Load ultra-short-term prediction method and system based on least squares support vector machine

Abstract

The present disclosure belongs to the technical field of power systems, and particularly relates to a load ultra-short-term prediction method and system based on a least squares support vector machine, which comprises the following steps: obtaining historical load of a load point; constructing a prediction model by using a least squares support vector machine; predicting the ultra-short-term load of the load point based on the obtained historical load and the constructed prediction model to obtain a first load ultra-short-term prediction result; calculating the error of the obtained first load ultra-short-term prediction result; predicting the obtained error to obtain an error prediction result; and obtaining a load ultra-short-term prediction result based on the obtained first load ultra-short-term prediction result and the error prediction result.

Description

TECHNICAL FIELD

[0001] The present disclosure belongs to the technical field of power systems, and particularly relates to a load ultra-short-term prediction method and system based on a least squares support vector machine. BACKGROUND

[0002] The statements in this section merely provide background information related to the present disclosure and do not necessarily constitute prior art.

[0003] Ultra-short-term load prediction is to predict the load change situation in the future 5 minutes to 1 hour, and the fast prediction speed and high precision are the significant features of the ultra-short-term load prediction. The traditional methods applied to the ultra-short-term load prediction mainly include time series method, curve extrapolation method and load derivation method, etc.; the ultra-short-term prediction methods based on modern intelligent technology include artificial neural network and support vector machine, etc. The traditional prediction method has simple model, fast prediction speed, and high prediction precision for smooth load sequence, but is not suitable for the prediction of load sequence with large volatility because it only focuses on the fitting of data, i.e. the search for load sequence law. The artificial intelligence prediction method is suitable for processing nonlinear and uncertain problems due to its strong self-adaptive and learning ability, but has the defects of long learning time and easy falling into local minimum, resulting in model divergence. SUMMARY

[0004] In order to solve the above problems, the present disclosure provides a load ultra-short-term prediction method and system based on a least squares support vector machine, which comprehensively considers the calculation speed and precision, adopts the least squares support vector machine for prediction according to the characteristics of fast calculation speed of the least squares support vector machine and better prediction effect of the least squares support vector machine on the load sequence with large volatility than the traditional prediction method; starts from the error aspect, uses the powerful nonlinear mapping ability of the BP neural network to predict the error of the least squares support vector machine prediction, corrects the predicted value through the error correction link, and improves the prediction precision.

[0005] According to some embodiments, the first aspect of the present disclosure provides a load ultra-short-term prediction method based on a least squares support vector machine, which adopts the following technical scheme:

[0006] A load ultra-short-term prediction method based on a least squares support vector machine, comprising:

[0007] obtaining historical load of a load point;

[0008] constructing a prediction model by using the least squares support vector machine;

[0009] predicting the ultra-short-term load of the load point based on the obtained historical load and the constructed prediction model, to obtain a first load ultra-short-term prediction result;

[0010] errors of the obtained first load ultra-short-term prediction result;

[0011] predicting the obtained errors to obtain error prediction results;

[0012] obtaining load ultra-short-term prediction results based on the obtained first load ultra-short-term prediction results and the error prediction results.

[0013] As a further technical limitation, in the process of constructing the prediction model, a least squares support vector machine regression algorithm based on a fast leave-one-out method is adopted to cut the obtained historical load samples into a plurality of sample subsets, the sample subsets are trained by a fast leave-one-out cross-validation method, and repeated iterations are performed until all the sample subsets are trained.

[0014] As a further technical limitation, the process of predicting the ultra-short-term load of the load point is:

[0015] inputting a training sample set of historical load and a test sample set of historical load;

[0016] cross-validation is performed by using a fast leave-one-out method to find an optimal parameter pair;

[0017] determining whether the found parameter is an optimal value, if yes, the next step is performed, otherwise, the previous step is returned to continue cross-validation;

[0018] outputting the best parameter pair based on the test sample set of historical load;

[0019] establishing a prediction model, inputting sample data, performing prediction, and obtaining a first load ultra-short-term prediction result.

[0020] As a further technical limitation, the FLOO method is used to cross-validate the obtained first load ultra-short-term prediction result to obtain errors of the first load ultra-short-term prediction result.

[0021] As a further technical limitation, in the process of predicting the obtained errors, a prediction error correction link is introduced to correct the errors of the obtained first load ultra-short-term prediction result, a BP neural network algorithm is used for rolling prediction to obtain error prediction results.

[0022] Further, the process of training the BP neural network is:

[0023] forming a training sample set of the BP neural network based on the obtained errors of the first load ultra-short-term prediction result;

[0024] constructing a BP neural network, training the BP neural network based on the formed training sample set of the BP neural network, and obtaining error prediction results.

[0025] As a further technical limitation, the load ultra-short-term prediction result is a cumulative sum of the first load ultra-short-term prediction result and the error prediction result.

[0026] According to some embodiments, a second aspect of the present disclosure provides a load ultra-short-term prediction system based on a least squares support vector machine, adopting the following technical solution:

[0027] A load ultra-short-term prediction system based on a least squares support vector machine comprises:

[0028] An acquisition module configured to acquire historical load of a load point;

[0029] A modeling module configured to build a prediction model by using a least squares support vector machine;

[0030] A first prediction module configured to predict ultra-short-term load of the load point based on the acquired historical load and the built prediction model, to obtain a first load ultra-short-term prediction result;

[0031] An error calculation module configured to calculate error of the obtained first load ultra-short-term prediction result;

[0032] A second prediction module configured to predict the obtained error, to obtain an error prediction result;

[0033] A prediction module configured to obtain a load ultra-short-term prediction result based on the obtained first load ultra-short-term prediction result and error prediction result.

[0034] According to some embodiments, a third aspect of the present disclosure provides a computer readable storage medium, adopting the following technical solution:

[0035] A computer readable storage medium having a program stored thereon, the program being executed by a processor to implement the steps in the load ultra-short-term prediction method based on a least squares support vector machine according to the first aspect of the present disclosure.

[0036] According to some embodiments, a fourth aspect of the present disclosure provides an electronic device, adopting the following technical solution:

[0037] An electronic device comprising a memory, a processor, and a program stored in the memory and executable on the processor, the processor implementing the steps in the load ultra-short-term prediction method based on a least squares support vector machine according to the first aspect of the present disclosure when executing the program.

[0038] Compared with the prior art, the present disclosure has the following beneficial effects:

[0039] The disclosure provides a load ultra-short-term algorithm based on a least square support vector machine. Since the algorithm has a large prediction error at a load inflection point, the disclosure introduces a prediction error correction link. A BP neural network algorithm is used to predict the prediction error of the least square support vector machine. The prediction result is the sum of the calculation results of the two algorithms. Actual examples are simulated to verify that the prediction accuracy of the algorithm after adding the correction link is improved. BRIEF DESCRIPTION OF DRAWINGS

[0040] The accompanying drawings, which form a part of this disclosure, are intended to provide further understanding of the disclosure and are incorporated herein in conjunction with the description of the exemplary embodiments of the present disclosure and their description. They do not constitute an improper limitation of the present disclosure.

[0041] Figure 1 is a flowchart of the load ultra-short-term prediction method based on the least square support vector machine in the first embodiment of the present disclosure;

[0042] Figure 2 is a flowchart of the least square support vector machine in the first embodiment of the present disclosure;

[0043] Figure 3 is a flowchart of the BP neural network training in the first embodiment of the present disclosure;

[0044] Figure 4 is a flowchart of the ultra-short-term load prediction in the first embodiment of the present disclosure;

[0045] Figure 5 is a working day load change curve schematic diagram in the first embodiment of the present disclosure;

[0046] Figure 6 is a weekend load change curve schematic diagram in the first embodiment of the present disclosure;

[0047] Figure 7 is a load ultra-short-term prediction curve schematic diagram on December 5 in the first embodiment of the present disclosure;

[0048] Figure 8 is a load ultra-short-term prediction error schematic diagram on December 5 in the first embodiment of the present disclosure;

[0049] Figure 9 is a load ultra-short-term prediction curve schematic diagram on Saturday in the first embodiment of the present disclosure;

[0050] Figure 10 is a load ultra-short-term prediction error schematic diagram on Saturday in the first embodiment of the present disclosure;

[0051] Figure 11 is a prediction curve and actual value curve schematic diagram on December 7 in the first embodiment of the present disclosure;

[0052] Figure 12is a prediction error schematic diagram on December 7 in the first embodiment of the disclosure;

[0053] Figure 13 is a prediction curve schematic diagram after error correction in the first embodiment of the disclosure;

[0054] Figure 14 is a prediction error schematic diagram in the first embodiment of the disclosure;

[0055] Figure 15 is a structural block diagram of a load ultra-short-term prediction system based on a least squares support vector machine in the second embodiment of the disclosure. DETAILED DESCRIPTION

[0056] The disclosure will be further described below in conjunction with the drawings and embodiments.

[0057] It should be noted that the following detailed description is exemplary and is intended to provide further explanation of the disclosure. Unless otherwise indicated, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which the disclosure belongs.

[0058] It should be noted that the terms used herein are only for the purpose of describing specific embodiments and are not intended to limit exemplary embodiments according to the disclosure. As used herein, the singular form is intended to include the plural form unless the context clearly indicates otherwise, and it should also be understood that when the terms "comprise" and / or "include" are used in the specification, there is a presence of the features, steps, operations, devices, components and / or combinations thereof.

[0059] The embodiments in the disclosure and the features in the embodiments can be combined with each other without conflict.

[0060] Embodiment one

[0061] The first embodiment of the disclosure introduces a load ultra-short-term prediction method based on a least squares support vector machine.

[0062] As Figure 1 shown, a load ultra-short-term prediction method based on a least squares support vector machine comprises:

[0063] Obtaining historical load of a load point;

[0064] Using a least squares support vector machine to construct a prediction model;

[0065] Based on the obtained historical load and the constructed prediction model, predicting the ultra-short-term load of the load point to obtain a first load ultra-short-term prediction result;

[0066] Calculating the error of the obtained first load ultra-short-term prediction result;

[0067] predicting the obtained error to obtain an error prediction result;

[0068] obtaining a load ultra-short-term prediction result based on the obtained first load ultra-short-term prediction result and the error prediction result.

[0069] Next, the prediction method in the present embodiment will be described in detail.

[0070] Support Vector Machine (SVM) is a prediction method based on statistics. The training problem is essentially a classical quadratic programming problem, which can avoid local optimal solution and has a unique global optimal solution, and can use many mature algorithms in optimization theory. In solving nonlinear problems, SVM can map nonlinear sample set to high-dimensional feature space through nonlinear mapping to obtain linearly separable data set, and use kernel function to replace the corresponding inner product operation in high-dimensional space. Support Vector Regression (SVR) algorithm is widely used in power system load prediction by introducing insensitive loss function and kernel function, and has good prediction performance and generalization ability.

[0071] Short-term load prediction based on SVM has higher prediction accuracy than traditional methods. It is based on Vapnik-Chervonenkis Dimension (VC) theory and structural risk minimization principle, and is ideal for solving small sample, nonlinear, high dimension and local minimum point problems. It has high fitting accuracy, strong generalization ability and global optimization, and fully considers various factors affecting load. The shortcomings are that the selection of parameters and kernel function should be considered, which is generally determined by experience and affects the prediction effect. Therefore, many improved algorithms are proposed, such as SVM based on linear programming, least squares support vector machine (LSSVM), weighted support vector machine (Weighted Support Vector Machine, W-SVM), etc. They can improve the performance of SVM to some extent, among which LS-SVM is the most commonly used method.

[0072] In the present embodiment, a least squares support vector machine regression algorithm based on fast leave-one-out method is used.

[0073] Given a training set where x i is called the i-th input, y i is called the target value corresponding to x i , and l is the number of samples. The goal of the regression problem is to determine the optimal function f(x) such that the regression function f(x) has the following form:

[0074]

[0075] In least squares support vector machine, the optimization problem corresponding to the regression problem is

[0076]

[0077] It can be seen that the constraint condition is different from that of support vector machine, and the constraint condition here is equality, while the constraint condition of support vector machine is inequality, where e i The loss function is represented by a quadratic function, which is also different from that of support vector machine. The Lagrange function corresponding to the least squares support vector machine is:

[0078]

[0079] The optimal value condition of formula (3) is:

[0080]

[0081] The equation can be obtained by writing formula (4) in matrix form and eliminating w and e as follows

[0082]

[0083] In the formula, α=[α1,α2,…α l ] T ;Y=[y1,y2,…y N ] T α and b can be obtained through formula (5).

[0084] It has been proved that the accuracy of the prediction algorithm has nothing to do with the selection of the kernel function. Common kernel functions include: linear kernel: k(x i ,x j )=x T i x j , polynomial kernel: k(x i ,x j )=(x T i x j ) d , Gaussian kernel: The Gaussian radial basis function RBF kernel is selected in this embodiment.

[0085] The model of least squares support vector machine regression is

[0086]

[0087] The matrix form of α i (i=1,2,…,l) and b is respectively

[0088]

[0089] As Figure 2 shown, the algorithm steps of the least squares support vector machine are as follows:

[0090] (1) input the training sample set M(x i ,y i ) and the test sample set N(x i ,y i );

[0091] (2) adopt the fast leave-one-out method for cross validation to find the optimal parameter pair gam and sig2;

[0092] (3) judge whether the parameters are optimal, if optimal, then proceed to the next step, otherwise return to the second step to continue cross validation;

[0093] (4) test the sample, and output the optimal parameter pair gam and sig2;

[0094] (5) establish a prediction model, input sample data, and make prediction.

[0095] The optimization algorithm of the least squares support vector machine adopts the fast leave-one-out method cross validation. Cross validation (CV) is also called cyclic estimation, which is mainly used in the performance evaluation application of modeling; its principle is: in a given number of samples, cut them into smaller subsets. Then take out most of the samples to model, and leave a small part of the sample to detect the established model, and evaluate the reliability of the model according to the error of the detection result, and then repeat the above process until all the samples are used for modeling and prediction.

[0096] Leave one out cross validation (LOOCV for short), namely leave one out method; its characteristics are that each time for validation, only one sample is left for detection, and all the remaining samples are used for modeling; then repeat the process until all the samples are used for detection once. Therefore, leave one out method is suitable for small sample data detection. The fast leave-one-out method has small calculation amount, which makes the online selection of model parameters possible. The prediction algorithm in the embodiment needs to optimize the selected parameters γ and σ.

[0097] When the number of samples is l, for convenience of calculation, let P=H -1 , the matrix is reblocked into

[0098]

[0099] Then, there are formula (9) and formula (10) respectively

[0100]

[0101]

[0102] In the formula, And Respectively, α l And y l Remove the first row, that is, When FLOO is used for cross-validation, the parameters of the LSSVR model after the ith iteration are And Then, when the first sample is excluded, there is formula (11)

[0103]

[0104] Therefore, the estimated value of the first sample based on the new model parameters obtained by FLOO is

[0105]

[0106] Substitute formula (9) and formula (10) into formula (12) in turn, and rearrange to obtain

[0107]

[0108] At the same time, for formula (8), there is formula (14) by the inverse of block matrix theorem

[0109]

[0110] In the formula Therefore, the prediction error of the first sample using FLOO for cross-validation can be represented as

[0111]

[0112] In the formula The first row and the first column of Q -1 Note that when solving the linear equations of formula (5), changing the order of each equation does not affect the final solution, so the prediction error of the ith sample based on the FLOO method can be represented as

[0113]

[0114] For Q of formula (8), the inverse of block matrix formula can be obtained Q -1 And the relationship between P l ​

[0115]

[0116] where is a scalar. Thus, Q -1 and P l The previous element on the diagonal satisfies the following relationship

[0117]

[0118] where Substituting equation (18) into equation (16) gives

[0119]

[0120] Thus, the full-sample total error based on FLOO is

[0121]

[0122] Since a i and P l,ii are known, the computation of s and o is very small, thus The computational complexity of is about O(l). When computing based on FLOO, there is only one inversion (i.e., computing P l ), which is 1 / l of the ordinary LOO algorithm.

[0123] The advantage of BP neural network is that it has strong autonomous learning ability, and no accurate mathematical model and physical model need to be established before network training. Through training, it can realize any complex nonlinear mapping function. In the embodiment, the BP neural network rolling prediction method is used to predict the error of the load. According to the similarity and regularity of the load prediction error, the error at the same time of the previous four days is taken as the input to predict the error at the same time of the current day. Therefore, in order to ensure the consistency of the network model, in the selection of the training sample structure, the input training sample of the i-th day at t time is set as X i,t = (e i-4,t , e i-3,t , e i-2,t , e i-1,t ), wherein: e i-4,t is the prediction error at the same time of the previous four days, e i-3,t is the prediction error at the same time of the previous three days, e i-2,t is the prediction error at the same time of the previous two days, and e i-1,t is the prediction error at the same time of the previous one day.

[0124] It is evident that at each time step, the input used to train the BP neural network is four, thus the input layer of the BP neural network is set to contain four neurons, and these four input layer neurons correspond to four network inputs, i.e., e. i-4,t ,e i-3,t ,e i-2,t ,e i-1,t The output layer contains one neuron, and one output neuron corresponds to one output e. i,t This refers to the predicted error value at time t on day i. Since there is still no reliable theoretical support for the selection of the number of hidden layers, through repeated experiments, it was found that the training effect is optimal when the number of hidden layer neurons is 8. Therefore, the BP neural network structure was ultimately determined to be a 4-8-1 type, i.e., 4 input neurons, 8 hidden layer neurons, and 1 output neuron. The training process of the BP neural network involves finding a nonlinear mapping relationship where the prediction error at time t on the previous 4 days is used as input to obtain the prediction error at time t on the predicted day.

[0125] like Figure 3 The training process of the BP neural network at each prediction time shown is as follows:

[0126] Prepare historical load data L for 96 time points per day for n consecutive days. i,t Historical data is read in and least squares support phasor machine is used to predict the load at each time point on day n, resulting in the corresponding load forecast value L'. i,t ;

[0127] The prediction error e at each time point on day n is obtained. i,t =L i,t -L' i,t And store the prediction error;

[0128] According to the defined training sample structure, the training sample set X of the BP neural network is formed by the stored prediction errors. i,t ;

[0129] For each time step, 96 backpropagation (BP) neural networks are constructed, each corresponding to one of the 96 time steps, with X as the input. i,t The output is e i,t Train the BP neural network;

[0130] Save the 96 trained BP neural networks to prepare for the load forecasting phase.

[0131] Based on the completed BP network training, the method of this embodiment can be used to perform ultra-short-term load forecasting at time t. The basic process is as follows: Figure 4 As shown; the operational process for ultra-short-term load forecasting is as follows:

[0132] Prepare the prediction sample for the least square support vector machine and the error prediction sample for the BP neural network model at the i-th day and t time;

[0133] Input the prediction sample at the i-th day and t time, and obtain the load prediction value l i,t at t time, which is not corrected by error;

[0134] Load the trained BP neural network corresponding to t time, and input the error prediction sample into the network to obtain the error prediction value e i,t at the i-th day and t time;

[0135] Add the load prediction value l i,t obtained by the LSSVM prediction and the error prediction value e i,t obtained by the corresponding BP neural network to obtain the final ultra-short-term load prediction value L i,t at the i-th day and t time. i,t i,t .

[0136] The three-phase total active power data of a 35kV bus load point in a certain distribution network area are used to analyze the prediction method described in this embodiment. One set of data is obtained every 15 minutes, and there are 96 sets of data in one day. The sample data are divided into two parts. One part is 12.3-12.7 weekdays, and the other part is 12.1-12.2, 12.8-12.9, and 12.15-12.6 consecutive weekends. The active power changes of the load point during weekdays and weekends are shown in Figure 5 and Figure 6 . During weekdays, the load changes have obvious regularity. The active power is very large and reaches the peak value at about 9 o'clock in the morning and from 1 o'clock to 3 o'clock in the afternoon. The active power suddenly decreases at about 11 o'clock in the morning and 6:30 in the afternoon, which are also the two points with large fluctuations. During weekends, it can be seen that the load change on Saturday is the same as that on weekdays, and the load on Sunday is generally lower than that on weekdays.

[0137] In this embodiment, the least square support vector machine algorithm is used to predict the load of the load point. The prediction period is 15 minutes. The data of the previous four time points [t1, t2, t3, t4] are used to predict the load at the next time point t5. The data are shifted, and the true value at t5 time and the values predicted at t2, t3, and t4 times are used to predict the value at the next time. For weekdays, the training data use 96 data of December 4, and the test data use 96 data of December 5 for prediction. For weekends, the training data use 96 load data of December 1, Saturday, and the test data use 96 load data of December 8, Saturday. The prediction curves and prediction errors are shown in Figure 7 , Figure 8 ,​Figure 9 and Figure 10 as shown.

[0138] Based on Figure 7 , Figure 8 , Figure 9 and Figure 10 , it can be seen that the prediction effect of LSSVM prediction algorithm is good, basically in line with the change trend of load, and the prediction error is relatively small. The average relative error rate on December 5 is 3.514%, the average relative error rate on December 6 is 4.413%, and the average relative error rate on December 7 is 4.443%. Other prediction results are not shown one by one, and the relative error rate basically maintains between 3% and 5%. For the load prediction on December 8, Saturday, because the load change on December 2, Saturday, is very similar to the load change on December 8, the prediction effect is very good, the average relative error rate is 2.3657%, and the maximum relative error rate is 6.7263%. However, the prediction error of least square support vector machine is still large at the inflection point and in the time period with large volatility, and the relative error rate even reaches 12%. Therefore, BP neural network is used to correct the error, and the prediction curve is shown in Figure 11 , Figure 12 , Figure 13 and Figure 14 .

[0139] From Figure 11 , Figure 12 , Figure 13 and Figure 14 , it can be seen that after adding error correction on December 7, the average relative error rate decreases from 4.443% to 1.832%, and the maximum relative error rate decreases from 11.815% to 6.731%. The prediction accuracy is obviously improved, and the relative error rate is basically below 4%. The prediction error of almost every moment is reduced, especially the prediction accuracy at the inflection point is well improved, and the error is obviously reduced. However, despite this, the relative error rate at the inflection point is still relatively high in terms of numerical value. From the above results, it can be seen that the load ultra-short-term prediction accuracy in the distribution network is lower than that in the power grid, which is closely related to the randomness of the load connected to the distribution network. The load change regularity in the distribution network is not as strong as that in the large power grid, especially on weekends, the load change is very different. Therefore, it is very important to select load data similar to the prediction day as training samples to improve the prediction accuracy. As the core link of situation prediction, the prediction accuracy of load ultra-short-term prediction directly determines the accuracy of future situation prediction of the distribution network. The current prediction accuracy cannot fully meet the demand of distribution network situation prediction, and improving the load ultra-short-term prediction accuracy will still be the key work of situation prediction.

[0140] This embodiment is based on the load ultra-short-term algorithm using least squares support vector machine. Since the algorithm has a large prediction error at the load inflection point, a prediction error correction stage is introduced. The BP neural network algorithm is used to predict the prediction error of the least squares support vector machine. The prediction result is the sum of the calculation results of the two algorithms. Through simulation analysis, it is verified that the prediction accuracy of the algorithm is improved after adding the correction stage.

[0141] Example 2

[0142] Embodiment 2 of this disclosure introduces a load ultra-short-term forecasting system based on least squares support vector machine.

[0143] like Figure 15 The load ultra-short-term forecasting system shown includes:

[0144] The acquisition module is configured to acquire the historical load of the load point;

[0145] The modeling module is configured to build a predictive model using least squares support vector machines.

[0146] The first forecasting module is configured to forecast the ultra-short-term load of the load point based on the acquired historical load and the constructed forecasting model, and obtain the first ultra-short-term load forecasting result.

[0147] An error calculation module is configured to calculate the error of the obtained first ultra-short-term load forecast result;

[0148] The second prediction module is configured to predict the obtained error and obtain the error prediction result.

[0149] The prediction module is configured to obtain the ultra-short-term load prediction result based on the obtained first ultra-short-term load prediction result and the error prediction result.

[0150] The detailed steps are the same as those of the load ultra-short-term forecasting method based on least squares support vector machine provided in Example 1, and will not be repeated here.

[0151] Example 3

[0152] Embodiment 3 of this disclosure provides a computer-readable storage medium.

[0153] A computer-readable storage medium having a program stored thereon that, when executed by a processor, implements the steps in the load ultra-short-term forecasting method based on least squares support vector machine as described in Embodiment 1 of this disclosure.

[0154] The detailed steps are the same as those of the load ultra-short-term forecasting method based on least squares support vector machine provided in Example 1, and will not be repeated here.

[0155] Embodiment four

[0156] Embodiment four of the present disclosure provides an electronic device.

[0157] An electronic device includes a memory, a processor, and a program stored in the memory and executable on the processor, and the processor implements the steps in the load ultra-short-term prediction method based on the least squares support vector machine as described in embodiment one of the present disclosure when executing the program.

[0158] The detailed steps are the same as the load ultra-short-term prediction method based on the least squares support vector machine provided in embodiment one, and will not be repeated here.

[0159] The above only describes the preferred embodiments of the present disclosure and is not intended to limit the present disclosure. Those skilled in the art can make various modifications and changes to the present disclosure. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present disclosure shall be included in the protection scope of the present disclosure.

Claims

1. A load ultra-short-term prediction method based on a least squares support vector machine, characterized by, The application relates to a method for predicting a short-term load of a load point. The method comprises the following steps: acquiring historical load of the load point; adopting a least square support vector machine to build a prediction model; in the process of building the prediction model, a least square support vector machine regression algorithm based on a fast leave-one-out method is adopted, the acquired historical load sample is cut into a plurality of sample subsets, the sample subsets are trained through a fast leave-one-out cross validation method, and iteration is repeated until all the sample subsets are trained; based on the acquired historical load and the built prediction model, a short-term load of the load point is predicted to obtain a first short-term load prediction result; the error of the least square support vector machine prediction is predicted by using the powerful nonlinear mapping capability of a BP neural network, the predicted value is corrected through an error correction link, and the prediction accuracy is improved; the error of the obtained first short-term load prediction result is calculated; In the process of predicting the obtained error, a prediction error correction link is introduced to correct the error of the obtained first load ultra-short-term prediction result, and the error prediction result is obtained through the obtained error is predicted to obtain an error prediction result; a neural network algorithm rolling prediction. the first short-term load prediction result and the error prediction result are used to obtain a short-term load prediction result.

2. The load ultra-short-term forecasting method based on least squares support vector machine according to claim 1, characterized in that, The process of predicting the short-term load of the load point comprises the following steps: inputting a training sample set of historical load and a test sample set of historical load; adopting a fast leave-one-out method for cross validation to find an optimal parameter pair; judging whether the found parameter is an optimal value; if yes, the next step is performed; otherwise, the previous step is returned to continue cross validation; based on the test sample set of historical load, the optimal parameter pair is outputted; a prediction model is built, sample data is inputted, prediction is performed, and a first short-term load prediction result is obtained.

3. The load ultra-short-term forecasting method based on least squares support vector machine according to claim 1, characterized in that, The first load ultra-short-term prediction result is obtained by using the first load ultra-short-term prediction method. FLOO The error of the first load ultra-short-term prediction result is obtained by cross-validation on the obtained first load ultra-short-term prediction result.

4. The load ultra-short-term forecasting method based on least squares support vector machine according to claim 1, characterized in that, The the first short-term load prediction result and the error prediction result are used to obtain a short-term load prediction result. The process of neural network training is: Based on the error of the obtained first load ultra-short-term prediction result, form The application relates to a method for predicting a short-term load of a load point. a training sample set of the neural network; constructing The method comprises the following steps: a neural network based on the formed a acquiring module configured to acquire historical load of the load point; training sample set for training the neural network a modeling module configured to adopt a least square support vector machine to build a prediction model; the neural network, obtaining an error prediction result.

5. The load ultra-short-term forecasting method based on least squares support vector machine according to claim 1, characterized in that, in the process of building the prediction model, a least square support vector machine regression algorithm based on a fast leave-one-out method is adopted, the acquired historical load sample is cut into a plurality of sample subsets, the sample subsets are trained through a fast leave-one-out cross validation method, and iteration is repeated until all the sample subsets are trained; 6. A load ultra-short-term prediction system based on least squares support vector machines, employing the load ultra-short-term prediction method based on least squares support vector machines according to any one of claims 1 to 5, characterized by, a first prediction module configured to, based on the acquired historical load and the built prediction model, predict a short-term load of the load point to obtain a first short-term load prediction result; the error of the least square support vector machine prediction is predicted by using the powerful nonlinear mapping capability of a BP neural network, the predicted value is corrected through an error correction link, and the prediction accuracy is improved; an error calculation module configured to calculate the error of the obtained first short-term load prediction result; a second prediction module configured to predict the obtained error to obtain an error prediction result; a prediction module configured to, based on the obtained first short-term load prediction result and the error prediction result, obtain a short-term load prediction result. the first short-term load prediction result and the error prediction result are used to obtain a short-term load prediction result. FLOO ​ ​ 7. The load ultra-short-term forecasting system based on least squares support vector machine according to claim 6, characterized in that, ​ 8. The load ultra-short-term forecasting system based on least squares support vector machine as claimed in claim 6, characterized in that, The first load ultra-short-term prediction result is obtained by using the first load ultra-short-term prediction method. ​ The error of the first load ultra-short-term prediction result is obtained by cross-validation on the obtained first load ultra-short-term prediction result.

9. A computer-readable storage medium having stored thereon a program, characterized in that, The program, when executed by the processor, implements the steps in the load ultra-short-term prediction method based on the least squares support vector machine according to any one of claims 1-5.

10. An electronic device comprising a memory, a processor, and a program stored on the memory and executable on the processor, characterized in that, The processor implements the steps in the load ultra-short-term prediction method based on the least squares support vector machine according to any one of claims 1-5 when executing the program.

Images