Multi-scene load prediction and evaluation method, device and system for intelligent power distribution network

Abstract

The invention discloses an intelligent power distribution network multi-scene load prediction and evaluation method, device and system, and the method comprises the steps: inputting the obtained multi-scene power grid load data into a feature coarse extraction model, and generating coarse features; inputting the coarse feature into a CRBM network, and generating a deep load feature by using the CRBM network with the goal of minimizing a pre-constructed energy feature function; wherein input parameters of the energy characteristic function comprise a visible layer state, a hidden layer state and a condition layer state of the CRBM network; and inputting the deep load characteristics into a pre-constructed load prediction model to obtain a load prediction value, and completing load prediction evaluation. The method is accurate in prediction result and has good prediction stability.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of power grid load prediction, and particularly relates to a multi-scenario load prediction evaluation method, device and system for intelligent distribution networks. BACKGROUND

[0002] With the rapid development of renewable energy technology, the penetration rate of new energy in distribution networks is continuously increasing. The output characteristics of new energy access modes represented by distributed photovoltaic and wind farms are greatly affected by weather, climate and environmental changes, and show significant volatility and uncertainty. At the same time, a large number of power electronic equipment-based electric terminals are connected, resulting in more complex load structure and more nonlinear dynamic response behavior in intelligent distribution networks. This diversified and non-stable load evolution trend puts higher requirements on the dispatching accuracy and system stability of power grid operation. Therefore, building an efficient and reliable load prediction mechanism for intelligent distribution networks is a key link to ensure power supply continuity and safe and stable operation of the system. As a basic link in power system operation, the accuracy of load prediction will directly affect key decisions such as power balance regulation, economic dispatching plan development and energy storage resource allocation. In the context of high penetration rate of new energy in distribution networks, traditional load prediction techniques gradually reveal their limitations.

[0003] Currently, most load prediction methods are based on machine learning and deep neural network techniques, such as BP neural network, long short-term memory network (LSTM), and Transformer model. These methods can learn the time series characteristics of load data to some extent and achieve short-term prediction goals. However, these models usually rely heavily on data volume and sample quality, and when dealing with massive, multi-scale and strongly nonlinear time series data, they often have problems such as long training time, high computational resource consumption and poor model stability. In practical applications, especially when facing sudden load disturbances, seasonal changes or special events such as holidays, the prediction accuracy and generalization ability of these methods often cannot meet the actual operational requirements. SUMMARY

[0004] To solve the above problems, the application provides a multi-scenario load prediction evaluation method, device and system for intelligent distribution networks, which has accurate prediction results and good prediction stability.

[0005] To achieve the above technical purposes and effects, the application realizes the following technical solutions:

[0006] In a first aspect, the application provides a multi-scenario load prediction evaluation method for intelligent distribution networks, comprising:

[0007] inputting the obtained multi-scenario power grid load data into a feature rough extraction model to generate rough features;

[0008] The coarse features are input into the CRBM network to minimize the pre-constructed energy feature function, and the CRBM network is used to generate deep load features; wherein, the input parameters of the energy feature function include the visible layer state, hidden layer state and conditional layer state of the CRBM network;

[0009] The deep load characteristics are input into a pre-built load prediction model to obtain load prediction values, thus completing the load prediction assessment.

[0010] In conjunction with the first aspect, optionally, the feature coarse extraction model includes convolutional layer units, pooling layer units, fully connected layers, and residual blocks;

[0011] The data processing formula for the convolutional layer unit is:

[0012] ,

[0013] In the formula, Indicates the first The first convolutional layer Each feature map; Indicates the first The convolutional layer and the first Convolutional kernels between convolutional layers; Indicates the first The first convolutional layer Each feature map; Indicates the first Bias terms for each convolutional layer; Indicates the first The nonlinear activation function of each convolutional layer For the first The total number of feature maps in each convolutional layer;

[0014] The data processing formula for the pooling layer unit is:

[0015] ,

[0016] In the formula, Indicates the first The first pooling layer Each feature map and They represent the first The weights and biases of each pooling layer; Indicates pooling operation; Represents a non-linear activation function;

[0017] The data processing formula for the fully connected layer is:

[0018] ,

[0019] wherein, is the output data of the full connection layer, is the weight matrix of the full connection layer, is the bias term of the full connection layer.

[0020] With reference to the first aspect, optionally, the residual block comprises a first convolutional layer, a second convolutional layer, a first batch normalization layer, a second batch normalization layer and a ReLU activation function, and a data processing formula of the residual block is:

[0021] ,

[0022] wherein, is the output data of the residual block, representing a coarse feature; , are weight matrices of the first convolutional layer and the second convolutional layer respectively, , represent normalization operations of the first batch normalization layer and the second batch normalization layer respectively, , represent convolution operations of the first convolutional layer and the second convolutional layer respectively.

[0023] With reference to the first aspect, optionally, a mathematical expression of the energy characteristic function is:

[0024] ,

[0025] wherein,

[0026] ,

[0027] ,

[0028] ,

[0029] wherein, represents the energy characteristic function, represents a visible layer state, represents a hidden layer state, represents a conditional layer state, is a gating factor of the visible layer, represents a visible layer bias, is a gating factor of the hidden layer, represents a hidden layer bias, , , represent the first to the th neurons of the visible layer, the hidden layer and the conditional layer respectively, , , ,​ , respectively represent the neuron set of the visible layer, the hidden layer and the conditional layer; is the connection weight of the visible layer and the hidden layer, is the connection weight of the visible layer and the conditional layer, is the connection weight of the conditional layer and the hidden layer; is the enhancement item energy for supplementing high-order nonlinear interaction; represents the interaction factor weight, represents the hyperbolic tangent function, is the gating factor is the weight parameter for the visible layer state, is the gating factor is the weight parameter for the conditional layer state, is the gating factor is the weight parameter for the hidden layer state, is the gating factor is the weight parameter for the conditional layer state, is the Sigmoid function for the gating factor calculation.

[0030] In combination with the first aspect, optionally, the deep load feature generation method comprises:

[0031] calculating the probability The calculation formula of the probability is as follows:

[0032] ,

[0033] wherein, is the output data of the CRBM network, represents all possible visible layer state combinations, represents all possible hidden layer state combinations, and the denominator represents the summation of all possible states , .

[0034] The probability under the minimum case is taken as the deep load feature.

[0035] In combination with the first aspect, optionally, the load prediction model comprises a BiLSTM model and a pre-trained Adaboost model;

[0036] The BiLSTM model performs forward propagation and backward propagation on the received deep load feature respectively, and weights the results of the forward propagation and the backward propagation to obtain an initial load prediction value;

[0037] The initial load forecast is input into the pre-trained Adaboost model to obtain the final load forecast.

[0038] In conjunction with the first aspect, optionally, the Adaboost model includes several weak regressors arranged in parallel, and the training method of the Adaboost model includes:

[0039] Input the output data of the BiLSTM model into the corresponding weak regressor;

[0040] Calculate the absolute value of the prediction error absolute value The calculation formula is: , This represents the actual load value. This represents the initial load forecast value output by the BiLSTM model. The deep load characteristics are the input.

[0041] According to the absolute value Update the weights of the corresponding weak regressor using the following formula: , Indicates the first The weights of the weak regressor.

[0042] In conjunction with the first aspect, optionally, the output signal of the Adaboost model is:

[0043] ,

[0044] In the formula, This is the output signal of the Adaboost model. For symbolic functions, This represents the total number of weak regressors.

[0045] Secondly, the present invention provides a smart distribution network multi-scenario load forecasting and assessment device, comprising:

[0046] The coarse feature generation module is used to input the acquired multi-scenario power grid load data into the coarse feature extraction model to generate coarse features;

[0047] A deep load feature generation module is used to input the coarse features into a CRBM network to generate deep load features by minimizing the pre-constructed energy feature function; wherein, the input parameters of the energy feature function include the visible layer state, hidden layer state, and conditional layer state of the CRBM network.

[0048] The load forecasting module is used to input the deep load characteristics into a pre-built load forecasting model to obtain load forecast values ​​and complete the load forecasting assessment.

[0049] In a third aspect, the present application provides a smart power distribution network multi-scenario load prediction and evaluation system, comprising a storage medium and a processor;

[0050] The storage medium is used to store instructions;

[0051] The processor is used to operate according to the instructions to execute the method according to any one of the first aspect.

[0052] Compared with the prior art, the present application has the following beneficial effects:

[0053] The present application proposes to input the obtained multi-scenario power grid load data into a feature rough extraction model to generate rough features; input the rough features into a CRBM network, so that the pre-constructed energy feature function is minimized as the target, and the deep load features are generated by using the CRBM network; wherein the input parameters of the energy feature function include the visible layer state, the hidden layer state and the conditional layer state of the CRBM network; input the deep load features into the pre-constructed load prediction model to obtain the load prediction value, complete the load prediction and evaluation, the required computing power platform is low, the prediction result is accurate, has good stability, and has application potential in the field of smart power distribution network multi-scenario load prediction and evaluation. BRIEF DESCRIPTION OF DRAWINGS

[0054] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed in the embodiments will be briefly introduced as follows. Obviously, the drawings in the following description are only some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained without creative labor on the basis of these drawings, wherein:

[0055] Figure 1 The overall framework diagram of the smart power distribution network multi-scenario load prediction and evaluation method of an embodiment of the present application;

[0056] Figure 2 The network structure diagram of the feature rough extraction model of an embodiment of the present application;

[0057] Figure 3 The structure diagram of the CRBM network of an embodiment of the present application;

[0058] Figure 4 The structure diagram of the load prediction model of an embodiment of the present application;

[0059] Figure 5 The CNN iterative convergence curve diagram of an embodiment of the present application;

[0060] Figure 6The annual load prediction graph under different algorithms of an embodiment of the present application;

[0061] Figure 7 The working day and weekend working condition load prediction graph of an embodiment of the present application;

[0062] Figure 8 The working day and weekend working condition error curve graph of an embodiment of the present application;

[0063] Figure 9 The dry season and rainy season working condition load prediction graph of an embodiment of the present application;

[0064] Figure 10 The dry season and rainy season working condition error curve graph of an embodiment of the present application;

[0065] Figure 11 The different algorithm index comparison graph of an embodiment of the present application. DETAILED DESCRIPTION

[0066] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by a person of ordinary skill in the art without creative work fall within the protection scope of the present application.

[0067] In addition, if the present application embodiments involve the description of "first", "second", etc., the description of "first", "second", etc. is only for the purpose of description, and cannot be understood as indicating or implying the relative importance of the indicated technical features or implicitly indicating the number of the indicated technical features. Therefore, the features limited by "first", "second" can explicitly or implicitly include at least one of the features. In addition, the technical solutions of each embodiment can be combined with each other, but it must be based on the fact that a person of ordinary skill in the art can realize it. When the combination of technical solutions contradicts each other or cannot be realized, it should be considered that the combination of technical solutions does not exist and is not within the protection scope of the present application.

[0068] Embodiment 1

[0069] In the embodiments of the present application, an intelligent power distribution network multi-scene load prediction evaluation method is provided, as shown in Figure 1 includes the following steps:

[0070] (1) input the obtained multi-scene power grid load data into a feature rough extraction model to generate rough features;

[0071] (2) The coarse features are input into the CRBM network to minimize the pre-constructed energy feature function, and the deep load features are generated using the CRBM network; wherein, the input parameters of the energy feature function include the visible layer state, hidden layer state, and conditional layer state of the CRBM network; see the structure diagram of the CRBM network. Figure 3 ;

[0072] (3) Input the deep load characteristics into the pre-built load prediction model to obtain the load prediction value and complete the load prediction assessment.

[0073] In one specific embodiment of the present invention, such as Figure 2 As shown, the feature coarse extraction model includes convolutional layer units, pooling layer units, fully connected layers, and residual blocks, which are used for coarse extraction of load features;

[0074] The data processing formula for the convolutional layer unit is:

[0075] ,

[0076] In the formula, Indicates the first The j-th feature map of a convolutional layer; Indicates the first The convolutional layer and the first Convolutional kernels between convolutional layers; Indicates the first The j-th feature map of a convolutional layer; Indicates the first Bias terms for each convolutional layer; Indicates the first The non-linear activation function of each convolutional layer is used to transform the linear output after convolution into a non-linear form of the result. Indicates the first The total number of feature maps in each convolutional layer;

[0077] The data processing formula for the pooling layer unit is:

[0078] ,

[0079] In the formula, Indicates the first The first pooling layer Each feature map and They represent the first The weights and biases of each pooling layer; This indicates a pooling operation, which can be either max pooling or average pooling in practice. Represents a nonlinear activation function;

[0080] The data processing formula of the full connection layer is:

[0081] ,

[0082] In the formula, is the output data of the full connection layer, is the weight matrix of the full connection layer, is the bias term of the full connection layer. The full connection layer is used for compressing and nonlinearly mapping information, and enhancing feature discrimination.

[0083] In one specific embodiment of the embodiment of the present application, the residual block includes a first convolutional layer, a second convolutional layer, a first batch normalization layer, a second batch normalization layer and a ReLU activation function, and the data processing formula of the residual block is:

[0084] ,

[0085] In the formula, represents the output data of the residual block, and represents the coarse feature, , are the weight matrices of the first convolutional layer and the second convolutional layer respectively, , represent the normalization operations of the first batch normalization layer and the second batch normalization layer respectively, , represent the convolution operations of the first convolutional layer and the second convolutional layer respectively.

[0086] In one specific embodiment of the embodiment of the present application, the mathematical expression of the energy feature function is:

[0087] ,

[0088] In the formula,

[0089] ,

[0090] ,

[0091] ,

[0092] In the formula, represents the energy feature function, represents the visible layer bias, represents the visible layer state, represents the hidden layer bias, represents the hidden layer state, represents the conditional layer state, , , respectively represent the visible layer, the hidden layer, and the conditional layer 、 、 respectively represent the visible layer, the hidden layer, and the conditional layer 、 、 respectively represent the visible layer, the hidden layer, and the conditional layer is a gating factor for the visible layer is a gating factor for the hidden layer, which is used to realize adaptive regulation of the activation strength of each neuron, and embodies the importance weight of the feature to the prediction target is a connection weight for the visible layer and the hidden layer is a connection weight for the visible layer and the conditional layer is a connection weight for the conditional layer and the hidden layer and realize the cross-time step association of 、 、 represents an interaction factor weight represents a hyperbolic tangent function is a gating factor for the weight parameter of the visible layer state is a gating factor for the weight parameter of the conditional layer state is a gating factor for the weight parameter of the hidden layer state is a gating factor for the weight parameter of the conditional layer state is a Sigmoid function used for gating factor calculation is an enhancement item energy used to supplement high-order nonlinear interaction represents a conditional regulation cross term, which enhances the guidance of the historical state to the current prediction state by introducing conditional modulation. The hidden layer state fuses the visible layer state and the conditional layer state through the weight to form a time-dependent item; the visible layer bias and the hidden layer bias in the energy function respectively adjust the activation threshold of each layer to ensure the stability of feature transmission.

[0093] In one specific embodiment of the embodiment of the present application, the generation method of the deep load feature comprises:

[0094] calculating a probability The calculation formula of the probability is as follows:

[0095] ,​

[0096] wherein, is the output data of the CRBM network, represents all possible visible layer state combinations, represents all possible hidden layer state combinations, and the denominator represents the sum of the probabilities of all possible states as a normalization term; , all possible states are summed up as a normalization term;

[0097] the probability in the minimum case is a deep load feature.

[0098] In the above scheme, the smaller the probability is, the greater the energy is, which realizes the limitation of the feature representation result in a more stable condition, obtains a deep load feature, and approximates the real load distribution, while improving the calculation efficiency in capturing deep information of data features.

[0099] In one specific embodiment of the embodiment of the present application, as shown in the figure, Figure 4 the load prediction model comprises a BiLSTM model and a pre-trained Adaboost model;

[0100] The BiLSTM model respectively performs forward propagation and backward propagation on the received deep load feature, and weights the results of forward propagation and backward propagation to obtain an initial load prediction value;

[0101] The initial load prediction value is input into the pre-trained Adaboost model to obtain a final load prediction value.

[0102] In the above scheme, the BiLSTM-Adaboost is used to construct the load prediction model, to capture the forward and backward load time sequence features, to reduce the overfitting risk in model training by using dynamic adjustment of sample weights, and to enhance the robustness of the model.

[0103] In one specific embodiment of the embodiment of the present application, the LSTM is composed of an input gate, a forgetting gate and an output gate, which jointly control and adjust the state of the memory cell.

[0104] The state of the hidden layer of the BiLSTM model at each time is composed of three parts. One part is forward propagation, denoted by , one part is backward propagation, denoted by , and the input information at the current time . The specific formula is as follows:

[0105] ,

[0106] ,

[0107] The final output of the BiLSTM model is obtained by weighted summation of forward and backward propagation:

[0108]

[0109] in, and These represent the states of the forward hidden layer and the backward hidden layer, respectively. and These are the weights used to weight the outputs of the forward and reverse hidden layers. It is the bias term at the current moment. This is the output of the BiLSTM model.

[0110] In one specific embodiment of the present invention, the Adaboost model includes several weak regressors arranged in parallel, and the training method of the Adaboost model includes:

[0111] The Adaboost model includes several weak regressors arranged in parallel, and the training method of the Adaboost model includes:

[0112] Input the output data of the BiLSTM model into the corresponding weak regressor;

[0113] Calculate the absolute value of the prediction error absolute value The calculation formula is: , This represents the actual load value. This represents the initial load forecast value output by the BiLSTM model. The deep load characteristics are the input.

[0114] According to the absolute value Update the weights of the corresponding weak regressor using the following formula: , Indicates the first The weights of the weak regressor.

[0115] In one specific embodiment of the present invention, the output signal of the Adaboost model is:

[0116] ,

[0117] In the formula, This is the output signal of the Adaboost model. For symbolic functions, This represents the total number of weak regressors.

[0118] In the above scheme, the Adaboost algorithm first takes the output of the BiLSTM as the input of the weak regressor; then trains the weak regressor, calculates the error rate, and updates the weight of the classifier according to the error rate; and finally linearly combines the weak classifiers into a strong classifier, that is, the Adaboost model, which can weight and integrate multiple prediction results to realize multi-scenario load prediction evaluation.

[0119] In the implementation process, the intelligent power distribution network multi-scenario load prediction evaluation method in the embodiment of the application is packaged into an RCHYQ model. For the load prediction effect under different working conditions, the RCHYQ model in the embodiment of the application is used to predict and evaluate the annual load data of the Panama power grid operator, taking four conditions as examples. The four conditions are: weekdays, weekends, dry season, and rainy season. The intelligent power distribution network data is classified, the prediction curve is compared with the real curve, and a classic model is introduced for horizontal comparison, which shows that the intelligent power distribution network multi-scenario load prediction evaluation method in the embodiment of the application not only has stability, but also can minimize the error and achieve the best prediction effect.

[0120] In one example, the convergence process curve of the feature rough extraction model with respect to the convolutional neural network (CNN) part is as shown in Figure 5 . The horizontal axis represents the number of iterations, and the vertical axis is the root mean square error. In the initial stage of training, the error is high, but it decreases rapidly and is maintained below 0.02, the overall change tends to be stable, and the convergence effect is good.

[0121] In this embodiment, part of the parameter settings are shown in Table 1.

[0122] Table 1

[0123]

[0124] After completing the model training, the test set is used to verify the load prediction under four conditions, and the prediction results are as shown in Figures 7-10 .

[0125] To verify the performance of the method, four indicators widely used in power load prediction are used to evaluate the accuracy of the model, that is, the mean square error , the root mean square error , the mean absolute error , and the determination coefficient R 2 , the formulas are as follows:

[0126]

[0127]

[0128]

[0129]

[0130] In the formula, The total number of samples; For the first The true value of each sample; For the first Predicted values ​​for each sample; This is the sample mean; see details below. Figure 11 .

[0131] The coarse feature extraction model provides richer feature information, reducing the computational burden on subsequent LSTM, effectively reducing computational redundancy, shortening runtime, and improving the overall computational efficiency of load forecasting. Table 2 shows the comparison results of different algorithms before and after incorporating the CRBM calculation method.

[0132] Table 2

[0133]

[0134] The results show that the multi-scenario load forecasting and evaluation method for smart distribution networks proposed in this invention is superior to other models in terms of evaluation indicators and operating speed, and has good performance and high accuracy in smart distribution network load forecasting.

[0135] like Figure 6 The diagram shows the annual load forecast under different algorithms. It is clear that the algorithm in this embodiment of the invention is significantly better than the forecasting effect of other algorithms.

[0136] Example 2

[0137] Based on the same inventive concept as Embodiment 1, this embodiment of the invention provides a smart distribution network multi-scenario load forecasting and assessment device, comprising:

[0138] The coarse feature generation module is used to input the acquired multi-scenario power grid load data into the coarse feature extraction model to generate coarse features;

[0139] A deep load feature generation module is used to input the coarse features into a CRBM network to generate deep load features by minimizing the pre-constructed energy feature function; wherein, the input parameters of the energy feature function include the visible layer state, hidden layer state, and conditional layer state of the CRBM network.

[0140] The load forecasting module is used to input the deep load characteristics into a pre-built load forecasting model to obtain load forecast values ​​and complete the load forecasting assessment.

[0141] The rest are the same as in Example 1.

[0142] Example 3

[0143] The embodiment of the present application provides a smart power distribution network multi-scene load prediction and evaluation system, comprising a storage medium and a processor.

[0144] The storage medium is used for storing instructions.

[0145] The processor is used for operating according to the instructions to execute the method according to any one of the embodiments 1.

[0146] Those skilled in the art should understand that the embodiments of the present application can be provided as a method, a system, or a computer program product. Therefore, the present application can adopt a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can adopt a computer program product in the form of being implemented on one or more computer usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer usable program codes.

[0147] The present application is described with reference to flowcharts and / or block diagrams according to the method, equipment (system), and computer program product of the embodiments of the present application. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, and the combination of the flows and / or blocks in the flowcharts and / or block diagrams can be implemented by computer program instructions. These computer program instructions can be provided to a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to produce a machine, so that the instructions executed by the computer or other programmable data processing devices produce a device implemented in the flowcharts and / or block diagrams. Figure 1 The device for implementing the functions specified in one or more flows and / or blocks. Figure 1 The device for implementing the functions specified in one or more flows and / or blocks.

[0148] These computer program instructions can also be stored in a computer readable storage medium capable of guiding the computer or other programmable data processing devices to work in a specific way, so that the instructions stored in the computer readable storage medium produce a product including instruction devices, which implement the flowcharts and / or block diagrams. Figure 1 The device for implementing the functions specified in one or more flows and / or blocks. Figure 1 The device for implementing the functions specified in one or more flows and / or blocks.

[0149] These computer program instructions can also be loaded into the computer or other programmable data processing devices, so that a series of operation steps are performed on the computer or other programmable devices to produce a computer implemented process, so that the instructions executed on the computer or other programmable devices provide a process for implementing the flowcharts and / or block diagrams. Figure 1 The device for implementing the functions specified in one or more flows and / or blocks. Figure 1 The device for implementing the functions specified in one or more flows and / or blocks.

[0150] The embodiments of the present application are described above with reference to the drawings, but the present application is not limited to the above-described specific embodiments, and the above-described specific embodiments are merely illustrative, not restrictive, and a person of ordinary skill in the art can make many forms under the inspiration of the present application without departing from the purpose of the present application and the scope protected by the claims, which all belong to the protection of the present application.

[0151] The basic principles and main features of the present application and the advantages of the present application are shown and described above. It should be understood by those skilled in the art that the present application is not limited to the above-described embodiments, and the above-described embodiments and descriptions in the specification are only illustrative of the principles of the present application, and various changes and improvements can be made to the present application without departing from the spirit and scope of the present application, which all fall within the scope of the claimed present application. The scope of protection of the present application is defined by the appended claims and their equivalents.

Claims

1. A smart power distribution network multi-scenario load prediction evaluation method, characterized in that, The method comprises the following steps: The obtained multi-scenario power grid load data is input into a feature rough extraction model to generate rough features; The rough features are input into a CRBM network to minimize a pre-constructed energy feature function, and deep load features are generated by using the CRBM network; wherein the input parameters of the energy feature function include the visible layer state, the hidden layer state and the conditional layer state of the CRBM network; The deep load features are input into a pre-constructed load prediction model to obtain load prediction values, and load prediction evaluation is completed. 2.The smart power grid multi-scenario load forecasting evaluation method according to claim 1, characterized in that: The feature rough extraction model comprises a convolution layer unit, a pooling layer unit, a full connection layer and a residual block; The data processing formula of the convolution layer unit is: ， In the formula, Indicates the first The first convolutional layer Each feature map; Indicates the first The convolutional layer and the first Convolutional kernels between convolutional layers; Indicates the first The first convolutional layer Each feature map; Indicates the first Bias terms for each convolutional layer; Indicates the first The nonlinear activation function of each convolutional layer For the first The total number of feature maps in each convolutional layer; The data processing formula of the pooling layer unit is: ， wherein, represents the i-th feature map of the i-th pooling layer, represents the i-th feature map of the i-th pooling layer, represents the i-th feature map of the i-th pooling layer, and represents the i-th feature map of the i-th pooling layer, represents the i-th feature map of the i-th pooling layer; represents the pooling operation; represents the non-linear activation function; The data processing formula of the full connection layer is: ， wherein is the output data of the fully connected layer, is the weight matrix of the fully connected layer, is the bias term of the fully connected layer. 3.The method of claim 2, wherein: The residual block comprises a first convolution layer, a second convolution layer, a first batch normalization layer, a second batch normalization layer and a ReLU activation function, and the data processing formula of the residual block is: ， wherein, is the output data of the residual block, representing the coarse feature; , are weight matrices of the first and second convolution layers, respectively, , represent normalization operations of the first and second batch normalization layers, respectively, , represent convolution operations of the first and second convolution layers, respectively.

4. The method of claim 1, wherein the method further comprises: The mathematical expression of the energy feature function is: ， wherein, ， ， ， wherein represents the energy characteristic function, represents the visible layer state, represents the hidden layer state, represents the conditional layer state, is the gating factor for the visible layer, represents the visible layer bias, is the gating factor for the hidden layer, represents the hidden layer bias, , , represents the first , , neurons of the visible layer, hidden layer, and conditional layer, respectively, , , represents the set of neurons of the visible layer, hidden layer, and conditional layer, respectively; is the connection weight between the visible layer and the hidden layer, is the connection weight between the visible layer and the conditional layer, is the connection weight between the conditional layer and the hidden layer; is the enhancement energy for supplementing high-order nonlinear interactions; represents the interaction factor weight, represents the hyperbolic tangent function, is the gating factor is the weight parameter for the visible layer state, is the gating factor is the weight parameter for the conditional layer state, is the gating factor is the weight parameter for the hidden layer state, is the gating factor is the weight parameter for the conditional layer state, is the Sigmoid function for the gating factor calculation.

5. The method of claim 4, wherein the method further comprises: The generation method of the deep load features comprises: Computing the probability The formula for computing the probability is: ， wherein, is the output data of the CRBM network, denotes all possible visible layer state combinations, denotes all possible hidden layer state combinations, the denominator denotes the sum over , all possible states; probabilities minimally as deep load features. 6.The smart grid multi-scenario load forecasting evaluation method of claim 1, wherein: The load prediction model comprises a BiLSTM model and a pre-trained Adaboost model; The BiLSTM model performs forward propagation and backward propagation on the received deep load features respectively, and weights the results of the forward propagation and the backward propagation to obtain initial load prediction values; The initial load prediction values are input into the pre-trained Adaboost model to obtain final load prediction values.

7. The method of claim 6, wherein the method further comprises: The Adaboost model comprises a plurality of weak regressors arranged in parallel, and the training method of the Adaboost model comprises: The output data of the BiLSTM model is input into the corresponding weak regressor; The absolute value of the prediction error is calculated , the absolute value The calculation formula is: , represents the true load value, represents the initial load prediction value output by the BiLSTM model, is the input deep load feature; According to the absolute value Update the weights of the corresponding weak regressor using the following formula: , Indicates the first The weights of the weak regressor. 8.The method of claim 7, wherein the method further comprises: The output signal of the Adaboost model is: ， wherein is an output signal of the Adaboost model, is a sign function, is the total number of weak regressors.

9. A smart power distribution network multi-scenario load forecasting evaluation device, characterized in that, The method comprises the following steps: The rough feature generation module is configured to input the obtained multi-scenario power grid load data into a feature rough extraction model to generate rough features; The deep load feature generation module is configured to input the rough features into a CRBM network to minimize a pre-constructed energy feature function, and generate deep load features by using the CRBM network; wherein the input parameters of the energy feature function include the visible layer state, the hidden layer state and the conditional layer state of the CRBM network; The load prediction module is configured to input the deep load features into a pre-constructed load prediction model to obtain load prediction values, and complete load prediction evaluation.

10. A smart power distribution network multi-scenario load forecasting evaluation system, characterized in that, The storage medium and the processor are included; The storage medium is configured to store instructions; The processor is configured to operate according to the instructions to execute the method according to any one of claims 1-8.

Images