A method for evaluating interruptible load demand response potential

Abstract

The application discloses a kind of interruptible load demand response potential evaluation method, the method comprises the following steps: collecting user power consumption behavior data and pre-processing to obtain pre-processed user power consumption behavior data;The pre-processed user power consumption behavior data is clustered to divide the user into different categories, and the user in each category has similar power consumption behavior;Interruptible load demand prediction model is constructed, and the power consumption behavior data of different categories of users is analyzed according to the interruptible load demand prediction model to quantify the interruptible power consumption load of user;The interruptible load of different user categories obtained by prediction is weighted and summed to obtain the interruptible load demand response potential evaluation result.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of interruptible load demand evaluation, and particularly relates to a method for evaluating interruptible load demand response potential. BACKGROUND

[0002] In modern power systems, with the rapid development of renewable energy and the year-by-year growth of power load demand, power systems are facing unprecedented challenges. Traditional power system dispatching methods mainly rely on the regulation ability of the power generation side, but with the increasing penetration of intermittent renewable energy such as wind and solar energy in the power grid, the regulation ability of the power generation side is limited. Due to the instability and intermittency of renewable energy, the imbalance between power supply and demand gradually appears, and the flexibility requirement of power systems is increasing.

[0003] Under this background, demand response gradually becomes an important means to improve the flexibility of power systems. Demand response refers to stimulating users to actively change their power consumption behavior during power grid load stress or peak price through incentive measures or price signals, thereby helping the power grid to achieve supply and demand balance. Interruptible load is the core of demand response. Interruptible load refers to the load that power users can temporarily interrupt or reduce when the power grid needs. By reasonably evaluating the demand response potential of interruptible load, power systems can better manage the power supply and demand contradiction and achieve more efficient dispatching. SUMMARY

[0004] To solve the above technical problems, the present application provides a method for evaluating the demand response potential of interruptible load, which comprises:

[0005] S1: collecting user power consumption behavior data and preprocessing to obtain preprocessed user power consumption behavior data, and using Gaussian weighted K nearest neighbor algorithm for missing data filling in preprocessing;

[0006] S2: clustering the preprocessed user power consumption behavior data to divide users into different categories, comprising:

[0007] S21: randomly initializing user category centers according to a pre-set number of user categories C;

[0008] S22: randomly initializing membership matrix U=[u ij ];

[0009] Wherein:

[0010] u ij represents the membership degree of the power consumption behavior data of the i-th user to the j-th user category, and the membership degree satisfies the constraint condition, and the calculation formula is:

[0011]

[0012] S23: updating the center of each user category center according to the membership of the power consumption behavior data, and the updating calculation formula is:

[0013]

[0014] wherein:

[0015] v j represents the center of the jth user category;

[0016] s i represents the power consumption behavior data of the ith user;

[0017] u ij represents the membership of the power consumption behavior data of the ith user belonging to the user category j;

[0018] m represents a fuzzification coefficient;

[0019] N represents the number of user power consumption behavior data;

[0020] S24: updating the membership of the sample according to the optimized DTW distance between the center of the user category and the user power consumption behavior data, and the membership updating formula is:

[0021]

[0022] wherein:

[0023] DTW(s i ,v j ) represents the optimized DTW distance between the power consumption behavior data s i of the ith user and the center v j of the jth user category;

[0024] S25: repeatedly performing the step S23 and the step S24 until the membership matrix U converges, and dividing the users into different categories according to the calculated membership, wherein the convergence is that the membership change after each iteration is lower than a preset threshold;

[0025] The users within each category have similar power consumption behavior, and the fuzzy clustering based on the optimized dynamic bending distance is a measurement method of the similar power consumption behavior;

[0026] S3: constructing an interruptible load demand prediction model, comprising:

[0027] S31: performing stationary detection on the power consumption behavior data of the users in different categories by using ADF test, and if the power consumption behavior data of the users is not stationary, performing first-order difference processing to obtain the stationary power consumption behavior data of the users;

[0028] S32: calculating the autoregressive order and the moving average order of the interruptible load demand prediction model according to the stationary user electricity consumption behavior data by using the autocorrelation function and the partial autocorrelation function;

[0029] S33: obtaining the optimal model parameters by optimizing the model by using the improved Newton algorithm;

[0030] S34: instantiating the model according to the optimal model parameters obtained by optimization, and performing electricity load prediction by using the model instance, and taking the difference between the predicted user electricity load and the actual user electricity load as the interruptible load demand of the corresponding user;

[0031] The model quantifies the interruptible electricity load of users by analyzing the electricity consumption behavior data of different categories of users.

[0032] S4: obtaining the interruptible load demand response potential evaluation result by weighted summing the interruptible load of different categories of users obtained by prediction.

[0033] Preferably, the user electricity consumption behavior data in the S1 step includes time stamp, electricity load, temperature and electricity price.

[0034] Preferably, the preprocessing adopts the Gaussian weighted K nearest neighbor algorithm for missing data filling, including:

[0035] S11: initializing the filling window K, and determining the neighbor data according to the filling window K with the to-be-filled data position as the center;

[0036] S12: calculating the weight of each neighbor data corresponding to the missing data by using the Gaussian weighting function, and the closer the neighbor data, the greater the weight, and the calculation formula is:

[0037]

[0038] Wherein:

[0039] x represents the to-be-filled user electricity consumption behavior data;

[0040] x j represents the user electricity consumption behavior data at time j;

[0041] w j represents the weight of the user electricity consumption behavior data at time j

[0042] d(x,x j ) represents the time interval between the user electricity consumption behavior data at time j and the to-be-filled user electricity consumption behavior data;

[0043] σ represents the bandwidth parameter;

[0044] exp represents the exponential function;

[0045] S13: Weighted filling the missing data according to the calculated weight of the neighbor data, the calculation formula is:

[0046]

[0047] Wherein:

[0048] Indicates the calculated missing data;

[0049] S14: Repeat the execution of steps S12 to S13 on the missing data in the user electricity consumption behavior data until all the missing data is filled to obtain the preprocessed user electricity consumption behavior data.

[0050] Preferably: the optimized DTW distance between the user electricity consumption behavior data in the S24 step comprises:

[0051] The boundary constraint DTW distance is used to limit the search range of the DTW path, reduce the calculation complexity, and the specific process comprises:

[0052] S24.1: Initialize the cumulative cost matrix D for storing the distance between the first user electricity consumption behavior data s1 and the second user electricity consumption behavior data s2, and the initial state of the cost matrix is set as D(i,j)=∞, which represents that the distance between the first user i time before electricity consumption behavior data and the second user j time before electricity consumption behavior data is infinite size, and the initial strip boundary constraint width parameter w is initialized;

[0053] S24.2: Calculate the data point distance d(i,j) between the first user electricity consumption behavior data i time data and the second user electricity consumption behavior data j time data, and the calculation formula is:

[0054] d(i,j) = (s1(i)-s2(j)) 2

[0055] Wherein:

[0056] s1(i) represents the first user electricity consumption behavior data i time data;

[0057] s2(j) represents the second user electricity consumption behavior data j time data;

[0058] S24.3: Iteratively update the cumulative cost matrix, take the D(L1,L2) value in the calculated cumulative cost matrix as the optimized DTW distance between the first user electricity consumption behavior data and the second user electricity consumption behavior data, and the update calculation formula is:

[0059] D(i, j) = d(i, j) + min(D(i - 1, j), D(i, j - 1), D(i - 1, j - 1))

[0060] wherein:

[0061] The iterative update is only iterated within the strip boundary constraint region satisfying |i-j|≤w, and other positions remain unchanged, L1 represents the length of the first user electricity behavior data, and L2 represents the length of the second user electricity behavior data.

[0062] Preferably, the optimal model parameters in the S33 step are obtained by optimizing the model using the improved quasi-Newton algorithm, comprising:

[0063] S33.1: constructing a target function according to the interruptible load demand prediction model, and the calculation formula is:

[0064]

[0065] wherein:

[0066] N ′ represents the total number of user electricity behavior data contained in the user category;

[0067] θ represents the interruptible load demand prediction model parameter;

[0068] ∈ i (θ) represents the prediction deviation of the interruptible load demand prediction model with θ as the parameter;

[0069] S33.2: randomly initializing the model parameter θ0 and the Hessian matrix B0, and calculating the initial model prediction deviation ∈(θ0) and the gradient initializing the maximum number of iterations;

[0070] S33.3: calculating the search direction according to the Hessian matrix and the gradient calculated in the current iteration round, and the calculation formula is:

[0071] p k =-B k g k

[0072] wherein:

[0073] B k represents the Hessian matrix calculated in the kth iteration round;

[0074] g k represents the gradient calculated in the kth iteration round;

[0075] S33.4: updating the model parameter according to the calculated search direction, and the calculation formula is:

[0076] θ k+1 = θ k + α k p k

[0077] wherein:

[0078] p k denotes the search direction obtained by kth iteration calculation;

[0079] θ k denotes the model parameter of kth iteration;

[0080] θ k+1 denotes the model parameter of k+1th iteration;

[0081] α k denotes the step size parameter of kth iteration obtained by line search method;

[0082] S33.5: updating the gradient and Hessian matrix according to the calculated model parameter;

[0083] S33.6: iteratively performing steps S33.3 to S33.5 until the maximum iteration number is reached, and outputting the optimal model parameter.

[0084] Preferably, the step S33.5 of updating the gradient and Hessian matrix according to the calculated model parameter comprises:

[0085] A1: updating the gradient g k+1 according to the calculated model parameter, and calculating the parameter difference s k = θ k+1 - θ k and the gradient difference y k = g k+1 - g k ;

[0086] wherein:

[0087] s k denotes the parameter difference value of kth iteration;

[0088] y k denotes the gradient difference value of kth iteration;

[0089] A2: estimating the updated Hessian matrix using the calculated parameter difference value and gradient difference value, and the calculation formula is:

[0090]

[0091] wherein:

[0092] B k denotes the Hessian matrix obtained by kth iteration calculation;

[0093] B k+1 represents the Hessian matrix obtained by k+1 rounds of iterative calculation;

[0094] T represents the transpose operation.

[0095] Preferably, the weighted summation of the predicted interruptible loads of different user categories in step S4 to obtain the interruptible load demand response potential evaluation result includes:

[0096] S41: Calculate the weight parameters of user interruptible load for different user categories using AHP analytical hierarchy process;

[0097] S42: Obtain an interruptible load demand response potential evaluation result by performing weighted summation based on the calculated user interruptible load weight parameter and the user interruptible load.

[0098] Technical effects and advantages of the present invention:

[0099] 1. In the data preprocessing stage, the present invention uses the Gaussian weighted K nearest neighbor algorithm to fill in missing data. Compared with the traditional KNN, the use of Gaussian weighting can better consider the distance weight between adjacent data points, reduce filling errors, and improve the accuracy and robustness of preprocessed data.

[0100] 2. In the clustering stage, the present invention uses a fuzzy clustering algorithm that optimizes dynamic bending distance to capture users' electricity usage behavior patterns, which improves computing efficiency and can more accurately characterize electricity usage behavior patterns.

[0101] 3. Aiming at the problem of predicting interruptible load, the present invention introduces the ARIMA model to perform periodic modeling based on historical electricity consumption behavior data, which has good adaptability and accuracy in predicting user interruptible load.

[0102] 4. The present invention utilizes an improved quasi-Newton algorithm to perform model optimization, thereby improving model optimization efficiency and achieving rapid demand response. BRIEF DESCRIPTION OF THE DRAWINGS

[0103] Figure 1 It is a flowchart of this application; DETAILED DESCRIPTION

[0104] The application is further described in detail below with the accompanying drawings and specific embodiments. The embodiments of the application are given for illustrative and descriptive purposes only and are not intended to be exhaustive or to limit the application to the forms disclosed. Many modifications and variations will be apparent to those of ordinary skill in the art. Embodiments are chosen and described in order to best explain the principles of the application and its practical application to thereby enable others skilled in the art to best utilize the application in various embodiments and with various modifications as are suited to the particular use contemplated.

[0105] Referring to Figure 1 In the embodiment, a method for evaluating interruptible load demand response potential is provided. The execution subject of the method includes, but is not limited to, at least one of electronic devices such as a server and a terminal, which can be configured to execute the method provided in the embodiment. In other words, the method can be executed by software or hardware installed in a terminal device or a server device. The software can be a blockchain platform. The server includes, but is not limited to, a single server, a server cluster, a cloud server, or a cloud server cluster.

[0106] A method for evaluating interruptible load demand response potential includes the following steps:

[0107] S1: Collect user electricity consumption behavior data and pre-process to obtain pre-processed user electricity consumption behavior data, the user electricity consumption behavior data including timestamp, electricity load, temperature, and electricity price, and the pre-processing uses a Gaussian weighted K nearest neighbor algorithm to fill in missing data.

[0108] In the S1 step, the pre-processing uses a Gaussian weighted K nearest neighbor algorithm to fill in missing data, including:

[0109] S11: Initialize the filling window K, and determine the neighbor data according to the filling window K with the to-be-filled data position as the center;

[0110] S12: Calculate the weight of each neighbor data corresponding to the missing data using a Gaussian weighting function, and the closer the neighbor data, the greater the weight, and the calculation formula is:

[0111]

[0112] Wherein:

[0113] x represents the to-be-filled user electricity consumption behavior data;

[0114] x j represents the user electricity consumption behavior data at time j;

[0115] w j represents the weight of the user electricity consumption behavior data at time j

[0116] d(x, x j ) represents the time interval between the user's power consumption behavior data at time j and the to-be-filled user's power consumption behavior data;

[0117] σ represents the bandwidth parameter;

[0118] exp represents the exponential function;

[0119] S13: The to-be-filled data is filled according to the weight of the calculated neighbor data, and the calculation formula is:

[0120]

[0121] Wherein:

[0122] represents the calculated to-be-filled data;

[0123] S14: The missing data in the user's power consumption behavior data is repeatedly executed according to steps S12 to S13 until all missing data is filled to obtain the preprocessed user's power consumption behavior data.

[0124] S2: The preprocessed user's power consumption behavior data is clustered to divide the users into different categories, and the users in each category have similar power consumption behaviors, wherein the fuzzy clustering based on the optimized dynamic bending distance is a measurement method of the similar power consumption behaviors.

[0125] The step S2 of clustering the preprocessed user's power consumption behavior data to divide the users into different categories comprises:

[0126] S21: Randomly initialize the user category center according to the pre-set number of user categories C;

[0127] S22: Randomly initialize the membership matrix U=[u ij ];

[0128] Wherein:

[0129] u ij represents the membership degree of the power consumption behavior data of the i-th user to the j-th user category, and the membership degree satisfies the constraint condition, and the calculation formula is:

[0130]

[0131] S23: Update the center of each user category center according to the membership degree of the power consumption behavior data, and the update calculation formula is:

[0132]

[0133] Wherein:

[0134] v j represents the center of the jth user category;

[0135] s i represents the power consumption behavior data of the ith user;

[0136] u ij represents the membership degree of the power consumption behavior data of the ith user belonging to the user category j;

[0137] m represents a fuzzification coefficient;

[0138] N represents the number of user power consumption behavior data;

[0139] S24: updating the membership degree of the sample according to the optimized DTW distance between the center of the user category and the user power consumption behavior data, and the membership degree updating formula is:

[0140]

[0141] wherein:

[0142] DTW(s i ,v j ) represents the optimized DTW distance between the power consumption behavior data s i of the ith user and the center v j of the jth user category;

[0143] S25: repeatedly performing the step S23 and the step S24 until the membership degree matrix U converges, and dividing the users into different categories according to the calculated membership degree, wherein the convergence is that the membership degree change after each iteration is lower than a preset threshold.

[0144] The optimized DTW distance between the user power consumption behavior data in the S24 step comprises:

[0145] The DTW distance with the boundary constraint limits the search range of the DTW path, and reduces the calculation complexity, and the specific process comprises:

[0146] S24.1: initializing an accumulated cost matrix D for storing the distance between the first user power consumption behavior data s1 and the second user power consumption behavior data s2, and setting the initial state of the cost matrix as D(i,j) = ∞, which represents that the distance between the power consumption behavior data of the first user i before the time and the power consumption behavior data of the second user j before the time is infinite, and initializing a strip boundary constraint width parameter w;

[0147] S24.2: calculating the data point distance d(i,j) between the data of the first user power consumption behavior data i time and the data of the second user power consumption behavior data j time, and the calculation formula is:

[0148] d(i,j)=(s1(i)-s2(j)) 2

[0149] in:

[0150] s1(i) represents the electricity usage behavior data of the first user at time i;

[0151] s2(j) represents the electricity usage behavior data of the second user at time j;

[0152] S24.3: Iteratively update and calculate the cumulative cost matrix, and use the D(L1, L2) value in the calculated cumulative cost matrix as the optimized DTW distance between the first user's electricity usage behavior data and the second user's electricity usage behavior data. The updated calculation formula is:

[0153] D(i,j)=d(i,j)+min(D(i-1,j),D(i,j-1),D(i-1,j-1))

[0154] in:

[0155] The iterative update is performed only within the strip boundary constraint area satisfying |ij|≤w, and other positions remain unchanged. L1 represents the length of the first user's electricity usage behavior data, and L2 represents the length of the second user's electricity usage behavior data.

[0156] S3: Construct an interruptible load demand prediction model, which quantifies the user's interruptible power load by analyzing the power consumption behavior data of different categories of users.

[0157] The interruptible load demand forecasting model is constructed in step S3, including:

[0158] S31: Use ADF test to detect the stability of different categories of user electricity usage behavior data. If the user electricity usage behavior data is not stable, perform first-order difference processing to obtain the stable user electricity usage behavior data;

[0159] S32: Calculate the autoregressive order and moving average order of the interruptible load demand prediction model based on the stabilized user electricity consumption behavior data using the autocorrelation function and the partial autocorrelation function;

[0160] S33: Use the improved quasi-Newton algorithm to optimize the model to obtain the optimal model parameters and improve the efficiency of model parameter solution;

[0161] S34: Instantiate the model according to the optimal model parameters obtained by the optimization solution, and use the model instance to predict the power load, and use the difference between the predicted user power load and the actual user power load as the interruptible load demand of the corresponding user.

[0162] The S33 step optimizes the model to obtain optimal model parameters by using an improved Newton algorithm, including:

[0163] S33.1: Construct a target function according to the interruptible load demand prediction model, and the calculation formula is:

[0164]

[0165] Wherein:

[0166] N ′ represents the total number of user behavior data contained in the user category;

[0167] θ represents the interruptible load demand prediction model parameter;

[0168] ∈ i (θ) represents the prediction error of the interruptible load demand prediction model with θ as the parameter;

[0169] S33.2: Randomly initialize the model parameter θ0 and the Hessian matrix B0, and calculate the initial model prediction error ∈(θ0) and the gradient Initialize the maximum number of iterations;

[0170] S33.3: Calculate the search direction according to the Hessian matrix and the gradient calculated in the current iteration round, and the calculation formula is:

[0171] p k =-B k g k

[0172] Wherein:

[0173] B k represents the Hessian matrix calculated in the kth iteration round;

[0174] g k represents the gradient calculated in the kth iteration round;

[0175] S33.4: Update the model parameter according to the calculated search direction, and the calculation formula is:

[0176] θ k+1 =θ k +α k p k

[0177] Wherein:

[0178] p k represents the search direction calculated in the kth iteration round;

[0179] θ k represents the model parameter in the kth round;

[0180] θ k+1 denotes the model parameters of k+1th iteration;

[0181] α k denotes the step size parameter of kth iteration calculated by line search method;

[0182] S33.5: updating the gradient and Hessian matrix according to the calculated model parameters;

[0183] S33.6: iteratively performing steps S33.3 to S33.5 until the maximum number of iterations is reached, and outputting the optimal model parameters.

[0184] The step S33.5 of updating the gradient and Hessian matrix according to the calculated model parameters comprises:

[0185] A1: updating the gradient g k+1 according to the calculated model parameters, and calculating the parameter difference s k = θ k+1 - θ k and the gradient difference y k = g k+1 - g k ;

[0186] wherein:

[0187] s k denotes the parameter difference value of kth iteration;

[0188] y k denotes the gradient difference value of kth iteration;

[0189] A2: estimating the updated Hessian matrix using the calculated parameter difference value and gradient difference value, and the calculation formula is:

[0190]

[0191] wherein:

[0192] B k denotes the Hessian matrix calculated by kth iteration;

[0193] B k+1 denotes the Hessian matrix calculated by k+1th iteration;

[0194] T denotes the transpose operation.

[0195] S4: performing weighted summation on the predicted interruptible load amount of different user categories to obtain the interruptible load demand response potential evaluation result, comprising:

[0196] S41: calculating the weight parameters of the interruptible load of different user categories by using the AHP method;

[0197] S42: obtaining the evaluation result of the interruptible load demand response potential by weighted summation according to the calculated weight parameters of the interruptible load of the user and the interruptible load of the user.

[0198] Obviously, the described embodiments are only 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 skilled in the art and related fields without creative labor should belong to the protection scope of the present application. The structures, devices and operation methods not specifically described and explained in the present application are implemented according to the conventional means in the art, unless specifically described and limited.

Claims

1. A method for evaluating the potential of interruptible load demand response, characterized in that: The method comprises: S1: Collect user electricity usage behavior data and preprocess it to obtain preprocessed user electricity usage behavior data. The preprocessing uses the Gaussian weighted K nearest neighbor algorithm to fill in missing data; S2: Cluster the pre-processed user electricity usage behavior data to divide users into different categories, including: S21: Randomly initialize the user category center according to the preset number of user categories C; S22: Randomly initialize the membership matrix U = [u ij ]; in: u ij The membership degree of the electricity consumption behavior data of the i-th user belongs to the user category j. The membership degree satisfies the constraint condition and is calculated as follows: S23: Update the center of each user category center according to the membership degree of the electricity consumption behavior data. The update calculation formula is: in: v j represents the center of the j-th user category; s i represents the electricity consumption behavior data of the i-th user; u ij Indicates the membership degree of the electricity consumption behavior data of the i-th user to user category j; m represents the fuzzification coefficient; N represents the number of user electricity consumption behavior data; S24: Update the membership of the sample based on the optimized DTW distance between the center of the user category and the user's electricity consumption behavior data. The membership update formula is: in: DTW(s i ,v j ) represents the electricity consumption behavior data s of the i-th user i and the center v of the jth user category j The optimized DTW distance between them; S25: Repeat steps S23 and S24 until the membership matrix U converges, and divide the users into different categories according to the calculated membership. Convergence means that the membership change after each iteration is lower than a preset threshold. Users in each category have similar electricity usage behaviors, wherein fuzzy clustering based on optimized dynamic bending distance is a measurement method for the similar electricity usage behaviors; S3: Build an interruptible load demand forecasting model, including: S31: Use ADF test to detect the stability of different categories of user electricity usage behavior data. If the user electricity usage behavior data is not stable, perform first-order difference processing to obtain the stable user electricity usage behavior data; S32: Calculate the autoregressive order and moving average order of the interruptible load demand prediction model based on the stabilized user electricity consumption behavior data using the autocorrelation function and the partial autocorrelation function; S33: Optimize the model using the improved quasi-Newton algorithm to obtain the optimal model parameters; S34: Instantiate the model according to the optimal model parameters obtained through optimization, and use the model instance to perform power load forecasting, and use the difference between the predicted user power load and the actual user power load as the interruptible load demand of the corresponding user; The model quantifies the user's interruptible power load by analyzing the power consumption behavior data of different categories of users; S4: Perform weighted summation on the predicted interruptible loads of different user categories to obtain an evaluation result of the interruptible load demand response potential.

2. The method for evaluating interruptible load demand response potential according to claim 1, characterized in that: The user's electricity usage behavior data in step S1 includes timestamp, electricity load, temperature and electricity price.

3. The method for evaluating interruptible load demand response potential according to claim 1, wherein: The preprocessing uses the Gaussian weighted K-nearest neighbor algorithm to fill in missing data, including: S11: Initialize the filling window K, and determine the neighboring data based on the filling window K with the position of the data to be filled as the center; S12: Use the Gaussian weighting function to calculate the weight of each neighbor data corresponding to the missing data. The closer the distance, the greater the weight of the neighbor data. The calculation formula is: in: x represents the user's electricity consumption behavior data to be filled; x j represents the user's electricity consumption behavior data at time j; w j Represents the weight of the user's electricity consumption behavior data at time j d(x,x j ) represents the time interval between the user's electricity usage behavior data at time j and the user's electricity usage behavior data to be filled; σ represents the bandwidth parameter; exp represents the exponential function; S13: Perform weighted filling on the data to be filled according to the calculated weight of the neighboring data. The calculation formula is: in: Indicates the calculated data to be filled; S14: Repeat steps S12 to S13 for the missing data in the user's electricity usage behavior data until all missing data are filled in to obtain the pre-processed user's electricity usage behavior data.

4. The method for evaluating interruptible load demand response potential according to claim 1, wherein: The optimized DTW distance between the user electricity usage behavior data in step S24 includes: The DTW distance with boundary constraints is used to limit the search range of the DTW path and reduce the computational complexity. The specific process includes: S24.1: Initialize a cumulative cost matrix D for storing the distance between the first user's electricity usage behavior data s1 and the second user's electricity usage behavior data s2. The initial state of the cost matrix is ​​set to D(i, j) = ∞, indicating that the distance between the first user's electricity usage behavior data before time i and the second user's electricity usage behavior data before time j is infinite. Initialize the strip boundary constraint width parameter w. S24.2: Calculate the data point distance d(i, j) between the first user's electricity usage behavior data at time i and the second user's electricity usage behavior data at time j. The calculation formula is: d(i,j)=(s1(i)-s2(j)) 2 in: s1(i) represents the electricity usage behavior data of the first user at time i; s2(j) represents the electricity usage behavior data of the second user at time j; S24.3: Iteratively update and calculate the cumulative cost matrix, and use the D(L1, L2) value in the calculated cumulative cost matrix as the optimized DTW distance between the first user's electricity usage behavior data and the second user's electricity usage behavior data. The updated calculation formula is: D(i,j)=d(i,j)+min(D(i-1,j),D(i,j-1),D(i-1,j-1)) in: The iterative update is performed only within the strip boundary constraint area satisfying |ij|≤w, and other positions remain unchanged. L1 represents the length of the first user's electricity usage behavior data, and L2 represents the length of the second user's electricity usage behavior data.

5. The method for evaluating interruptible load demand response potential according to claim 1, wherein: In step S33, the improved quasi-Newton algorithm is used to optimize the model to obtain the optimal model parameters, including: S33.1: Construct the objective function based on the interruptible load demand forecast model. The calculation formula is: in: N ′ Indicates the total number of user electricity usage behavior data included in the user category; θ represents the interruptible load demand forecasting model parameter; ∈ i (θ) represents the forecast deviation of the interruptible load demand forecasting model with θ as the parameter; S33.2: Randomly initialize the model parameters θ0 and the Hessian matrix B0, and calculate the initial model prediction bias ∈ (θ0) and gradient Initialize the maximum number of iterations; S33.3: Calculate the search direction based on the Hessian matrix and gradient calculated in the current iteration round. The calculation formula is: p k =-B k g k in: B k Represents the Hessian matrix obtained by k rounds of iterative calculation; g k Represents the gradient calculated after k rounds of iterations; S33.4: Update the model parameters based on the calculated search direction. The calculation formula is: i k+1 =θ k +a k p k in: p k Represents the search direction obtained by k rounds of iterative calculation; θ k represents the model parameters of k rounds; θ k+1 represents the model parameters of the k+1 round; α k represents the step size parameter calculated by the line search method in round k; S33.5: Update the gradient and Hessian matrix based on the calculated model parameters; S33.6: Iterate steps S33.3 to S33.5 until the maximum number of iterations is reached and output the optimal model parameters.

6. The method for evaluating interruptible load demand response potential according to claim 5, characterized in that: In step S33.5, the gradient and Hessian matrix are updated according to the calculated model parameters, including: A1: Update the gradient g based on the calculated model parameters k+1 , and calculate the parameter difference s k =θ k+1 -θ k and the gradient difference y k =g k+1 -g k ; in: s k represents the parameter difference of k rounds; y k Represents the gradient difference of k rounds; A2: Use the calculated parameter difference and gradient difference to estimate and update the Hessian matrix. The calculation formula is: in: B k Represents the Hessian matrix obtained by k rounds of iterative calculation; B k+1 represents the Hessian matrix obtained by k+1 rounds of iterative calculation; T represents the transpose operation.

7. The method for evaluating interruptible load demand response potential according to claim 1, characterized in that: In step S4, the weighted sum of the predicted interruptible loads of different user categories is performed to obtain an interruptible load demand response potential evaluation result, including: S41: Calculate the weight parameters of user interruptible load for different user categories using AHP analytical hierarchy process; S42: Obtain an interruptible load demand response potential evaluation result by performing weighted summation based on the calculated user interruptible load weight parameter and the user interruptible load.

Images