A hybrid interval-point forecasting method for demand response baseline load

Abstract

The present invention discloses a hybrid interval-point forecasting method for demand response baseline load. The method divides historical baseline load and meteorological characteristic data into a training set and a test set. A convolutional neural network model is trained on the training set to extract the complex relationship between meteorological characteristics and baseline load. The meteorological characteristic data is input into the test set to obtain a baseline load prediction curve. The training and forecasting steps are repeated to obtain sample data and a confidence interval model for the baseline load forecast in the test set. A point forecasting model is obtained by calculating the root mean square error (RMSE) between the mean and median functions and the actual baseline load values. Finally, a baseline load forecast for a future time period is performed based on future meteorological characteristic data and the hybrid interval-point forecasting model for baseline load, with interval and point forecast results output. The method of the present invention not only reflects the uncertainty of baseline load forecasts through interval forecast results, but also provides point forecast results to facilitate the calculation of user load reduction.

Description

Technical Field

[0001] The present invention relates to the technical field of power load demand side management, and in particular to a hybrid interval-point forecasting method for demand response baseline load. Background Art

[0002] Against the backdrop of the development of the Energy Internet and smart grids, demand response, as a flexible electricity demand management strategy, has become a key technology for improving power system flexibility and reliability. Through pricing and incentive mechanisms, demand response guides users to adjust their electricity usage behavior and optimize their patterns, effectively balancing electricity supply and demand, alleviating grid pressure, and reducing operating costs. During the implementation of demand response, baseline load forecasting has become a key technology to evaluate its effectiveness. Its goal is to predict users' normal electricity usage behavior without participating in demand response. By comparing the baseline load with the actual load after the response is implemented, the user's load reduction can be calculated, thereby determining the reward amount for participating users in demand response. Therefore, the accuracy of the baseline load forecast is directly related to the user's economic benefits and willingness to respond.

[0003] However, due to the nonlinear and non-stationary nature of power load data and the complex influence of external environmental factors (such as meteorological conditions), the baseline load for demand response is difficult to accurately predict. To facilitate practical applications, traditional baseline load forecasting typically uses point prediction, which outputs a single, fixed forecast value. However, due to data noise, model uncertainty, and unpredictable user behavior, the results of point predictions cannot fully characterize the fluctuation range and potential risks of baseline load. Therefore, there is an urgent need to introduce interval prediction methods to more comprehensively reflect the uncertainty of baseline load. At the same time, for ease of application, it is necessary to provide point prediction results based on interval predictions. Summary of the Invention

[0004] The purpose of the present invention is to provide a hybrid interval-point forecasting method for demand response baseline load to solve the above defects.

[0005] In order to achieve the above object, the present invention provides the following technical solutions:

[0006] A hybrid interval-point forecasting method for demand response baseline load includes the following steps:

[0007] S1. Obtain historical baseline load and historical meteorological characteristic data, the total length of the data is T;

[0008] S2. Divide the historical baseline load and historical meteorological characteristic data into a training set and a test set. The time period of the training set is 1 to W, and the time period of the test set is W+1 to T.

[0009] S3. Build a convolutional neural network model and set its parameters;

[0010] S4, using the historical meteorological characteristic data and historical baseline load data in the training set as independent variables and dependent variables, respectively, to train the convolutional neural network model;

[0011] S5. Input the meteorological characteristic data in the test set into the convolutional neural network model and calculate the baseline load forecast curve of the test set for the i-th time I sim Indicates the total number of calculations, i indicates the current number of calculations, and i is initialized to 1;

[0012] S6. If i≤I sim If yes, assign i+1 to i and then execute step S4; otherwise, execute the next step S7;

[0013] S7. Load prediction based on test set baseline The data were used as random samples and the confidence interval of baseline load was obtained through statistical calculation;

[0014] S8. Load prediction based on test set baseline The data is taken as a random sample, and two statistics of baseline load are obtained through statistical calculation, namely the mean and median

[0015] S9. Calculate the root mean square error (RMSE) between the mean of the baseline load and the actual value of the baseline load. ave , the root mean square error (RMSE) between the median of the baseline load and the actual value of the baseline load med ; Compare RMSE ave and RMSE med Size, if RMSE ave If smaller, use the mean As a point forecast estimate of baseline load; if RMSE med If smaller, use the median as a point forecast estimate of baseline load;

[0016] S10. Based on the future meteorological characteristic data and the above-established demand response baseline load hybrid interval-point prediction model, the baseline load for the future period is predicted, and the prediction results are output: the confidence interval prediction value and the point prediction value of the baseline load.

[0017] Preferably, in step S1, the historical baseline load and historical meteorological characteristic data include: historical baseline load L(t) and historical meteorological characteristic M(t) = [K max (t),K min (t),K mean (t),Hmax (t),H min (t),H mean (t)]; where t = 1, 2, ..., T, T represents the total duration; K max (t) is the maximum temperature during the period t, K min (t) is the lowest temperature in the period t, K mean (t) is the average temperature during the period t, H max (t) is the maximum humidity during the period t, H min (t) is the minimum humidity during the period t, H mean (t) is the average humidity during the period t.

[0018] Preferably, in step S2, the historical baseline load and meteorological characteristic data are divided into a training set and a test set, as follows:

[0019] Training set: baseline load L(t) during period 1 to W, meteorological characteristics M(t) during period 1 to W = [K max (t),K min (t),K mean (t),H max (t),H min (t),H mean (t)]; where t = 1, 2, ..., W, and W represents the duration of the training data;

[0020] Test set: baseline load L(t) during (W+1)~T period, meteorological characteristics M(t) during (W+1)~T period = [K max (t),K min (t),K mean (t),H max (t),H min (t),H mean (t)]; where t = W+1, W+2, …, T, and T represents the total duration.

[0021] Preferably, in step S3, the parameters of the convolutional neural network model are set as follows: the input layer format is k rows, h columns, and z channels; the size of the neural network filter layer is k F ×h F , the number is n; the activation function selection of the activation layer; the discard probability of the Dropout layer is η a .

[0022] Preferably, in step S4, the convolutional neural network is trained using a stochastic gradient descent algorithm with a descent factor of F. drop The maximum number of training times and batches are E Max With S batch , the learning rate is v study.

[0023] Preferably, the specific steps of step S7 are as follows:

[0024] S71. Calculate the baseline load forecast curve of the test set in the period (W+1) to T according to formula (1) The mean function μ * (t), t=W+1,W+2,…,T;

[0025]

[0026] S72. Calculate the baseline load forecast curve of the test set within the period (W+1) to T according to formula (2) The standard deviation function σ * (t), t=W+1,W+2,…,T;

[0027]

[0028] S73. Calculate the baseline load forecast curve of the test set in the period (W+1) to T using equations (3) and (4). The confidence interval with confidence level α

[0029]

[0030] In formula (3) and formula (4), Z α / 2 is the standard normal quantile, and Z is determined by looking up the Z table α / 2 The value of and They are the lower and upper bounds of the confidence interval of the baseline load forecast curve of the test set in the period (W+1)~T.

[0031] Preferably, the specific steps of step S8 are as follows:

[0032] S81. Calculate I at t=W+1, W+2, ..., T according to formula (5) sim The mean of a random sample of data:

[0033]

[0034] S82, respectively set I at t=W+1, W+2, ..., T sim The random sample data are rearranged in order from small to large to obtain new random sample data, which is recorded as

[0035] If I sim If is an odd number, the test set baseline load prediction curve is calculated according to formula (6): the median of

[0036]

[0037] If I sim If is an even number, the test set baseline load prediction curve is calculated according to formula (7) the median of

[0038]

[0039] Preferably, in step S9, the RMSE ave , RMSE med , and its calculation formula is shown in formula (8) and (9), which are as follows:

[0040]

[0041] The beneficial effects of the present invention are:

[0042] (1) The hybrid interval-point prediction method for demand response baseline load of the present invention takes into account the impact of meteorological characteristics on user baseline load, quantifies the complex relationship between external meteorological factors such as temperature and humidity and baseline load, and is more in line with reality.

[0043] (2) The present invention provides a hybrid interval-point prediction method for demand response baseline load, which uses a convolutional neural network method to obtain the confidence interval prediction result of the baseline load, thereby being able to more comprehensively reflect the uncertainty of the baseline load prediction and fully characterize the fluctuation range and potential risks of the baseline load.

[0044] (3) The present invention provides a hybrid interval-point prediction method for demand response baseline load, which calculates the root mean square error between the mean and median statistics and the actual baseline load value in the test set, and selects the statistic with the smaller root mean square error as the point prediction method, which is convenient for accurately calculating the user's load reduction in demand response practice. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 Flowchart of the method of the present invention. DETAILED DESCRIPTION

[0046] The present invention is further described below with reference to the embodiments. It should be noted that these are merely examples and illustrations of the concept of the present invention. Those skilled in the art may make various modifications or additions to the described specific embodiments or replace them in a similar manner. As long as they do not deviate from the concept of the invention or exceed the scope defined by the claims, they should be deemed to fall within the scope of protection of the present invention.

[0047] Example 1:

[0048] Figure 1 Flowchart of the method of the present invention. Figure 1 As shown, a hybrid interval-point forecasting method for demand response baseline load includes the following steps:

[0049] S1. Obtain historical baseline load and historical meteorological characteristic data, the total length of the data is T;

[0050] S2. Divide the historical baseline load and historical meteorological characteristic data into a training set and a test set. The time period of the training set is 1 to W, and the time period of the test set is W+1 to T.

[0051] S3. Build a convolutional neural network model and set its parameters;

[0052] S4, using the historical meteorological characteristic data and historical baseline load data in the training set as independent variables and dependent variables, respectively, to train the convolutional neural network model;

[0053] S5. Input the meteorological characteristic data in the test set into the convolutional neural network model and calculate the baseline load forecast curve of the test set for the i-th time I sim Indicates the total number of calculations, i indicates the current number of calculations, and i is initialized to 1;

[0054] S6. If i≤I sim If yes, assign i+1 to i and then execute step S4; otherwise, execute the next step S7;

[0055] S7. Load prediction based on test set baseline The data were used as random samples and the confidence interval of baseline load was obtained through statistical calculation;

[0056] S8. Load prediction based on test set baseline The data is taken as a random sample, and two statistics of baseline load are obtained through statistical calculation, namely the mean and median

[0057] S9. Calculate the root mean square error (RMSE) between the mean of the baseline load and the actual value of the baseline load. ave , the root mean square error (RMSE) between the median of the baseline load and the actual value of the baseline load med ; Compare RMSE ave and RMSE med Size, if RMSE ave If smaller, use the mean As a point forecast estimate of baseline load; if RMSE med If smaller, use the median as a point forecast estimate of baseline load;

[0058] S10. Based on the future meteorological characteristic data and the above-established demand response baseline load hybrid interval-point prediction model, the baseline load for the future period is predicted, and the prediction results are output: the confidence interval prediction value and the point prediction value of the baseline load.

[0059] Example 2:

[0060] Figure 1 Flowchart of the method of the present invention. Figure 1 As shown, a hybrid interval-point forecasting method for demand response baseline load includes the following steps:

[0061] S1. Obtain historical baseline load and historical meteorological characteristic data.

[0062] Historical baseline load data:

[0063] Baseline load L(t), t = 1, 2, ..., T, where T represents the total duration.

[0064] Historical meteorological characteristic data:

[0065] Meteorological characteristics M(t)=[K max (t),K min (t),K mean (t),H max (t),H min (t),H mean (t)]; where K max (t) is the maximum temperature during the period t, K min (t) is the lowest temperature in the period t, K mean (t) is the average temperature during the period t, H max (t) is the maximum humidity during the period t, H min (t) is the minimum humidity during the period t, H mean (t) is the average humidity during the period t.

[0066] S2. Divide the historical baseline load and historical meteorological characteristic data into a training set and a test set. The data from the period 1 to W is used as the training set, and the data from the period (W+1) to T is used as the test set. The details are as follows:

[0067] Training set: baseline load L(t) during period 1 to W, meteorological characteristics M(t) during period 1 to W = [K max (t),K min (t),K mean (t),H max (t),H min (t),H mean (t)]; where t = 1, 2, ..., W, and W represents the duration of the training data;

[0068] Test set: baseline load L(t) during (W+1)~T period, meteorological characteristics M(t) during (W+1)~T period = [K max (t),K min (t),K mean (t),H max (t),H min (t),H mean (t)]; where t = W+1, W+2, …, T, and T represents the total duration.

[0069] In this embodiment, the proportion of the training set is 80%, and the proportion of the test set is 20%; and normalization preprocessing is performed on the training set data and the test set data.

[0070] S3. Build a convolutional neural network model and set its parameters.

[0071] The convolutional neural network model designed in this embodiment consists of an input layer, a convolution layer, an activation layer, a dropout layer, a fully connected layer, and an output layer. The input layer matches the size of the data set to the convolution layer by changing parameters such as the number of rows, columns, and channels of the input data. The convolution layer consists of multiple small filter layers, which perform convolution operations on the data set covered by them to extract local features. The activation layer enhances the network's expressive power through activation functions, allowing the network to learn more complex features. The dropout layer can reduce the risk of model overfitting. The fully connected layer integrates the features refined by each layer and ultimately outputs the network's prediction results. The output layer restores the original format of the data by changing parameters such as the number of rows, columns, and channels of the prediction results.

[0072] The parameters of the convolutional neural network model are as follows: the input layer format is k rows, h columns, and z channels; the size of the neural network filter layer is k F ×h F , the number is n; the activation function selection of the activation layer; the discard probability of the Dropout layer is η a ; The output layer format is x rows, y columns, and v channels.

[0073] In this embodiment, the input layer format is set to 6 rows, 1 column, and 1 channel; the size of the neural network filter layer is set to 3×1, and the number is 32; the activation layer adopts the ReLU activation function (which has unilateral inhibition characteristics and can prevent overfitting); the discarding probability of the Dropout layer is set to 0.2; and the output layer format is set to 1 row, 1 column, and 1 channel.

[0074] S4. Input the baseline load and meteorological characteristic data in the training set into the convolutional neural network model for training. The training process uses the stochastic gradient descent algorithm with a descent factor of F. drop The maximum number of training times and batches are EMax With S batch , the learning rate is v study .

[0075] In this embodiment, the drop factor is 0.5, the maximum number of training times is 200, the batch size is 30, and the learning rate is 0.01.

[0076] S5. Input the meteorological characteristic data in the test set into the convolutional neural network model and calculate the baseline load forecast curve of the test set for the i-th time period (W+1) to T.

[0077] Among them, I sim Represents the total number of calculations, i represents the current number of calculations, and i is initialized to 1.

[0078] S6. If i≤I sim If it is established, it means that the set total number of calculations has not been reached, then i is assigned a value of i+1, and then step S4 is executed; otherwise, it means that the set total number of calculations has been reached, and the next step S7 is executed.

[0079] S7. Load prediction based on test set baseline The data is used as a random sample, and the confidence interval of the test set baseline load forecast is obtained through statistical calculation. The specific steps are as follows:

[0080] S71. Calculate the baseline load forecast curve of the test set in the period (W+1) to T according to formula (1) The mean function μ * (t), t=W+1,W+2,…,T;

[0081]

[0082] S72. Calculate the baseline load forecast curve of the test set within the period (W+1) to T according to formula (2) The standard deviation function σ * (t), t=W+1,W+2,…,T;

[0083]

[0084] S73. Calculate the baseline load forecast curve of the test set in the period (W+1) to T using equations (3) and (4). The confidence interval with confidence level α

[0085]

[0086] In formula (3) and formula (4), Z α / 2 is the standard normal quantile, and Z is determined by looking up the Z table α / 2The value of and They are the lower and upper bounds of the confidence interval of baseline load in the period (W+1)~T respectively.

[0087] So far, the interval forecasting modeling of demand response baseline load has been completed.

[0088] S8. Load prediction based on test set baseline The data is used as a random sample, and two statistics of the test set baseline load forecast are obtained through statistical calculation, namely the mean and median The specific steps are as follows:

[0089] S81. Calculate I at t=W+1, W+2, ..., T according to formula (5) sim The mean of a random sample of data:

[0090]

[0091] S82, respectively set I at t=W+1, W+2, ..., T sim The random sample data are rearranged in order from small to large to obtain new random sample data, which is recorded as

[0092] If I sim If is an odd number, the median of the baseline load forecast curve of the test set is calculated according to formula (6):

[0093]

[0094] If I sim If is an even number, the median of the baseline load forecast curve of the test set is calculated according to formula (7):

[0095]

[0096] S9. Calculate the root mean square error (RMSE) between the mean of the baseline load and the actual value of the baseline load. ave , the root mean square error (RMSE) between the median of the baseline load and the actual value of the baseline load med .

[0097] First, RMSE ave , RMSE med , and its calculation formula is shown in formula (8) and (9), which are as follows:

[0098]

[0099] Then, choose the statistic with the smaller root mean square error as the estimate of the point forecast. That is, compare RMSEave and RMSE med Size: If RMSE ave If smaller, use the mean As a point forecast estimate of baseline load; if RMSE med If smaller, use the median Serves as a point forecast estimate of baseline load.

[0100] So far, the point forecast modeling of demand response baseline load has been completed.

[0101] Finally, based on the convolutional neural network model, a complete hybrid interval-point prediction model of demand response baseline load was established, which can predict future baseline load.

[0102] S10. Based on the future meteorological characteristic data and the above-established demand response baseline load hybrid interval-point prediction model, perform baseline load prediction for the future period and output the final prediction result.

[0103] The model output is denormalized to produce the final forecast, which includes a confidence interval and point forecast for the baseline load. The confidence interval reflects the uncertainty of the forecast, i.e., the range of fluctuation and potential risk of the baseline load. The point forecast can be used to calculate load reduction for users in practical applications.

[0104] It can be seen from Examples 1 and 2 that the hybrid interval-point prediction method for demand response baseline load of the present invention overcomes the deficiency of traditional baseline load point prediction that cannot reflect prediction uncertainty, can fully characterize the fluctuation range and potential risks of the baseline load, and provides point prediction results to facilitate the calculation of user load reduction.

[0105] First, historical meteorological characteristics and historical baseline load data are divided into a training set and a test set. A convolutional neural network is trained on the training set to extract the complex relationship between meteorological characteristics and baseline load. In the test set, meteorological characteristic test data is input to obtain the predicted value of the test set baseline load. Then, the training and prediction steps are repeated to obtain sample data for baseline load prediction. Next, a confidence interval prediction of the baseline load is obtained through statistical calculations. The baseline load statistics, mean and median, are calculated using the sample data for baseline load prediction. The root mean square error (RMSE) between these two statistics and the actual baseline load value in the test set is calculated, and the statistic with the smaller RMSE is used as the point prediction method. Finally, based on future meteorological characteristic data and the aforementioned baseline load hybrid interval-point prediction model, baseline load forecasts for future time periods are performed, and interval and point prediction results are output.

[0106] The hybrid interval-point forecasting method for demand response baseline load of the present invention takes into account the impact of meteorological characteristics on user baseline load, quantifies the complex relationship between external meteorological factors such as temperature and humidity and baseline load, and is more in line with reality.

[0107] The present invention provides a hybrid interval-point prediction method for demand response baseline load, which adopts a convolutional neural network method to obtain a confidence interval prediction result of the baseline load, thereby being able to more comprehensively reflect the uncertainty of the baseline load prediction and fully characterize the fluctuation range and potential risks of the baseline load.

[0108] The present invention provides a hybrid interval-point prediction method for demand response baseline load, which calculates the root mean square error between the mean and median statistics and the actual baseline load value in the test set, and selects the statistic with the smaller root mean square error as the point prediction method, so as to facilitate the accurate calculation of the user's load reduction in demand response practice.

[0109] The above is an exemplary description of the invention. Obviously, the specific implementation of the present invention is not limited to the above-mentioned method. As long as such non-substantial improvements are made by adopting the method concept and technical solution of the present invention, or the concept and technical solution of the invention are directly applied to other occasions without improvement, they are all within the scope of protection of the present invention.

Claims

1. A hybrid interval-point forecasting method for demand response baseline load, characterized in that: The following steps are involved: S1. Obtain historical baseline load and historical meteorological characteristic data, the total length of the data is T; S2. Divide the historical baseline load and historical meteorological characteristic data into a training set and a test set. The time period of the training set is 1 to W, and the time period of the test set is W+1 to T. S3. Build a convolutional neural network model and set its parameters; S4, using the historical meteorological characteristic data and historical baseline load data in the training set as independent variables and dependent variables, respectively, to train the convolutional neural network model; S5. Input the meteorological characteristic data in the test set into the convolutional neural network model and calculate the baseline load forecast curve of the test set for the i-th time i=1,2,...,I sim ,t=W+1,W+2,…,T;I sim Indicates the total number of calculations, i indicates the current number of calculations, and i is initialized to 1; S6. If i≤I sim If yes, assign i+1 to i, and then execute step S4; Otherwise, proceed to the next step S7; S7. Load prediction based on test set baseline i=1,2,...,I sim ,t=W+1,W+2,…,T data are used as random samples, and the confidence interval of baseline load is obtained through statistical calculation; S8. Load prediction based on test set baseline i=1,2,...,I sim ,t=W+1,W+2,…,T data are used as random samples, and two statistics of baseline load are obtained through statistical calculation, namely the mean and median S9. Calculate the root mean square error (RMSE) between the mean of the baseline load and the actual value of the baseline load. ave , the root mean square error (RMSE) between the median of the baseline load and the actual value of the baseline load med ; Compare RMSE ave and RMSE med Size, if RMSE ave If smaller, use the mean As a point forecast estimate of baseline load; if RMSE med If smaller, use the median as a point forecast estimate of baseline load; S10. Based on the future meteorological characteristic data and the above-established demand response baseline load hybrid interval-point prediction model, the baseline load for the future period is predicted, and the prediction results are output: the confidence interval prediction value and the point prediction value of the baseline load.

2. A hybrid interval-point forecasting method for demand response baseline load according to claim 1, characterized in that: In step S1, the historical baseline load and historical meteorological characteristic data include: historical baseline load L(t) and historical meteorological characteristic M(t) = [K max (t),K min (t),K mean (t),H max (t),H min (t),H mean (t)]; where t = 1, 2, ..., T, T represents the total duration; K max (t) is the maximum temperature during the period t, K min (t) is the lowest temperature in the period t, K mean (t) is the average temperature during the period t, H max (t) is the maximum humidity during the period t, H min (t) is the minimum humidity during the period t, H mean (t) is the average humidity during the period t.

3. A hybrid interval-point forecasting method for demand response baseline load according to claim 1, characterized in that: In step S2, the historical baseline load and historical meteorological characteristic data are divided into a training set and a test set, as follows: Training set: baseline load L(t) during period 1 to W, meteorological characteristics M(t) during period 1 to W = [K max (t),K min (t),K mean (t),H max (t),H min (t),H mean (t)]; where t = 1, 2, ..., W, and W represents the duration of the training data; Test set: baseline load L(t) during (W+1)~T period, meteorological characteristics M(t) during (W+1)~T period = [K max (t),K min (t),K mean (t),H max (t),H min (t),H mean (t)]; where t = W+1, W+2, …, T, and T represents the total duration.

4. A hybrid interval-point forecasting method for demand response baseline load according to claim 1, characterized in that: In step S3, the parameters of the convolutional neural network model are set as follows: The input layer format is k rows, h columns, and z channels; the size of the neural network filter layer is k F ×h F , the number is n; the activation function selection of the activation layer; the discard probability of the Dropout layer is η a .

5. A hybrid interval-point forecasting method for demand response baseline load according to claim 1, characterized in that: In step S4, the convolutional neural network is trained using a stochastic gradient descent algorithm with a descent factor of F. drop The maximum number of training times and batches are E Max With S batch , the learning rate is v study .

6. A hybrid interval-point forecasting method for demand response baseline load according to claim 1, characterized in that: The specific steps of step S7 are as follows: S71. Calculate the baseline load forecast curve of the test set within the period (W+1) to T according to formula (1) The mean function μ * (t), t=W+1,W+2,…,T; S72. Calculate the baseline load forecast curve of the test set within the period (W+1) to T according to formula (2) The standard deviation function σ * (t), t=W+1,W+2,…,T; S73. Calculate the baseline load forecast curve of the test set in the period (W+1) to T using equations (3) and (4). The confidence interval with confidence level α t=W+1,W+2,…,T; In formula (3) and formula (4), Z α / 2 is the standard normal quantile, and Z is determined by looking up the Z table α / 2 The value of and They are the lower and upper bounds of the confidence interval of the baseline load forecast curve of the test set in the period (W+1)~T.

7. A hybrid interval-point forecasting method for demand response baseline load according to claim 6, characterized in that: The specific steps of step S8 are as follows: S81. Calculate I at t=W+1, W+2, ..., T according to formula (5) sim The mean of a random sample of data: S82, respectively set I at t=W+1, W+2, ..., T sim The random sample data are rearranged in order from small to large to obtain new random sample data, which is recorded as i=1,2,…,I sim ,t=W+1,W+2,…,T; If I sim If is an odd number, the test set baseline load prediction curve is calculated according to formula (6): the median of If I sim If is an even number, the test set baseline load prediction curve is calculated according to formula (7) the median of 8. A hybrid interval-point forecasting method for demand response baseline load according to claim 7, characterized in that: In step S9, the RMSE ave , RMSE med , and its calculation formula is shown in formula (8) and (9), which are as follows:

Images