neural network unit is the intermediate prediction model.

4. The method for generating a virtual power plant dispatching strategy based on digital simulation according to claim 3 is characterized in that: The step of inputting the time series data into a pre-built LSTM neural network unit and outputting an initial prediction result includes: Based on the forward propagation and backward propagation processing of the forget gate, input gate, output gate and fully connected layer in the LSTM neural network unit, feature extraction is performed on the time series data to obtain the initial prediction result.

5. The method for generating a virtual power plant dispatching strategy based on digital simulation according to claim 2, characterized in that: The step of inputting the intermediate prediction result of the intermediate prediction model into the virtual simulation model comprises: The intermediate prediction results are input into the virtual simulation model constructed based on the following expression, and the corresponding simulation results are output: Among them, SOC(t) is the energy storage state in the virtual power plant system in period t, SOC(t+1) is the energy storage state in the virtual power plant system after update in period t+1, Pcharge(t) is the charging power of the energy storage system in period t, Pdischarge(t) is the discharging power of the energy storage system in period t, ηcharge is the charging efficiency, ηdischarge is the discharging efficiency, 1 / ηdischarge is used to characterize the compensation of energy loss during the discharge process, the period t is the first period, and the period t+1 is the second period.

6. The method for generating a virtual power plant dispatching strategy based on digital simulation according to any one of claims 1 to 5, characterized in that: The objective function is constructed based on minimizing the electricity cost of the target area and is expressed as: Among them, C grid (t) is the cost of electricity purchased from the power grid during period t, Egrid(t) is the amount of electricity purchased from the power grid during period t, and C loss (t) is the system loss cost of the virtual power plant in period t.

7. The method for generating a virtual power plant dispatching strategy based on digital simulation according to claim 6, characterized in that: The constraints include: The sum of the total power generation and charging capacity of the virtual power plant system in the target area is equal to the sum of the load power consumption, discharge and storage capacity, and the charging capacity includes photovoltaic charging and wind charging; The charging amount is within a set charging interval; The discharge amount is within a set discharge interval; The stored power is located in a set storage interval.

8. The method for generating a virtual power plant dispatching strategy based on digital simulation according to claim 7, characterized in that: When solving the objective function based on the set constraint conditions, outputting the solution result corresponding to the objective function includes: The objective function is solved based on a genetic algorithm, so that when the objective function satisfies the set constraint conditions, the solution result is output, and the solution result includes the start and stop time of the power generation equipment and the charging equipment in the target area.

9. A virtual power plant dispatching strategy generation device based on digital simulation, characterized in that: The device comprises: An acquisition module, used to acquire power data of a first period in a target area, wherein the power data includes charge and discharge capacity and energy storage capacity of each power device in the target area; A preprocessing module, used for preprocessing the power data; A prediction module, used for inputting the preprocessed power data into a pre-built target prediction model, and outputting predicted power data for a second period, wherein the target prediction model is built based on an LSTM neural network and a virtual simulation model of the power equipment in the target area, wherein the virtual simulation model is used to characterize the operating status of the power equipment in the target area, and the first period is earlier than the second period; A solution module is used to input the output predicted power data into a pre-constructed objective function, so that when the objective function is solved based on the set constraints, a solution result corresponding to the objective function is output, and the solution result represents the scheduling strategy of the power equipment in the target area in the second time period.

10. A computer device, characterized in that: The computer device includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the method for generating a virtual power plant dispatching strategy based on digital simulation as described in any one of claims 1 to 8.

Images