Demand side resource demand response transaction scale calculation method based on LassoNet and S-MLP

Abstract

The invention discloses a demand side resource demand response transaction scale calculation method based on LassoNet and S-MLP, and relates to the demand response field, and the method comprises the steps: classifying demand side resources according to power consumption types; the meteorological factors and the classified demand side resources are input into a LassoNet network model, and feature variables are screened through regularization; and splicing the classified demand side resources and the feature variables, carrying out position coding by adopting a position embedding method to form input data input 1, input 2 and input 3, inputting the input data input 1, input 2 and input 3 into a multi-layer perceptron model MLP, capturing a nonlinear relationship, and completing prediction of the demand response transaction scale of the demand side resources. According to the method, through the feature selection capability of the LassoNet model, the model calculation resources are optimized, and the prediction precision is improved; the position embedding method ensures the structured processing of the data, and enhances the understanding of the model on the complex relationship. Finally, through nonlinear feature extraction of the MLP model, high-precision prediction of the demand side resource response transaction scale is realized, and the stability of the system is improved.

Description

Technical Field

[0001] The present invention relates to the field of demand response, and in particular to a method for calculating the transaction scale of demand-side resource demand response based on LassoNet and S-MLP. Background Art

[0002] Demand side management (DSM) is a technical means to improve the efficiency and stability of the power system by guiding power consumers to actively adjust their electricity consumption behavior. In the demand response (DR) plan, power consumers can respond to the demand signals of the power system by reducing or shifting loads, thereby balancing the supply and demand relationship and improving the reliability and flexibility of the power grid.

[0003] However, existing demand response transaction scale estimation methods often rely on traditional statistical analysis and regression models, which have certain limitations when dealing with complex and changeable demand-side response behaviors. For example, traditional methods are difficult to effectively capture the nonlinear characteristics of demand-side resources and the complex relationship between high-dimensional data, resulting in low prediction accuracy and difficulty in meeting actual application needs.

[0004] In order to solve the above problems, machine learning technology has been introduced into the demand response field in recent years, aiming to improve the accuracy of demand response transaction scale calculation by building more complex and accurate prediction models. Among them, LassoNet is a deep learning method based on sparse feature selection, which can automatically screen out the key features that have the greatest impact on the prediction results; S-MLP (sequential multilayer perceptron) is a neural network model that can process time series data, which is suitable for processing spatiotemporal features in demand response transaction scale prediction.

[0005] Combining the LassoNet and S-MLP methods can fully utilize the advantages of both, and achieve accurate measurement of the transaction scale of demand-side resource demand response through sparse feature selection and time series feature modeling. This method can not only improve the accuracy and robustness of predictions, but also provide strong decision-making support for power market and grid operations. Summary of the invention

[0006] In view of the problems existing in the existing method for calculating the scale of next-day demand-side resource transactions, the present invention is proposed.

[0007] Therefore, the problem to be solved by the present invention is how to improve the prediction accuracy and calculation efficiency of the demand response transaction scale based on the demand-side resource demand response transaction scale calculation method of LassoNet and S-MLP.

[0008] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0009] In the first aspect, an embodiment of the present invention provides a method for calculating the scale of demand-side resource demand response transactions based on LassoNet and S-MLP, which includes classifying demand-side resources according to electricity consumption types; inputting meteorological factors and classified demand-side resources into the LassoNet network model, and filtering feature variables through regularization; splicing the classified demand-side resources and feature variables, and using a position embedding method to perform position encoding to form input data input1, input2 and input3; inputting the input data input1, input2 and input3 into a multi-layer perceptron model MLP to capture nonlinear relationships and complete the prediction of the scale of demand-side resource demand response transactions.

[0010] As a preferred solution of the method for calculating the transaction scale of demand-side resource demand response based on LassoNet and S-MLP of the present invention, the multilayer perceptron model MLP includes the following steps: In the MLP, the neurons perform weighted summation on the input signal and convert it into the output of the jth neuron in the lth layer through the activation function. The specific formula is as follows:

[0011]

[0012] in, The activation function is converted into the output of the jth neuron in the lth layer, ai (l-1) is the output of the i-th neuron in the l-1th layer, is the weight connecting the i-th neuron in the l-1th layer and the j-th neuron in the lth layer, is the bias term of the jth neuron in the lth layer;

[0013] Apply the activation function σ to the weighted sum To introduce nonlinearity, the specific formula is as follows:

[0014]

[0015] As a preferred solution of the method for calculating the transaction scale of demand-side resource demand response based on LassoNet and S-MLP of the present invention, it also includes, during the training process, minimizing the difference between the predicted output and the true label as the objective function L, and the specific formula is as follows:

[0016]

[0017] Where N is the number of training samples, is the kth target value of the nth sample, is the model's predicted value for the nth input sample at the kth output neuron, For the model, for the nth input sample xn The predicted value at the kth output neuron.

[0018] The regularization term L2 is introduced to optimize the network weights. The specific formula is as follows:

[0019]

[0020] Among them, Ω is the regularization term, λ is the regularization coefficient, L is the number of network layers, M is the number of neurons in the lth layer, is the weight connecting the i-th neuron in the l-1th layer and the j-th neuron in the lth layer.

[0021] As a preferred solution of the method for calculating the transaction scale of demand-side resource demand response based on LassoNet and S-MLP described in the present invention, it also includes: inputting input1 into the first multilayer perceptron, performing MLP1 processing, obtaining the first output output_1, and connecting output_1 with input2 to generate a first connection signal link_1; inputting link_1 into the second multilayer perceptron, performing MLP2 processing, obtaining the second output output_2, and connecting output_2 with input3 to generate a second connection signal link_2; inputting link_2 into the third multilayer perceptron, performing MLP3 processing, and generating a final output signal link; inputting the final output signal link into the linear layer for compression, and inputting the compressed signal into the fully connected layer to generate the final prediction result of the transaction scale of demand-side resource demand response.

[0022] As a preferred solution of the method for calculating the transaction scale of demand-side resource demand response based on LassoNet and S-MLP of the present invention, the objective function of the LassoNet network model includes a loss function and a regularization term; the LassoNet network model includes the following steps: predicting the difference between the predicted label and the real label of the LassoNet network model through the loss function, the specific formula is as follows:

[0023]

[0024] Among them, min θ,W L(θ,W) is to find the minimum value of the parameter θ and the first hidden layer weight W in the loss function L(θ,W), θ1 is the L1 norm of the network layer number of θ, ||Wj (1) || ∞ is the infinite norm of the jth weight vector of the first hidden layer weight W, λ is the regularization coefficient, |θ j | is the absolute value of the jth parameter of θ, and M is the maximum allowed value of the weight.

[0025] The weights of the Skip layer are penalized through the regularization term L1 norm to select feature variables; by controlling the sparsity of the Skip layer weights, full control over the network feature selection is achieved.

[0026] As a preferred solution of the demand-side resource demand response transaction scale estimation method based on LassoNet and S-MLP described in the present invention, wherein: the architecture of the LassoNet includes a Skip layer and a feedforward neural network of arbitrary depth; the Skip layer and the first hidden layer together capture any nonlinearity in the feature variables through a hierarchical soft threshold optimizer; and also includes that when M=0, all hidden units are inactive, and the LassoNet network model degenerates into a Lasso model that only uses Skip connections; when M tends to infinity, the LassoNet network model restores the standard unregularized feedforward neural network.

[0027] As a preferred solution of the method for calculating the transaction scale of demand-side resource demand response based on LassoNet and S-MLP in the present invention, the classification includes curtailable load, transferable load and energy storage resources; the curtailable load includes the power load temporarily reduced or suspended by the user in a demand response event; the transferable load includes the power load adjusted between different time periods; the energy storage resources include battery energy storage systems that provide electricity or absorb excess electricity when the power system needs it.

[0028] In the second aspect, an embodiment of the present invention provides a demand-side resource demand response transaction scale calculation system based on LassoNet and S-MLP, which includes: a classification module for classifying demand-side resources according to electricity consumption type; a screening module for inputting meteorological factors and classified demand-side resources into the LassoNet network model, and screening feature variables through regularization; an encoding module for splicing the classified demand-side resources and feature variables, and using a position embedding method for position encoding to form input data input1, input2 and input3; a capture module for inputting input data input1, input2 and input3 into a multilayer perceptron model MLP to capture nonlinear relationships and complete the prediction of the demand-side resource demand response transaction scale.

[0029] In a third aspect, an embodiment of the present invention provides a computer device, including a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program instructions are executed by the processor, the steps of the method for calculating the transaction scale of demand-side resource demand response based on LassoNet and S-MLP as described in the first aspect of the present invention are implemented.

[0030] In a fourth aspect, an embodiment of the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program instructions are executed by a processor, the steps of the method for calculating the transaction scale of demand-side resource demand response based on LassoNet and S-MLP as described in the first aspect of the present invention are implemented.

[0031] The beneficial effects of the present invention are as follows: the present invention classifies the demand-side resources according to the type of electricity consumption, and inputs the meteorological factors and the classified demand-side resources into the LassoNet network model to screen the characteristic variables, then splices the classified resources with the characteristic variables and encodes them using the position embedding method, and finally inputs these data into the multi-layer perceptron model MLP to capture nonlinear relationships, so as to achieve accurate prediction of the transaction scale of the demand-side resource demand response. This method improves the flexibility of system response by effectively managing and regulating different types of resources; optimizes the model computing resources and improves the prediction accuracy through the feature selection capability of the LassoNet model; the position embedding method ensures the structured processing of the data and enhances the model's understanding of complex relationships. Finally, through the nonlinear feature extraction of the MLP model, a high-precision prediction of the transaction scale of the demand-side resource response is achieved, and the economy and stability of the system are improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work. Among them:

[0033] Figure 1 This is a flow chart of the demand-side resource demand response transaction scale calculation method based on LassoNet and S-MLP in Example 1. DETAILED DESCRIPTION

[0034] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the accompanying drawings.

[0035] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0036] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive with other embodiments.

[0037] Example 1

[0038] Reference Figure 1 , which is the first embodiment of the present invention, and provides a method for calculating the transaction scale of demand-side resource demand response based on LassoNet and S-MLP, including:

[0039] S1: Classify demand-side resources according to electricity consumption type.

[0040] Specifically, the classification includes curtailable loads, transferable loads and energy storage resources; curtailable loads refer to loads that users can temporarily reduce or suspend during demand response events. For example, during peak power consumption, commercial buildings can temporarily shut down some lighting systems or air conditioning equipment.

[0041] Furthermore, shiftable load refers to the electricity load that can be flexibly adjusted between different time periods. For example, industrial users can shift certain non-urgent production tasks from peak hours to off-peak hours in response to electricity price incentives.

[0042] Furthermore, energy storage resources include battery storage systems and other forms of energy storage equipment that can provide power or absorb excess power when the power system needs it. For example, solar panels installed by home users combined with energy storage batteries can store excess power during the day and use it or sell it back to the grid at night.

[0043] S2: Input meteorological factors and classified demand-side resources into the LassoNet network model, and filter feature variables through regularization.

[0044] Specifically, in LassoNet, feature sparsity is achieved by adding a Skip layer (i.e., residual connection), which allows input features to directly skip certain hidden layers and directly affect the output. The objective function of the LassoNet network model includes a loss function and a regularization term; the loss function measures the difference between the model prediction and the true label; the regularization term penalizes the weights of the Skip layer through the L1 norm, encouraging the model to select fewer features.

[0045] The further LassoNet network model includes the following steps: predicting the difference between the predicted label and the true label of the LassoNet network model through the loss function, the specific formula is as follows:

[0046]

[0047] Among them, min θ,W L(θ,W) is to find the minimum value of the parameter θ and the first hidden layer weight W in the loss function L(θ,W), θ1 is the L1 norm of the network layer number of θ, ||Wj (1) || ∞ is the infinite norm of the jth weight vector of the first hidden layer weight W, λ is the regularization coefficient, |θ j | is the absolute value of the jth parameter of θ, and M is the maximum allowed value of the weight.

[0048] It should be noted that the objective function not only promotes the linear part of the signal over the nonlinear part, but also provides a natural framework for feature selection through a single coefficient (linear part).

[0049] Furthermore, the weights of the Skip layer are penalized by the regularization term L1 norm, the sparsity of the weights of the Skip layer is controlled, and the feature variables are selected to achieve complete control over the network feature selection; the architecture of LassoNet includes a Skip layer and a feedforward neural network of arbitrary depth; the Skip layer and the first hidden layer together capture any nonlinearity in the feature variables through a hierarchical soft threshold optimizer.

[0050] Specifically, when M = 0, all hidden units are inactive, and the LassoNet network model degenerates into a Lasso model that only uses Skip connections; when M tends to infinity, the LassoNet network model restores the standard unregularized feedforward neural network.

[0051] Furthermore, another important feature of the LassoNet method is its generalization ability in different data sets and tasks. Experimental results on multiple real data sets show that LassoNet significantly outperforms existing state-of-the-art methods in feature selection and regression tasks. In addition, LassoNet can be directly extended to deep networks and can be implemented by adding only a small amount of code to standard neural networks, which further demonstrates its flexibility and practicality. Meteorological factors such as electricity prices and response incentives are input into the LassoNet network together with three types of loads to screen out the best features for each type of load.

[0052] S3: Concatenate the classified demand-side resources and feature variables, and use the position embedding method to perform position encoding to form input data input1, input2, and input3.

[0053] S4: Input data input1, input2 and input3 are input into the multi-layer perceptron model MLP to capture nonlinear relationships and complete the prediction of the transaction scale of demand-side resource demand response;

[0054] Specifically, input1 is input into the first multilayer perceptron, MLP1 is performed to obtain the first output output_1, and output_1 is connected to input2 to generate a first connection signal link_1; link_1 is input into the second multilayer perceptron, MLP2 is performed to obtain the second output output_2, and output_2 is connected to input3 to generate a second connection signal link_2; link_2 is input into the third multilayer perceptron, MLP3 is performed to generate a final output signal link; the final output signal link is input into the linear layer for compression, and the compressed signal is input into the fully connected layer to generate the final demand-side resource demand response transaction scale prediction result.

[0055] Furthermore, a multilayer perceptron (MLP) is a classic feedforward artificial neural network that consists of several layers, including an input layer, one or more hidden layers, and an output layer. Each layer consists of multiple neurons, which are connected by weights. The main purpose of the MLP model is to learn the mapping relationship from input data to output data through these connections. In MLP, each neuron performs a weighted summation of the input signal and converts it into an output through an activation function.

[0056] Furthermore, the output of the jth neuron in layer l can be expressed by the following formula:

[0057]

[0058] in, The activation function is converted into the output of the jth neuron in the lth layer, ai (l-1) is the output of the i-th neuron in the l-1th layer, is the weight connecting the i-th neuron in the l-1th layer and the j-th neuron in the lth layer, is the bias term of the jth neuron in the lth layer.

[0059] Specifically, the activation function σ is usually a nonlinear function, such as Sigmoid, applied to the weighted sum To introduce nonlinearity, the specific formula is as follows:

[0060]

[0061] Furthermore, the training process of the MLP model usually uses the back propagation algorithm and the gradient descent method. During the training process, the objective function L is to minimize the difference between the predicted output and the true label. The specific formula is as follows:

[0062]

[0063] Where N is the number of training samples, is the kth target value of the nth sample, is the model's predicted value for the nth input sample at the kth output neuron, For the model, for the nth input sample x n The predicted value at the kth output neuron.

[0064] Furthermore, in order to better optimize the network weights, MLP usually introduces a regularization term to optimize the network weights. The specific formula is as follows:

[0065]

[0066] Among them, Ω is the regularization term, λ is the regularization coefficient, L is the number of network layers, M is the number of neurons in the lth layer, is the weight connecting the i-th neuron in the l-1th layer and the j-th neuron in the lth layer.

[0067] Furthermore, this embodiment also provides a demand-side resource demand response transaction scale calculation system based on LassoNet and S-MLP, including: a classification module, used to classify demand-side resources according to electricity consumption type; a screening module, used to input meteorological factors and classified demand-side resources into the LassoNet network model, and screen feature variables through regularization; an encoding module, used to splice the classified demand-side resources and feature variables, and use a position embedding method to perform position encoding to form input data input1, input2 and input3; a capture module, used to input input data input1, input2 and input3 into a multilayer perceptron model MLP, capture nonlinear relationships, and complete the prediction of the demand-side resource demand response transaction scale.

[0068] This embodiment also provides a computer device, which is applicable to the demand-side resource demand response transaction scale calculation method based on LassoNet and S-MLP, including a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute computer-executable instructions to implement the demand-side resource demand response transaction scale calculation method based on LassoNet and S-MLP as proposed in the above embodiment.

[0069] The computer device may be a terminal, and the computer device includes a processor, a memory, a communication interface, a display screen and an input device connected via a system bus. 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 and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be achieved through WIFI, an operator network, NFC (near field communication) or other technologies. The display screen of the computer device may be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device may be a touch layer covering the display screen, or a key, trackball or touchpad provided on the housing of the computer device, or an external keyboard, touchpad or mouse, etc.

[0070] The present embodiment also provides a storage medium having a computer program stored thereon, which implements the following steps when executed by a processor: classifying demand-side resources according to electricity consumption types; inputting meteorological factors and classified demand-side resources into a LassoNet network model, and filtering characteristic variables through regularization; concatenating the classified demand-side resources and characteristic variables, and position encoding them using a position embedding method to form input data input1, input2, and input3; inputting the input data input1, input2, and input3 into a multilayer perceptron model MLP to capture nonlinear relationships and complete the prediction of the scale of demand-side resource demand response transactions.

[0071] In summary, the present invention classifies demand-side resources according to electricity consumption types, and inputs meteorological factors and classified demand-side resources into the LassoNet network model to screen feature variables, then splices the classified resources with feature variables and encodes them using the position embedding method, and finally inputs these data into the multi-layer perceptron model MLP to capture nonlinear relationships, so as to achieve accurate prediction of the scale of demand-side resource demand response transactions. This method improves the flexibility of system response by effectively managing and regulating different types of resources; optimizes model computing resources and improves prediction accuracy through the feature selection capability of the LassoNet model; and the position embedding method ensures structured processing of data and enhances the model's understanding of complex relationships. Finally, through the nonlinear feature extraction of the MLP model, high-precision prediction of the scale of demand-side resource response transactions is achieved, improving the economy and stability of the system.

[0072] Example 2

[0073] Referring to Tables 1 and 2, which are the second embodiment of the present invention, this embodiment provides a method for calculating the transaction scale of demand-side resource demand response based on LassoNet and S-MLP. In order to verify the beneficial effects of the present invention, scientific demonstration is carried out through economic benefit calculation and simulation experiments.

[0074] Specifically, a demand-side resource trading dataset of a certain electricity market in the United States is used. The data types include industrial equipment, commercial air conditioning, lighting, industrial manufacturing equipment, battery energy storage systems, demand response incentives, etc. The meteorological data comes from the global solar radiation data base. The data types include temperature, dew point, wind speed, etc. The data sampling rate is 1h. The training set is from January 1, 2020 to November 30, 2020, and the test set is from December 1, 2020 to January 31, 2021.

[0075] Furthermore, this study uses three indicators to evaluate the performance of the model, including Mean Absolute Error (MAE), Mean Squared Error (MSE) and Root Mean Squared Error (RMSE). The specific formula is as follows:

[0076]

[0077]

[0078]

[0079] Among them, I i and Represent the true value and predicted value at time i respectively.

[0080] Furthermore, LassoNet is used to select features for different types of demand-side resources. First, a multi-layer perceptron (MLP) is constructed, and the L1 regularization term is integrated into the loss function of the neural network. By optimizing the network weights and feature coefficients during the training process, LassoNet compresses the coefficients of some features to zero, thus achieving feature selection for different demand-side resources.

[0081] Specifically, the experimental results are shown in Table 1. The features that contribute most to curtailable load, transferable load and energy storage resources are temperature, DCL and surface reflectivity, respectively. Taking these three features together with curtailable load, transferable load and energy storage resources as the input of the ST-MLP model can help the model better capture the relationship in the data and speed up the model training.

[0082] Table 1 Demand-side resource characteristics importance evaluation table

[0083]

[0084] Furthermore, in order to better compare the measurement capabilities of the demand-side resource demand response transaction scale based on LassoNet and S-MLP, this study also set up an ablation experiment to remove LassoNet, and set up comparative experiments of RFE-SMLP, LSTM, LassoNet-LSTM, RFE-LSTM and GRU models. As shown in Table 2, the various indicators of the proposed measurement method (the measurement method of the demand-side resource demand response transaction scale based on LassoNet and S-MLP) are better than other comparison models. The prediction method proposed in this paper performs better than other comparison models in all indicators. Its performance on the MSE indicator is 0.0006, which is 0.0003 lower than the performance of the SMLP model on MSE in the ablation experiment; its performance on the RMSE indicator is 0.0253, and its performance on the MAE indicator is 0.0215, which is at least 0.001 and 0.009 lower than other models.

[0085] Table 2 Performance index diagram of each model

[0086] Model MSE RMSE MAE LassoNet-SMLP 0.0006 0.0253 0.0215 SMLP 0.0009 0.0293 0.0236 RFE-SMLP 0.0007 0.0263 0.0224 LassoNet-LSTM 0.0009 0.0308 0.0260 RFE-LSTM 0.0018 0.0425 0.0354 LSTM 0.0021 0.0460 0.0344

[0087] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A method for calculating the transaction scale of demand-side resource demand response based on LassoNet and S-MLP, characterized by: include, Classify demand-side resources according to electricity usage type; The meteorological factors and classified demand-side resources are input into the LassoNet network model, and the characteristic variables are screened through regularization; The classified demand-side resources and feature variables are concatenated and positionally encoded using the position embedding method to form input data input1, input2, and input3; Input data input1, input2 and input3 are input into the multi-layer perceptron model MLP to capture nonlinear relationships and complete the prediction of the transaction scale of demand-side resource demand response.

2. The method for calculating the transaction scale of demand-side resource demand response based on LassoNet and S-MLP as claimed in claim 1, characterized in that: The multi-layer perceptron model MLP includes the following steps: In MLP, neurons perform weighted summation on input signals and convert them into the output of the jth neuron in layer l through an activation function. The specific formula is as follows: With j (l) =b j (l) +Σ i IN i,j (l) ai (l-1) ; in, The activation function is converted into the output of the jth neuron in the lth layer, ai (l-1) is the output of the i-th neuron in the l-1th layer, is the weight connecting the i-th neuron in the l-1th layer and the j-th neuron in the lth layer, is the bias term of the jth neuron in the lth layer; Apply the activation function σ to the weighted sum To introduce nonlinearity, the specific formula is as follows: a j (l) =σ(z j (l) )。 3. The method for calculating the transaction scale of demand-side resource demand response based on LassoNet and S-MLP as claimed in claim 2, characterized in that: Also includes, During the training process, the objective function L is to minimize the difference between the predicted output and the true label. The specific formula is as follows: Where N is the number of training samples, is the kth target value of the nth sample, is the model's predicted value for the nth input sample at the kth output neuron, a k (L) (x n ) is the model for n input samples x n The predicted value at the kth output neuron; Introduce regularization terms to optimize network weights. The specific formula is as follows: Among them, Ω is the regularization term, λ is the regularization coefficient, L is the number of network layers, M is the number of neurons in the lth layer, is the weight connecting the i-th neuron in the l-1th layer and the j-th neuron in the lth layer.

4. The method for calculating the transaction scale of demand-side resource demand response based on LassoNet and S-MLP as claimed in claim 3, characterized in that: Also includes, Input 1 is input into the first multilayer perceptron, and MLP1 is performed to obtain the first output output_1. At the same time, output_1 is connected to input 2 to generate the first connection signal link_1. Input link_1 to the second multilayer perceptron, perform MLP2 processing, obtain the second output output_2, and connect output_2 with input3 to generate a second connection signal link_2; Input link_2 into the third multi-layer perceptron for MLP3 processing to generate the final output signal link; The final output signal link is input into the linear layer for compression, and the compressed signal is input into the fully connected layer to generate the final prediction result of the demand-side resource demand response transaction scale.

5. The method for calculating the transaction scale of demand-side resource demand response based on LassoNet and S-MLP as claimed in claim 1, characterized in that: The objective function of the LassoNet network model includes a loss function and a regularization term; the LassoNet network model includes the following steps: The difference between the predicted label and the true label of the LassoNet network model is predicted by the loss function. The specific formula is as follows: Among them, min θ,W L(θ,W) is to find the minimum value of the parameter θ and the first hidden layer weight W in the loss function L(θ,W), θ1 is the L1 norm of the network layer number of θ, ||Wj (1) || ∞ is the infinite norm of the jth weight vector of the first hidden layer weight W, λ is the regularization coefficient, |θ j | is the absolute value of the jth parameter of θ, and M is the maximum allowed value of the weight; The weights of the Skip layer are penalized through the regularization term L1 norm, the sparsity of the Skip layer weights is controlled, and feature variables are selected to achieve complete control over the network feature selection.

6. The method for calculating the transaction scale of demand-side resource demand response based on LassoNet and S-MLP as claimed in claim 4, characterized in that: The architecture of the LassoNet includes a Skip layer and a feedforward neural network of arbitrary depth; the Skip layer and the first hidden layer together pass through a hierarchical soft threshold optimizer to capture any nonlinearity in the feature variables; and include, When M = 0, all hidden units are inactive, and the LassoNet network model degenerates into a Lasso model that only uses Skip connections; When M tends to infinity, the LassoNet network model recovers the standard unregularized feedforward neural network.

7. The method for calculating the transaction scale of demand-side resource demand response based on LassoNet and S-MLP as claimed in claim 1, characterized in that: The classification includes curtailable loads, transferable loads and energy storage resources; the curtailable loads include electricity loads that are temporarily reduced or suspended by users in demand response events; the transferable loads include electricity loads that are adjusted between different time periods; the energy storage resources include battery energy storage systems that provide electricity or absorb excess electricity when the power system needs it.

8. A demand-side resource demand response transaction scale calculation system based on LassoNet and S-MLP, based on the demand-side resource demand response transaction scale calculation method based on LassoNet and S-MLP according to any one of claims 1 to 7, characterized in that: include, A classification module is used to classify demand-side resources according to electricity consumption types; The screening module is used to input meteorological factors and classified demand-side resources into the LassoNet network model and screen feature variables through regularization; The encoding module is used to concatenate the classified demand-side resources and feature variables, and use the position embedding method to perform position encoding to form input data input1, input2 and input3; The capture module is used to input the input data input1, input2 and input3 into the multi-layer perceptron model MLP, capture the nonlinear relationship, and complete the prediction of the transaction scale of the demand-side resource demand response.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method for calculating the transaction scale of demand-side resource demand response based on LassoNet and S-MLP as described in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method for calculating the transaction scale of demand-side resource demand response based on LassoNet and S-MLP according to any one of claims 1 to 7 are implemented.

Images