Multi-model fusion power daily load online prediction method and system
Through multi-model fusion method, including autoencoder, unified information coefficient and hiking optimization algorithm, the problems of data quality and model parameter selection in power load prediction are solved, the reliability and stability of prediction are improved, and efficient power load is achieved predict.
Patent Information
- Application Number
- CN202411153000.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-21
- Publication Date
- 2024-11-22
- Estimated Expiration
- 2044-08-21
AI Technical Summary
In the prior art, the power load prediction model affects the reliability and stability of power load prediction due to factors such as data quality, model parameter selection and calculation complexity.
The online prediction method of daily load of power is adopted with multi-model fusion, including autoencoder detecting and filling outliers, unifying information coefficients to extract features, hiking optimization algorithm selects KELM model parameters, using KELM model to predict single-day load, and obtains it through rolling strategy. The next day's prediction results are made, and the model performance is finally evaluated by multiple indicators.
It improves the accuracy and stability of power load prediction, effectively avoids data anomalies, improves the prediction accuracy and nonlinear processing capabilities of the KELM model, reduces the computational burden, and avoids model overfitting.
Smart Images

Figure CN119004075A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of power load forecasting, and in particular to a multi-model fusion power daily load online forecasting method. Background Art
[0002] With the development of social economy and the improvement of living standards, the demand for electricity has shown an increasing trend; especially during peak hours, the fluctuation of power load has put forward higher requirements on the stability of the power grid; daily load forecasting is a key task in power system management, which involves the prediction of power load in the next day to ensure the reliability and stability of power supply; accurate daily load forecasting can not only help power companies effectively dispatch power resources, but also optimize power grid operation, improve energy efficiency and reduce operating costs. Summary of the invention
[0003] The technical problems of the present invention are: With the rapid development of machine learning and deep learning technologies, more and more prediction models have been introduced into the field of load forecasting. The actual application effects of these models have affected the reliability and stability of power load forecasting due to factors such as data quality, model parameter selection, and computational complexity.
[0004] The purpose of the present invention is to solve the above problems and proposes a multi-model fusion daily power load online prediction method; specifically, it includes: using autoencoder and moving average to improve the data set, using unified information coefficient to extract high correlation features, using walking optimization algorithm to realize KELM model parameter selection, using KELM model to determine the single-day load forecast, based on the rolling strategy to obtain the next day's forecast results, and finally using a multi-indicator evaluation model to predict the results.
[0005] The technical solution of the present invention is: A multi-model fusion power daily load online prediction method includes the following steps: S1: using an autoencoder to detect outliers on the collected historical daily load data, and using moving average to fill the outliers to obtain an initial data set D1; S2: extract the characteristic information of the power load in the data set D1 using the unified information coefficient to obtain the input data set D2; S3: Divide the input data set D2 into a training set, a test set, and a validation set according to a fixed ratio, use the training set and the walking optimization algorithm to select the parameters of the KELM model, and use the KELM model to determine the single-day load forecast; S4: Keep the sample size unchanged, use the rolling strategy to update the input data set D2, and repeat steps S1 to S3 until the next day's forecast results are obtained; S5: Use multiple indicators to evaluate the prediction results and adjust the initial parameters synchronously based on the prediction results.
[0006] Furthermore, in step S1, due to problems such as acquisition system failure and transmission delay, the data in the database contain outliers, which are detected using an autoencoder, specifically including encoder training, reconstruction error calculation, and outlier detection; Preferably, the principle of outlier detection by the autoencoder is as follows: 1) The encoder is used to map the input data to a low-dimensional representation, and the decoder is used to restore the low-dimensional representation to the original data space; 2) The training process is completed with the goal of minimizing the difference between the input data and the reconstructed data, using the mean square error as the loss function; 3) Use the autoencoder to encode and decode the new data and calculate the reconstruction error of each data point; 4) Compare the reconstruction error with the threshold and identify outliers.
[0007] Preferably, in step S1, moving average is used to fill the detected outliers, and the formula is as follows:
[0008] In the formula, o t is the data to be interpolated, r is the time window data size, o i is the known data in the window.
[0009] Furthermore, in step S2, the characteristic information of the power load is extracted using the unified information coefficient, including the following sub-steps: S21: Mutual Information Calculation:
[0010] In the formula, is mutual information, p ( x , y ) is the joint probability distribution function; p ( x )and p ( y ) are respectively X and Y The marginal probability distribution function of X and Y are two sets of eigenvectors, x and y is the feature vector X and Y The corresponding element.
[0011] S22: Variable division:
[0012]
[0013] In the formula, L x and L y They are X and Y The partition unit length of x max and x min Variables X The maximum and minimum values of ; y max and y min Variables Y The maximum and minimum values of ; m and n They are X and Y The number of segments; is the partition grid size; k Indicates the sample size.
[0014] S23: Unified information coefficient calculation:
[0015] In the formula, Represents unified information, m and n They are X and Y The number of segments, min means taking the minimum value.
[0016] Furthermore, in step S3, the training set is used for the model to learn the patterns and features of the data and adjust the internal parameters of the model; the validation set is used to test the model to prevent overfitting; and the test set is used to finally evaluate the performance of the model and verify the model's generalization ability on new and unseen data.
[0017] Preferably, the hiking optimization method is based on the Tobler hiking function, and determines the walking speed of the hiker by considering the elevation of the terrain and the walking distance, including: 1) Initialize the hiker positions and evaluate the best hiker position, which is defined as the leader; 2) Calculate the hiker's initial speed based on the terrain slope, including the following calculation formula:
[0018]
[0019] In the formula, dh and dx represent the altitude and horizontal distance covered by the hiker, respectively. is the slope angle of the terrain, Is a hiker i In the t The speed of iterations; is the slope of the terrain, e is the base of the natural logarithm function.
[0020] 3) Update hiker speed through the social thinking of the traveler group and the personal cognition of individual travelers, as follows:
[0021] In the formula, is a random number in [0,1]; is the speed of traveler i at the tth iteration; It is the position of the leader; is the position of traveler i at the tth iteration; is the scan factor of traveler i, and the scan factor is in [1,2]. The scan factor ensures that travelers do not stray too far from the leader so that they can see the direction of the leader and receive the leader's signal.
[0022] 4) Updates to hiker locations are as follows:
[0023] In the formula, For travelers i In the t +1 iteration position; 5) Update the fitness value of the hiker group and make iterative convergence judgment. Once the current number of iterations reaches the maximum number of iterations, the iterative process ends.
[0024] Preferably, the KELM model includes an ELM model and a kernel function, including: 1) ELM prediction output F ( x ) is expressed as follows:
[0025] In the formula, q is the input vector, H ( q ) is the hidden node output matrix, represents the output weight; 2) Introducing the regularization coefficient and the unit matrix, transforming the solution process, the least squares solution of the output weight is:
[0026] In the formula, G is the target vector, for The least squares solution of I is the identity matrix, C represents the regularization coefficient.
[0027] 3) Change the random mapping to kernel mapping, and the kernel matrix expression is as follows:
[0028] In the formula, For nuclear elements, is the kernel function, q i and q j Represents the experimental input vector.
[0029] 4) The kernel function selects the Gaussian kernel function, the formula is as follows:
[0030] In the formula, is the kernel parameter, q i and q j represents the experimental input vector, exp is a natural exponential function.
[0031] 5) The final output is:
[0032] In the formula, n is the sample size, I is the identity matrix, C is the regularization coefficient, K is the kernel function, For nuclear elements, T is the transposition symbol, G is the target vector, for KELM Output.
[0033] Further, in step S4, the time range of the input data set is d Day d + N Daily load data, forecast target is d + N +1 day load data. As the data set continues to grow, the time range of the input data set is updated using a rolling strategy. d+ 1st d + N+ 1-day load data, the forecast target is d +N+ 2-day load data, in order to balance the model training time and maintain the generalization of the model, including the following representation:
[0034] In the formula, d For one day, including multiple sampling times, N is the sample size, Y d , Y d+1 , Y d+2 , Y d+N , Y d+N+1 Respectively expressed as d , No. d +1, No. d +2, No. d + N and d + N +1 day's load.
[0035] Furthermore, in step S5, a multi-index evaluation model is used to evaluate the prediction results, including root mean square error, explained variance score and mean absolute percentage error, including: 1) Root mean square error:
[0036] In the formula, is the root mean square error, N represents the sample size; is the true value; is the predicted value; 2) Explainable variance score
[0037] In the formula, is the explained variance score, var(·) is the variance; 3) Mean absolute percentage error
[0038] Where: is the mean absolute percentage error, N represents the sample size; is the true value; is the predicted value.
[0039] Preferably, the root mean square error is used to measure the degree of deviation between the predicted value and the true value, and its value range is The smaller the value, the smaller the prediction error of the model, and the stronger the prediction ability of the model; the explained variance score ranges from 0 to 1. When EVS is 1, it means that the model perfectly predicts the data; conversely, it means that the model explains the data less. The range of mean absolute percentage error is , MAPE greater than 100% indicates an inferior model, otherwise, it indicates that the model has a higher prediction accuracy.
[0040] Compared with the prior art, the beneficial effects of the present invention include: 1) The present invention discloses an online prediction method for daily power load based on multi-model fusion, which uses autoencoders and moving averages to improve the initial data set, uses unified information coefficients to extract high-correlation input features, and uses a walking optimization algorithm to select KELM model parameters. The KELM model is used to determine the single-day load forecast, and the next-day forecast results are obtained based on a rolling strategy. Finally, the prediction results of the model are evaluated using multiple indicators; 2) Considering that errors are inevitable in the process of data collection and transmission, resulting in a small offset between the collected data and the real data, the data recognition strategy based on the autoencoder of the present invention can effectively capture the outliers in the data and effectively avoid data omission and false detection; 3) Different daily load sequences have similar variation patterns, that is, there is daily periodicity. The interpolation technology using moving average of the present invention can achieve similar periodicity of the interpolated sequences; 4) Different from the common correlation coefficient, the unified information coefficient of the present invention can measure the linear correlation and nonlinear correlation between variables, and has a certain robustness to noise data, which is conducive to the extraction of characteristic variables; 5) The present invention uses the walking optimization algorithm to select parameters of the prediction model, which improves the prediction accuracy of the KELM model and its ability to handle nonlinear problems, and has high adaptability and stability; 6) Existing prediction models often predict future load demand through one-time training, ignoring the dynamic characteristics of load series changing with social development. The present invention updates the input data set in real time while maintaining the sample size, which can reduce the computational burden and avoid the problem of model overfitting. BRIEF DESCRIPTION OF THE DRAWINGS
[0041] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.
[0042] Figure 1 This is a flow chart of power load forecasting according to an embodiment of the present invention; Figure 2 Schematic diagram of an autoencoder according to an embodiment of the present invention; Figure 3 A schematic diagram showing a comparison between the actual value and the predicted value of the power load of each model in an embodiment of the present invention; Figure 4 4 is a system module structure diagram of an embodiment of the present invention. DETAILED DESCRIPTION
[0043] Embodiment 1: like Figure 1 As shown, according to the method of the present invention, the power load in New England from January 1, 2024 to June 31, 2024 is simulated and predicted, the total sample capacity is 4368, the sampling frequency is 1h, and the characteristic variables include marginal price, dry bulb temperature and wet bulb temperature.
[0044] A multi-model fusion online prediction method for daily power load includes the following steps: S1: Based on the autoencoder, outlier detection is performed on the collected historical daily load data, and the outliers are filled using moving average to obtain the initial data set D1; S2: extracting characteristic information of the power load in the data set D1 using the unified information coefficient to obtain a characteristic data set D2; S3: Divide the feature data set D2 into a training set, a test set, and a validation set according to a fixed ratio, use the KELM model to build a load forecasting model, use the training set and the walking optimization algorithm to select the KELM model parameters, and use the KELM model to determine the single-day load forecast; S4: Keep the sample size unchanged, use the rolling strategy to update the data set D2, and repeat steps S1 to S3 until the next day's forecast results are obtained; S5: Use multiple indicators to evaluate the prediction results and adjust the initial parameters synchronously based on the prediction results.
[0045] Further, in step S1, due to problems such as acquisition system failure and transmission delay, the data in the database contain outliers, which are detected using an autoencoder, specifically including encoder training, reconstruction error calculation, and outlier detection; Preferably, Figure 2 This is a schematic diagram of the autoencoder structure. The principle of outlier detection is as follows: 1) The encoder is used to map the input data to a low-dimensional representation, and the decoder is used to restore the low-dimensional representation to the original data space; 2) The training process is completed with the goal of minimizing the difference between the input data and the reconstructed data, using the mean square error as the loss function; 3) Use the autoencoder to encode and decode the new data and calculate the reconstruction error of each data point; 4) Compare the reconstruction error with the threshold and identify outliers.
[0046] Preferably, in step S1, moving average is used to fill the detected outliers, and the formula is as follows:
[0047] In the formula, o t is the data to be interpolated, r is the time window data size, o i is the known data in the window.
[0048] Furthermore, in step S1, the abnormal data is filled by using a moving average, the load data of the previous day on the day when the abnormal value occurs is extracted, and the average value of the load of the previous day is calculated as the filling value.
[0049] Furthermore, in step S2, the characteristic information of the power load is extracted using the unified information coefficient, including the following sub-steps: S21: The mutual information is calculated as follows:
[0050] In the formula, is mutual information, p ( x , y ) is the joint probability distribution function; p ( x )and p ( y ) are respectively X and Y The marginal probability distribution function of X and Y are two sets of eigenvectors, x and y is the feature vector X and Y The corresponding element.
[0051] S22: The variable division results are as follows:
[0052]
[0053] In the formula, L x and L y They are X and Y The partition unit length of x max and x min Variables X The maximum and minimum values of ; y max and y min Variables YThe maximum and minimum values of ; m and n They are X and Y The number of segments; is the partition grid size; k Indicates the sample size.
[0054] S23: Unified information coefficient calculation:
[0055] In the formula, Represents unified information, m and n They are X and Y The number of segments, min means taking the minimum value.
[0056] Furthermore, in step S3, the training set is used for the model to learn the patterns and features of the data and adjust the internal parameters of the model; the validation set is used to prevent the model from overfitting; and the test set is used to finally evaluate the performance of the model and verify the model's generalization ability on new and unseen data.
[0057] Preferably, the hiking optimization method is based on the Tobler hiking function, and determines the walking speed of the hiker by considering the elevation of the terrain and the walking distance, including the following calculation formula: 1) Initialize the hiker positions and evaluate the best hiker position, which is defined as the leader; 2) Calculate the hiker's initial speed based on the terrain slope, including the following calculation formula:
[0058]
[0059] In the formula, dh and dx represent the altitude and horizontal distance covered by the hiker, respectively. is the slope angle of the terrain, is the speed of hiker i at iteration t; is the slope of the terrain, and e is the base of the natural logarithm function.
[0060] 3) Update the hiker speed through the social thinking of the traveler group and the personal cognition of individual travelers, and the calculation formula is:
[0061] In the formula, is a random number in [0,1]; is the speed of traveler i at the tth iteration; It is the position of the leader; is the position of traveler i at the tth iteration; is the scan factor of traveler i, and the scan factor is in [1,2]. The scan factor ensures that travelers do not stray too far from the leader so that they can see the direction of the leader and receive the leader's signal.
[0062] 4) Further, the hiker’s location is updated and the calculation formula is:
[0063] In the formula, For travelers i In the t +1 iteration position; 5) Update the fitness value of the hiker group and make iterative convergence judgment. Once the current number of iterations reaches the maximum number of iterations, the iterative process ends.
[0064] Preferably, the KELM model includes an ELM model and a kernel function, including the following calculation formula: 1) ELM prediction output F ( q ) is expressed as follows:
[0065] In the formula, q is the input vector, H ( x ) is the hidden node output matrix, represents the output weight; 2) Introducing the regularization coefficient and the unit matrix, transforming the solution process, the least squares solution of the output weight is:
[0066] In the formula, G is the target vector, for The least squares solution of I is the identity matrix, C represents the regularization coefficient.
[0067] 3) Change the random mapping to kernel mapping, and the kernel matrix expression is as follows:
[0068] In the formula, For nuclear elements, is the kernel function, q i and q j Represents the experimental input vector.
[0069] 4) The kernel function selects the Gaussian kernel function, the formula is as follows:
[0070] In the formula, is the kernel parameter, q i and q j represents the experimental input vector, exp is a natural exponential function.
[0071] 5) The final output calculation formula is:
[0072] In the formula, n is the sample size, I represents the identity matrix, C represents the regularization coefficient, K is the kernel function, For nuclear elements, T is the transposition symbol, G is the target vector, for KELM Output.
[0073] Further, in step S4, the time range of the input data set is d Day d + N Daily load data, forecast target is d + N +1 day load data. As the data set continues to grow, the time scale of the input data set is updated using a rolling strategy. d+ 1st to 2nd d + N+ 1 day load data, the forecast target is changed accordingly to d + N+ 2-day load data, in order to balance the model training time and maintain the generalization of the model, including the following representation:
[0074] In the formula, d For one day, including multiple sampling times, N is the sample size, Y d , Y d+1 , Y d+2 , Y d+N , Y d+N+1 Respectively expressed as d , No. d +1, No.d +2, No. d + N and d + N +1 day's load.
[0075] Furthermore, in step S5, a multi-index evaluation model is used to predict the results, including root mean square error, explained variance score and mean absolute percentage error, including the following calculation formula: 1) Root mean square error:
[0076] In the formula, is the root mean square error, N is the sample size; is the true value; is the predicted value.
[0077] 2) Explainable variance score
[0078] In the formula, Explained variance score, var is the variance.
[0079] 3) Mean absolute percentage error
[0080] Where: is the mean absolute percentage error, N represents the sample size; is the true value; is the predicted value.
[0081] Preferably, the root mean square error is used to measure the degree of deviation between the predicted value and the true value, and its value range is The smaller the value, the smaller the prediction error of the model, and the stronger the prediction ability of the model; the explained variance score ranges from 0 to 1. When EVS is 1, it means that the model perfectly predicts the data; conversely, it means that the model explains the data less. The range of mean absolute percentage error is , MAPE greater than 100% indicates an inferior model, otherwise, it indicates that the model has a higher prediction accuracy.
[0082] like Figure 3As shown in the figure, different models can achieve daily load forecasting, the input data remain the same, and the prediction time scale is 24 sampling points, that is, the daily load demand. Due to the problems of gradient disappearance and gradient explosion, the prediction error of RNN is large; LSTM and GRU are further optimized on the basis of RNN, and the load forecasting effect is effectively improved by improving the gate structure; the kernel extreme learning machine introduces Gaussian function on the basis of ordinary extreme learning machine, improves the nonlinear mapping ability, and has significant advantages in processing nonlinear complex relationships; the prediction result after KELM optimization by HOA is closest to the true value, and the load prediction accuracy is the highest, indicating that effective parameter selection plays an important role in improving the load forecasting effect, and further illustrates the necessity of parameter selection. As shown in Table 1, compared with the five single prediction models of RNN, LSTM, GRU, ELM and KELM, its RMSE is relatively reduced by 36.25%, 35.61%, 30.57%, 40.24% and 28.34%. The smaller the value, the higher the model accuracy; MAPE is relatively reduced by 33.95%, 32.19%, 21.05%, 39.09% and 18.26%. Similar to RMSE, the smaller the value, the better; EVS is relatively improved by 0.45%, 0.89%, 1.09%, 1.09% and 0.49%. Different from the above two indicators, the larger the value, the more data variance the model explains.
[0083] The embodiments of the present invention use an autoencoder to detect abnormal values of load data, a KELM model to build a load forecasting model, a walking optimization algorithm to optimize model parameter selection, a rolling strategy to update input data sets, and multiple indicators to evaluate model performance, so as to obtain online prediction results of power load, thereby producing a technical effect of one plus one being greater than two.
[0084] Table 1 Power load prediction performance of each model
[0085] Embodiment 2: The present invention discloses an online prediction system for daily power load based on multi-model fusion, such as Figure 4 As shown, including: Data detection and filling module: Use autoencoder to detect outliers in historical load data, give outlier data and location, interpolate outlier data through moving average, and obtain the initial data set; Feature extraction module: calculates the unified information coefficient between the feature variables and the load data in the initial data set, and selects the features and load data with high correlation to construct the input data set; Model training module: divide the input data set into training set, test set and validation set according to a fixed ratio, train the KELM model, and use the walking optimization algorithm to optimize the parameters to obtain the best model parameters; Daily load forecasting module: Based on the optimal model parameters, input the characteristic data of the day to be predicted to obtain the load data of the day to be predicted; Data update module: Uses a rolling strategy to update the input data set, re-executes data detection, data filling, feature extraction, model training and prediction processes, and obtains the load forecast results for the next day; Model update module: multiple indicators are used to evaluate the prediction performance of load forecasting results, and the initial parameter settings are adjusted according to the evaluation results.
[0086] It is particularly noted that the above examples are only some of the better examples of the present invention. The present invention is described in detail with reference to the above embodiments. Those skilled in the art should understand that the specific implementation modes of the present invention may be modified or replaced by equivalents, and any modification or equivalent replacement that does not depart from the spirit and scope of the present invention shall be included in the scope of protection of the claims of the present invention.
Claims
1. A multi-model fusion power daily load online prediction method, characterized in that: The following steps are involved: S1: Use the autoencoder to detect outliers in the collected historical daily load data, and fill in the outliers through moving average to obtain the initial data set D1; S2: extracting characteristic information of power load in the initial data set D1 using the unified information coefficient to obtain the input data set D2; S3: Divide the input data set D2 into a training set, a test set, and a validation set according to a fixed ratio, use the KELM model to build a load forecasting model, use the training set and the walking optimization algorithm to select the KELM model parameters, and realize the single-day load forecasting through the KELM model; S4: Keep the sample size unchanged, use the rolling strategy to update the KELM after parameter optimization to obtain the load forecast value, and repeat steps S1 to S3 until the forecast result for the next day is obtained; S5: Use multiple indicators to evaluate the prediction results and adjust the initial parameters synchronously based on the prediction results.
2. According to the multi-model fusion online prediction method for daily power load in claim 1, in step S1, the autoencoder performs outlier detection on the collected historical daily load data, including encoder training, reconstruction error calculation and outlier detection.
3. According to the method for online prediction of daily power load by multi-model fusion in claim 1, in step S2, the method of extracting characteristic information of power load by using unified information coefficient comprises the following sub-steps: S21: Mutual Information Calculation: In the formula, represents mutual information, p ( x , y ) is the joint probability distribution function; p ( x )and p ( y ) are respectively X and Y The marginal probability distribution function of X and Y are two sets of eigenvectors, x is the feature vector X The corresponding elements, y is the feature vector Y The corresponding element; S22: Variable division: In the formula, L x and L y They are X and Y The partition unit length of x max and x min Variables X The maximum and minimum values of ; y max and ymin Variables Y The maximum and minimum values of ; m and n They are X and Y The number of segments; is the partition grid size; k is the sample size; S23: Unified information coefficient calculation: In the formula, Represents unified information, m and n They are X and Y The number of segments, min Indicates taking the minimum value.
4. According to the multi-model fusion online prediction method for daily power load of claim 1, in step S3, it is characterized in that: The walking optimization algorithm includes: 1) Initialize the hiker positions and evaluate the best hiker position, which is defined as the leader; 2) Calculate the hiker's initial speed based on the terrain slope, including the following calculation formula: In the formula, dh and dx represent the altitude and horizontal distance covered by the hiker, respectively. is the slope angle of the terrain, Is a hiker i In the t The speed of iterations; is the slope of the terrain, e is the base of the natural logarithm function; 3) Update hiker speed through the social thinking of the traveler group and the personal cognition of individual travelers, as follows: In the formula, is a random number in [0,1]; For travelers i In the t The speed of iterations; It is the position of the leader; For travelers i In the t The position of the iteration; Is a traveler i The scanning factor SF is in [1,2]. SF ensures that the traveler does not stray too far from the leader and receives the leader's signal. 4) Further, the hiker's location is updated as follows: In the formula, For travelers i In the t +1 iteration position; 5) Update the fitness value of the hiker group and make an iterative convergence judgment. When the current number of iterations reaches the maximum number of iterations, the iterative process ends.
5. According to the multi-model fusion online prediction method for daily power load of claim 1, in step S3, it is characterized in that: The KELM model includes an ELM model and a kernel function, including: 1) ELM prediction output F ( q ) is expressed as follows: In the formula, q is the input vector, H is the hidden node output matrix, is the output weight; is the output function; 2) Introducing the regularization coefficient and the unit matrix, transforming the solution process, the least squares solution of the output weight is: In the formula, G is the target vector, for The least squares solution of I is the unit matrix, C is the regularization coefficient; 3) Change the random mapping to kernel mapping, and the kernel matrix expression is as follows: In the formula, For nuclear elements, is the kernel function, q i and q j represents the experimental input vector; 4) The kernel function selects the Gaussian kernel function, the formula is: In the formula, is the kernel parameter, q i and q j represents the experimental input vector, exp is the natural exponential function; 5) The final output is: In the formula, n is the sample size, I is the identity matrix, C is the regularization coefficient, K is the kernel function, For nuclear elements, T is the transposition symbol, G is the target vector, for KELM Output.
6. According to the multi-model fusion online prediction method for daily power load of claim 1, in step S4, it is characterized in that: The method to obtain the next day's forecast results is: the original sample size is N , time range d ~ d + N , the rolling strategy is used to update the feature data set D2. The specific process is to remove the time in D2 d Load data for the day, add time at the end of D2 for the first d + N +1 day's load data, the updated calculation formula is: In the formula, d For one day, including multiple sampling times, N is the sample size, Y d , Y d+1 , Y d+2 , Y d+N , Y d+N+1 Respectively expressed as d , No. d +1, No. d +2, No. d + N and d + N +1 day's load.
7. The method for online prediction of daily power load by multi-model fusion according to claim 1, wherein in step S5, The load forecasting result of step S4 is used to evaluate the forecasting accuracy of the model, specifically including: 1) Root mean square error: In the formula, is the root mean square error, N is the sample size; is the true value; is the predicted value; 2) Explainable variance score In the formula, Explained variance score, var is the variance; 3) Mean absolute percentage error Where: is the mean absolute percentage error, N is the sample size; is the true value; is the predicted value.
8. A multi-model fusion power daily load online prediction system, characterized in that: include: Data detection and filling module: Use autoencoder to detect outliers in historical load data, give outlier data and location, interpolate outlier data through moving average, and obtain the initial data set; Feature extraction module: calculates the unified information coefficient between the feature variables and the load data in the initial data set, and selects the features and load data with high correlation to construct the input data set; Model training module: divide the input data set into training set, test set and validation set according to a fixed ratio, train the KELM model, and use the walking optimization algorithm to optimize the parameters to obtain the best model parameters; Daily load forecasting module: Based on the optimal model parameters, input the characteristic data of the day to be predicted to obtain the load data of the day to be predicted; Data update module: Uses a rolling strategy to update the input data set, re-executes data detection, data filling, feature extraction, model training and prediction processes, and obtains the load forecast results for the next day; Model update module: multiple indicators are used to evaluate the prediction performance of load forecasting results, and the initial parameter settings are adjusted according to the evaluation results.
Citation Information
Patent Citations
Short-term load prediction method based on variant selection and Gaussian process regression
CN106971240A
Method and system of predicting electric system load based on wavelet noise reduction and EMD-arima
US20220045509A1
Prediction method for charging loads of electric vehicles with consideration of data correlation
US20230074700A1
AU2020104000A4
Cited By
Building cold load cycle prediction method based on boundary feature protection and HOA-LightGBM model
CN120781493A
Hiking time dynamic prediction method for emergency rescue in mountainous area
CN120849777A
