Short-term load prediction method and system based on similar day and combination model
By using a method based on similar days and combined models, and employing the Pearson correlation coefficient method, improved variational mode decomposition, and least squares support vector machine model, the accuracy and adaptability issues of short-term load forecasting are solved, achieving more efficient power load forecasting.
Patent Information
- Application Number
- CN202410092491.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-01-23
- Publication Date
- 2026-02-13
AI Technical Summary
Existing short-term load forecasting methods suffer from low accuracy, low adaptability, and low reliability. In particular, they are unable to effectively capture the meteorological influencing factors of diverse loads when faced with complex power systems, weather changes, and market demands.
A method based on similar days and combination models is adopted, and the meteorological influencing factors of multivariate loads are determined by the Pearson correlation coefficient method. Historical datasets are constructed, and the intrinsic mode function components are predicted by an improved variational mode decomposition and an improved least squares support vector machine model, outputting the optimal short-term load prediction results.
It improves the accuracy and reliability of short-term load forecasting, enhances the adaptability of the model, and is better able to cope with changes in different regions, seasons and market environments, providing more accurate and reliable forecast results.
Smart Images
Figure CN121525907A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of power technology, specifically to a short-term load forecasting method and system based on a similar day and combination model. Background Technology
[0002] Short-term load forecasting has always been a key issue in power system management and energy planning. With the continuous growth of electricity demand and the increasing complexity of power systems, accurate short-term load forecasting has become increasingly important. In the power industry, there are many traditional methods for short-term load forecasting, including time series analysis, regression analysis and statistical models. These methods are usually based on historical data and some conventional characteristics, such as time, date and seasonality, but they perform poorly when dealing with complex changing patterns and external factors.
[0003] With the advancement of technology, some progress has been made in the field of short-term load forecasting, but challenges still exist. Traditional forecasting methods are usually based on historical data and statistical models, which perform poorly when faced with complex power systems, weather changes, and market demands. In recent years, the application of machine learning and data science has begun to emerge in short-term load forecasting, providing new opportunities to improve accuracy.
[0004] Traditional methods often fail to effectively capture the complex changing patterns of power load, resulting in low accuracy of predictions. This makes it difficult for power systems to cope with sudden events and demand fluctuations. Existing methods typically rely on specific modeling assumptions and parameter settings, making it difficult to adapt to changes in different regions, seasons, and market environments. This limits their flexibility and adaptability in practical applications. Because traditional methods have high requirements for data quality and availability, they are susceptible to interference from noise, missing data, or anomalous data, which reduces the reliability of prediction models. Summary of the Invention
[0005] In view of the above-mentioned problems, the present invention is proposed.
[0006] Therefore, the technical problem solved by this invention is that existing short-term load forecasting methods suffer from low accuracy, low adaptability, and low reliability, as well as the problem of how to identify the meteorological influencing factors of multiple loads.
[0007] To address the aforementioned technical problems, this invention provides the following technical solution: a short-term load forecasting method based on similar days and a combined model, comprising determining the meteorological influencing factors of multivariate loads using the Pearson correlation coefficient method, selecting similar days for meteorological factors, load date types, and pricing mechanisms, constructing a historical dataset, decomposing the historical dataset into intrinsic mode function components at different frequencies using an improved variational mode decomposition method, predicting the intrinsic mode function components using an improved least squares support vector machine model, and outputting the optimal short-term load forecasting result.
[0008] As a preferred embodiment of the short-term load forecasting method based on a similar day and combination model described in this invention, the determination of meteorological influencing factors of multivariate loads includes determining the meteorological influencing factors of multivariate loads using the Pearson correlation coefficient method.
[0009] Data was collected on meteorological factors affecting multiple loads, including temperature, humidity, and wind speed data. NaN and blank values were removed from the temperature, humidity, and wind speed data.
[0010] The significant impact of meteorological factors on power load was determined using the Pearson correlation coefficient method. The temperature correlation coefficient was calculated and expressed as follows:
[0011]
[0012] Where r1 is the temperature correlation coefficient, T o The temperature value at time point 0. For the average temperature, L o Let be the power load value at time point 0. This represents the average power load.
[0013] The humidity correlation coefficient is calculated as follows:
[0014]
[0015] Where r2 is the humidity correlation coefficient, H o The humidity value at time point 0. This represents the average humidity level.
[0016] The wind speed correlation coefficient is expressed as follows:
[0017]
[0018] Where r3 is the wind speed correlation coefficient, W o Let be the wind speed value at time point 0. This represents the average wind speed.
[0019] By statistically analyzing temperature, humidity, and wind speed, thresholds θ1, θ2, and θ3 were set to indicate the significant impact of temperature, humidity, and wind speed.
[0020] The correlation coefficients for temperature, humidity, and wind speed were evaluated separately, and the significant impact of each meteorological factor was calculated and expressed as follows:
[0021]
[0022] Among them, S n Let r be the significance value of the nth meteorological factor. n Let θ be the nth correlation coefficient. n γ is the threshold for the nth significant effect. n To adjust the parameters of the curve.
[0023] When the significance value of the temperature correlation coefficient assessment is greater than 0.5, the temperature factor data in the meteorological factors are retained; when the significance value of the temperature correlation coefficient assessment is less than or equal to 0.5, the temperature factor data in the meteorological factors are not retained.
[0024] When the significance value of the humidity correlation coefficient assessment is greater than 0.5, the humidity factor data in the meteorological factors is retained; when the significance value of the humidity correlation coefficient assessment is less than or equal to 0.5, the humidity factor data in the meteorological factors is not retained.
[0025] When the significance value of the wind speed correlation coefficient assessment is greater than 0.5, the wind speed factor data in the meteorological factors is retained; when the significance value of the wind speed correlation coefficient assessment is less than or equal to 0.5, the wind speed factor data in the meteorological factors is not retained.
[0026] As a preferred embodiment of the short-term load forecasting method based on similar days and combination models described in this invention, the construction of the historical dataset includes selecting similar days based on meteorological factors, load date types, and price mechanisms to construct the historical dataset.
[0027] Load date types include weekdays, weekends, and public holidays.
[0028] The overall similarity between each day and the predicted day is calculated as follows:
[0029] S(D,d)=(w wt ·f wt (D,d)+w ld ·f ld (D,d)+w price ·f price (D,d))·P(D,d)
[0030] Where S(D,d) is the comprehensive similarity between the predicted day D and the historical day d, f wt (D,d) is the similarity function of meteorological factors, f ld (D,d) is a complex similarity function for the date type, f price (D,d) is the price mechanism similarity function, w wt As the weight of meteorological factors, w ld For the load date type weight, w price P(D,d) represents the weight of the price mechanism and P(D,d) represents the penalty term.
[0031] The similarity function of meteorological factors is expressed as:
[0032]
[0033] Among them, THW c Let σTHW be the data value of the c-th meteorological factor. c Let be the standard deviation of the c-th meteorological factor data.
[0034] dynamically adjust w based on historical data wt w ld and w price The obtained S(D,d) values are selected. If there are jt extreme weather or special events predicted on a given day, and jt equals 0 and the meteorological factor S(D,d) value is greater than or equal to 0.6, then historical days d and predicted days D with S(D,d) values greater than or equal to 0.6 are selected as similar days. If jt equals 0 and the meteorological factor S(D,d) value is less than 0.6, data including sudden events and natural disasters in the collected meteorological factor data are deleted, and the significant impact of meteorological factors on power load is recalculated using the Pearson correlation coefficient method. The comprehensive similarity value between the obtained predicted day D and historical day d is recalculated as S(D,d)2. If the S(D,d)2 value is greater than or equal to 0.6, then historical days d and predicted days D with S(D,d)2 values greater than or equal to 0.6 are selected as similar days. If the S(D,d)2 value is less than 0.6, then historical days d and predicted days D with S(D,d)2 values less than 0.6 are not selected as similar days, and the selection of similar days is repeated.
[0035] When jt is greater than or equal to 1 and the meteorological factor S(D,d) value is greater than or equal to 0.75, historical days d and predicted days D with S(D,d) values greater than or equal to 0.75 are selected as similar days. When jt is greater than or equal to 1 and the meteorological factor S(D,d) value is less than 0.75, the collected meteorological factor data is expanded by deleting data that does not include sudden events and natural disasters. The comprehensive similarity value between the predicted day D and the historical day d is recalculated as S(D,d)3. When the S(D,d)3 value is greater than or equal to 0.75, historical days d and predicted days D with S(D,d)3 are selected as similar days. When the S(D,d)3 value is less than 0.75, historical days d and predicted days D with S(D,d)3 values less than 0.75 are not selected as similar days, and the next similar day with predicted day D is selected.
[0036] The selected similar day data are used to construct a historical dataset.
[0037] As a preferred embodiment of the short-term load forecasting method based on similar days and combination models described in this invention, the improved variational mode decomposition includes an adaptive frequency bandwidth selection mechanism based on data characteristics.
[0038] The adaptive frequency bandwidth selection mechanism dynamically adjusts the frequency bandwidth as follows:
[0039] BW=β(Var(d),Skew(d),Kurt(d))·SD(f)+δ·Entropy(f)
[0040] Where BW is the adaptive frequency bandwidth, Var(d) is the variance of historical day d, Skew(d) is the skewness of historical day d, Kurt(d) is the kurtosis of historical day d, Entropy(f) is the information entropy of frequency f, β is the composite function of variance, skewness and kurtosis adjustment, δ is the adjustment parameter, and SD(f) is the standard deviation of frequency sequence f.
[0041] The composite function β is expressed as:
[0042] β(Var(d),Skew(d),Kurt(d))
[0043] =α·exp(-γ1·Var(d)-γ2·|Skew(d)|-γ3·Kurt(d))
[0044] Among them, α, γ1, γ2 and γ3 are set parameters.
[0045] As a preferred embodiment of the short-term load forecasting method based on similar days and combination models described in this invention, the intrinsic mode function components include improved variational mode decomposition, which decomposes the historical dataset into intrinsic mode function components at different frequencies.
[0046] The characteristics of the historical dataset include variance Var(d), skewness Skew(d), kurtosis Kurt(d), and information entropy Entropy(f).
[0047] The frequency bandwidth BW is dynamically calculated using an improved variational mode decomposition method on the historical dataset, and is expressed as:
[0048]
[0049] Where V represents a set of IMF components obtained from variational mode decomposition, u k (t) represents each IMF component in V, {u k} represents the IMF component, {ω kLet} be the center frequency of the IMF component, δ(t) be the Dirac delta function, Re{·} be the real part operation, Hilbert{·} be the Hilbert transform, K be the number of IMF components, g(BW) be the function for adaptive frequency bandwidth BW adjustment, and exp(-jω) be the frequency response of the IMF component. k t) is a complex exponential function. This is the derivative with respect to time.
[0050] As a preferred embodiment of the short-term load forecasting method based on a similar day and combination model described in this invention, the improved least squares support vector machine model includes the introduction of an adaptive learning rate.
[0051] During model training, the learning rate is adaptively adjusted based on the prediction error. The improved least squares support vector machine model is represented as follows:
[0052]
[0053] Where J(w,b,cx; α(t)) is the objective function of the improved least squares support vector machine model, w is the model weight, b is the bias term, cx is the error term, α(t) is the adaptive learning rate, t is the number of iterations, and C(t) is the regularization parameter adjusted with the number of iterations. This is the sum of squared errors for all training samples.
[0054] The constraints of the improved least squares support vector machine model are expressed as follows:
[0055] y z =w·φ(Combine(F1,F2,…,F K ))+b+cx i
[0056] y z Let φ be the target value of the z-th training sample, and let φ be the kernel function mapping. Combine(F1,F2,…,F…) K ) is a combination function that combines the features F1, F2, ..., F of V. K Combined to form a single predictor variable, F k cx is the feature set extracted from the k-th IMF component. i Let be the error term for the i-th sample.
[0057] As a preferred embodiment of the short-term load forecasting method based on similar days and combination models described in this invention, the optimal short-term load forecasting result includes using an improved least squares support vector machine model to predict the intrinsic mode function components and output the optimal short-term load forecasting result.
[0058] Extract time-domain and spectral features from V, and integrate the time-domain and spectral features into a comprehensive feature set F. k .
[0059] The intrinsic mode function components V are input to the improved least squares support vector machine model, and the adaptive learning mechanism and the combination function Combine(F1,F2,…,F) are used to achieve this. K Train the improved least squares support vector machine model.
[0060] Inputting the historical IMF components of similar days to the predicted day into the trained improved least squares support vector machine model, the improved least squares support vector machine model outputs the short-term load forecast result, which is expressed as follows:
[0061]
[0062] in, This is the predicted result.
[0063] The output short-term load forecast results This is considered the optimal short-term load forecast result.
[0064] Another objective of this invention is to provide a short-term load forecasting system based on a similarity day and combination model, which can predict the intrinsic mode function components by using an improved least squares support vector machine model and output the optimal short-term load forecasting results, thus solving the problem of low reliability in current short-term load forecasting methods.
[0065] As a preferred embodiment of the short-term load forecasting system based on similar days and combination models described in this invention, the system includes: a historical dataset module, an intrinsic mode function component module, and an optimal short-term load forecasting module. The historical dataset module is used to determine the meteorological influencing factors of multivariate loads using the Pearson correlation coefficient method, select similar days for meteorological factors, load date types, and pricing mechanisms, and construct a historical dataset. The intrinsic mode function component module is used to decompose the historical dataset into intrinsic mode function components at different frequencies using an improved variational mode decomposition method. The optimal short-term load forecasting module is used to predict the intrinsic mode function components using an improved least squares support vector machine model and output the optimal short-term load forecasting result.
[0066] A computer device includes a memory and a processor, the memory storing a computer program, the processor executing the computer program to implement a short-term load forecasting method based on a similar day and combination model.
[0067] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of a short-term load forecasting method based on a similar day and combination model.
[0068] The beneficial effects of this invention are as follows: This invention provides a short-term load forecasting method based on similar days and a combined model. It determines the meteorological influencing factors of multivariate loads using the Pearson correlation coefficient method, selects similar days for meteorological factors, load date types, and pricing mechanisms, constructs a historical dataset, identifies and determines the meteorological influencing factors in multivariate loads, and filters similar days to provide key input features for subsequent load forecasting models. By filtering and matching similar meteorological and load data, the accuracy of the forecasting model is improved. An improved variational mode decomposition method is used to decompose the historical dataset into intrinsic mode function components at different frequencies, enhancing the feature extraction capability of the historical dataset and providing more representative and information-rich inputs for subsequent models, thus providing more adaptive data for model construction. An improved least squares support vector machine model is used to predict the intrinsic mode function components, outputting the optimal short-term load forecasting results. This provides a more adaptive forecasting model, making short-term load forecasting more accurate and reliable, improving the accuracy and efficiency of short-term load forecasting, and providing trustworthy forecasting results for practical applications. This invention achieves better results in terms of accuracy, adaptability, and reliability. Attached Figure Description
[0069] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0070] Figure 1 The first embodiment of the present invention provides an overall flowchart of a short-term load forecasting method based on a similar day and combination model.
[0071] Figure 2 The following is an overall flowchart of a short-term load forecasting system based on a similar day and combination model, provided for the third embodiment of the present invention. Detailed Implementation
[0072] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of the present invention.
[0073] Example 1
[0074] Reference Figure 1 As an embodiment of the present invention, a short-term load forecasting method based on a similar day and combination model is provided, comprising:
[0075] S1: The meteorological influencing factors of multivariate loads are determined by the Pearson correlation coefficient method. Similar days are selected for meteorological factors, load date types and price mechanisms to construct a historical dataset.
[0076] Furthermore, the meteorological influencing factors of multivariate loads were determined using the Pearson correlation coefficient method.
[0077] Data was collected on meteorological factors affecting multiple loads, including temperature, humidity, and wind speed data. NaN and blank values were removed from the temperature, humidity, and wind speed data.
[0078] The significant impact of meteorological factors on power load was determined using the Pearson correlation coefficient method. The temperature correlation coefficient was calculated and expressed as follows:
[0079]
[0080] Where r1 is the temperature correlation coefficient, T o The temperature value at time point 0. For the average temperature, L o Let be the power load value at time point 0. This represents the average power load.
[0081] The humidity correlation coefficient is calculated as follows:
[0082]
[0083] Where r2 is the humidity correlation coefficient, H o The humidity value at time point 0. This represents the average humidity.
[0084] The wind speed correlation coefficient is expressed as follows:
[0085]
[0086] Where r3 is the wind speed correlation coefficient, W o Let be the wind speed value at time point 0. This represents the average wind speed.
[0087] By statistically analyzing temperature, humidity, and wind speed, thresholds θ1, θ2, and θ3 were set to indicate the significant impact of temperature, humidity, and wind speed.
[0088] The correlation coefficients for temperature, humidity, and wind speed were evaluated separately, and the significant impact of each meteorological factor was calculated and expressed as follows:
[0089]
[0090] Among them, S n Let r be the significance value of the nth meteorological factor. n Let θ be the nth correlation coefficient. n γ is the threshold for the nth significant effect. n To adjust the parameters of the curve.
[0091] When the significance value of the temperature correlation coefficient assessment is greater than 0.5, the temperature factor data in the meteorological factors are retained; when the significance value of the temperature correlation coefficient assessment is less than or equal to 0.5, the temperature factor data in the meteorological factors are not retained.
[0092] When the significance value of the humidity correlation coefficient assessment is greater than 0.5, the humidity factor data in the meteorological factors is retained; when the significance value of the humidity correlation coefficient assessment is less than or equal to 0.5, the humidity factor data in the meteorological factors is not retained.
[0093] When the significance value of the wind speed correlation coefficient assessment is greater than 0.5, the wind speed factor data in the meteorological factors is retained; when the significance value of the wind speed correlation coefficient assessment is less than or equal to 0.5, the wind speed factor data in the meteorological factors is not retained.
[0094] It should be noted that r1, r2, and r3 are the temperature correlation coefficient, humidity correlation coefficient, and wind speed correlation coefficient, respectively, and their values range from [-1, 1]. o H o and W o These represent the temperature, humidity, and wind speed values at time point 0. as well as These represent the overall average temperature, overall average humidity, and overall average wind speed, respectively. o and Let r1, r2, and r3 be the power load value and the average power load at time point 0, respectively. Their range depends on the actual power load data. r1, r2, and r3 reflect the degree of linear correlation between each meteorological factor and the power load. θ n The value range of γ is [0,1]. n The parameter S is used to adjust the steepness of the curve. nThe value range is [0,1].
[0095] It should also be noted that similar days were selected based on meteorological factors, load date types, and pricing mechanisms to construct a historical dataset.
[0096] Load date types include weekdays, weekends, and public holidays.
[0097] The overall similarity between each day and the predicted day is calculated as follows:
[0098] S(D,d)=(w wt ·f wt (D,d)+w ld ·f ld (D,d)+w price ·f price (D,d))·P(D,d)
[0099] Where S(D,d) is the comprehensive similarity between the predicted day D and the historical day d, f wt (D,d) is the similarity function of meteorological factors, f ld (D,d) is a complex similarity function for the date type, f price (D,d) is the price mechanism similarity function, w wt As the weight of meteorological factors, w ld For the load date type weight, w price P(D,d) represents the weight of the price mechanism and P(D,d) represents the penalty term.
[0100] The similarity function of meteorological factors is expressed as:
[0101]
[0102] Among them, THW c Let σTHW be the data value of the c-th meteorological factor. c Let be the standard deviation of the c-th meteorological factor data.
[0103] dynamically adjust w based on historical data wt w ld and w priceThe obtained S(D,d) values are selected. If there are jt extreme weather or special events predicted on a given day, and jt equals 0 and the meteorological factor S(D,d) value is greater than or equal to 0.6, then historical days d and predicted days D with S(D,d) values greater than or equal to 0.6 are selected as similar days. If jt equals 0 and the meteorological factor S(D,d) value is less than 0.6, data including sudden events and natural disasters in the collected meteorological factor data are deleted, and the significant impact of meteorological factors on power load is recalculated using the Pearson correlation coefficient method. The comprehensive similarity value between the obtained predicted day D and historical day d is recalculated as S(D,d)2. If the S(D,d)2 value is greater than or equal to 0.6, then historical days d and predicted days D with S(D,d)2 values greater than or equal to 0.6 are selected as similar days. If the S(D,d)2 value is less than 0.6, then historical days d and predicted days D with S(D,d)2 values less than 0.6 are not selected as similar days, and the selection of similar days is repeated.
[0104] When jt is greater than or equal to 1 and the meteorological factor S(D,d) value is greater than or equal to 0.75, historical days d and predicted days D with S(D,d) values greater than or equal to 0.75 are selected as similar days. When jt is greater than or equal to 1 and the meteorological factor S(D,d) value is less than 0.75, the collected meteorological factor data is expanded by deleting data that does not include sudden events and natural disasters. The comprehensive similarity value between the predicted day D and the historical day d is recalculated as S(D,d)3. When the S(D,d)3 value is greater than or equal to 0.75, historical days d and predicted days D with S(D,d)3 are selected as similar days. When the S(D,d)3 value is less than 0.75, historical days d and predicted days D with S(D,d)3 values less than 0.75 are not selected as similar days, and the next similar day with predicted day D is selected.
[0105] The selected similar day data are used to construct a historical dataset.
[0106] It should also be noted that jt represents the total number of extreme weather and special events on prediction day D, collecting relevant power load data and pricing mechanism information. P(D,d) is a penalty term used to handle extreme weather or special events. In the case of extreme weather or special events, the similarity calculation requires a more stringent standard. For example, 0.75 means that only under very similar conditions will historical days be selected as training data for the prediction model. The adjustment of weights can ensure that the model more accurately reflects the impact of special weather conditions or events on power load. The collected meteorological factor data is expanded to a wider time range than the previous statistics, thus expanding the collected data.
[0107] S2: Improved variational mode decomposition is used to decompose the historical dataset into intrinsic mode function components at different frequencies.
[0108] Furthermore, the improved variational mode decomposition includes the use of an adaptive frequency bandwidth selection mechanism based on data characteristics.
[0109] The adaptive frequency bandwidth selection mechanism dynamically adjusts the frequency bandwidth as follows:
[0110] BW=β(Var(d),Skew(d),Kurt(d))·SD(f)+δ·Entropy(f)
[0111] Where BW is the adaptive frequency bandwidth, Var(d) is the variance of historical day d, Skew(d) is the skewness of historical day d, Kurt(d) is the kurtosis of historical day d, Entropy(f) is the information entropy of frequency f, β is the composite function of variance, skewness and kurtosis adjustment, δ is the adjustment parameter, and SD(f) is the standard deviation of frequency sequence f.
[0112] The composite function β is expressed as:
[0113] β(Var(d),Skew(d),Kurt(d))
[0114] =α·exp(-γ1·Var(d)-γ2·|Skew(d)|-γ3·Kurt(d))
[0115] Among them, α, γ1, γ2 and γ3 are set parameters.
[0116] It should be noted that this mechanism will enable the improved variational mode decomposition to adjust the frequency bandwidth in the variational mode decomposition more precisely according to the characteristics of the data. The improved mechanism will consider multiple statistical characteristics of the data, thereby effectively handling nonlinearity and nonstationarity. δ is an adjustment parameter used to control the influence of information entropy on bandwidth. α, γ1, γ2, and γ3 are pre-set parameters that determine the sensitivity of bandwidth to data characteristics.
[0117] It should also be noted that an improved variational mode decomposition is used to decompose the historical dataset into intrinsic mode function components at different frequencies.
[0118] The characteristics of the historical dataset include variance Var(d), skewness Skew(d), kurtosis Kurt(d), and information entropy Entropy(f).
[0119] The frequency bandwidth BW is dynamically calculated using an improved variational mode decomposition method on the historical dataset, and is expressed as:
[0120]
[0121] Where V represents a set of IMF components obtained from variational mode decomposition, uk (t) represents each IMF component in V, {u k} represents the IMF component, {ω k Let} be the center frequency of the IMF component, δ(t) be the Dirac delta function, Re{·} be the real part operation, Hilbert{·} be the Hilbert transform, K be the number of IMF components, g(BW) be the function for adaptive frequency bandwidth BW adjustment, and exp(-jω) be the frequency response of the IMF component. k t) is a complex exponential function. This is the derivative with respect to time.
[0122] It should also be noted that IMF refers to intrinsic mode function components. The goal of the optimization process is to find a set of IMF components {u k} and their respective center frequencies {ω k The process of finding the optimal solution involves minimizing the value of the entire formula. The optimal solution refers to a series of IMF components and their corresponding center frequencies. These components and frequencies together constitute the output of the VMD method. Each IMF component u k (t) all correspond to a specific center frequency ω. k Hilbert{·} is the Hilbert transform, used to extract analytic signals from IMF components; δ(t) is the Dirac delta function, used to center IMF components in the frequency domain; Re{·} is the real part operation, which refers to the process of extracting the real part from a complex number. The derivative with respect to time is used to assess the rate of change of the IMF components.
[0123] S3: An improved least squares support vector machine model is used to predict the intrinsic mode function components and output the optimal short-term load forecast results.
[0124] Furthermore, the improved least squares support vector machine model includes the introduction of an adaptive learning rate.
[0125] During model training, the learning rate is adaptively adjusted based on the prediction error. The improved least squares support vector machine model is represented as follows:
[0126]
[0127] Where J(w,b,cx; α(t)) is the objective function of the improved least squares support vector machine model, w is the model weight, b is the bias term, cx is the error term, α(t) is the adaptive learning rate, t is the number of iterations, and C(t) is the regularization parameter adjusted with the number of iterations. This is the sum of squared errors for all training samples.
[0128] The constraints of the improved least squares support vector machine model are expressed as follows:
[0129] y z =w·φ(Combine(F1,F2,…,F K ))+b+cx i
[0130] y z Let φ be the target value of the z-th training sample, and let φ be the kernel function mapping. Combine(F1,F2,…,F…) K ) is a combination function that combines the features F1, F2, ..., F of V. K Combined to form a single predictor variable, F k cx is the feature set extracted from the k-th IMF component. i Let be the error term for the i-th sample.
[0131] It should be noted that C(t) is a regularization parameter that adjusts with the number of iterations. Combined with the adaptive learning rate, the constraint is part of the optimization problem of the improved least squares support vector machine model. It expresses the constraints of the model and is used to guide the optimization of model parameters during training.
[0132] It should also be noted that an improved least squares support vector machine model is used to predict the intrinsic mode function components, outputting the optimal short-term load forecast results.
[0133] Extract time-domain and spectral features from V, and integrate the time-domain and spectral features into a comprehensive feature set F. k .
[0134] The intrinsic mode function components V are input to the improved least squares support vector machine model, and the adaptive learning mechanism and the combination function Combine(F1,F2,…,F) are used to achieve this. K Train the improved least squares support vector machine model.
[0135] Inputting the historical IMF components of similar days to the predicted day into the trained improved least squares support vector machine model, the improved least squares support vector machine model outputs the short-term load forecast result, which is expressed as follows:
[0136]
[0137] in, This is the predicted result.
[0138] The output short-term load forecast results This is considered the optimal short-term load forecast result.
[0139] It should also be noted that during the training process, the model is trained based on the adaptive learning rate α(t) and the adjusted regularization parameter C(t). The trained model is then used to predict the short-term load. The short-term load prediction result represents the output used to predict the short-term load result, i.e., the calculation formula used for prediction.
[0140] Example 2
[0141] One embodiment of the present invention provides a short-term load forecasting method based on a similar day and combination model. To verify the beneficial effects of the present invention, scientific demonstration is carried out through economic benefit calculations and simulation experiments.
[0142] First, electricity load data and relevant meteorological data, including temperature, humidity, and wind speed, were collected. NaN values and blank values were removed. The correlation coefficients between temperature, humidity, wind speed, and electricity load were calculated using the Pearson correlation coefficient method. These coefficients were significant (greater than 0.5), indicating that meteorological factors have a significant impact on electricity load. Based on the load date type and pricing mechanism, the comprehensive similarity between each day and the forecast day was calculated. Dynamically adjusted historical data and thresholds were used to ensure the selection of similar historical days. The selected similar day data were used to construct a historical dataset. An adaptive frequency bandwidth selection mechanism based on data features was adopted to decompose the historical dataset into intrinsic mode function components at different frequencies. An improved least squares support vector machine model was trained by introducing an adaptive learning rate. The historical intrinsic mode function components of similar days were input to predict the short-term load forecast results.
[0143] Refer to Table 1, experimental data table of similar day multivariate load short-term forecast.
[0144] Table 1 Experimental Data on Short-Term Forecasting of Multiple Loads on Similar Days
[0145]
[0146] Through the analysis and comparison of the above experimental data, the short-term load forecasting model has shown high accuracy in comparison with the actual power load. Compared with the existing technology, the model is more accurate in predicting power load and has excellent adaptability. It can select similar days according to different load date types and pricing mechanisms, thus better adapting to various situations. While solving the problems of the existing technology, the present invention has higher innovation and advantages. Therefore, our invention is inventive.
[0147] Example 3
[0148] Reference Figure 2 As an embodiment of the present invention, a short-term load forecasting system based on a similar day and combination model is provided, including a historical dataset module, an intrinsic mode function component module, and an optimal short-term load forecasting module.
[0149] The historical dataset module is used to determine the meteorological influencing factors of multivariate loads using the Pearson correlation coefficient method. It selects similar days for meteorological factors, load date types, and pricing mechanisms to construct the historical dataset. The intrinsic mode function component module is used to decompose the historical dataset into intrinsic mode function components at different frequencies using an improved variational mode decomposition. The optimal short-term load forecasting module is used to forecast the intrinsic mode function components using an improved least squares support vector machine model and output the optimal short-term load forecasting results.
[0150] If a function is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.
[0151] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-included system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device.
[0152] More specific examples (a non-exhaustive list) of computer-readable media include: electrical connections (electronic devices) having one or more wires, portable computer disk drives (magnetic devices), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Furthermore, computer-readable media can even be paper or other suitable media on which programs can be printed, because programs can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in computer memory.
[0153] It should be understood that various parts of the present invention can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented using software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc. It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.
[0154] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.
Claims
1. A short-term load forecasting method based on a similar day and combination model, characterized in that, include: The meteorological influencing factors of multivariate loads were determined by the Pearson correlation coefficient method. Similar days were selected for meteorological factors, load date types, and pricing mechanisms to construct a historical dataset. An improved variational mode decomposition method is used to decompose the historical dataset into intrinsic mode function components at different frequencies; An improved least squares support vector machine model is used to predict the intrinsic mode function components, and the optimal short-term load forecast results are output.
2. The short-term load forecasting method based on a similar day and combination model as described in claim 1, characterized in that: The determination of meteorological influencing factors of multivariate loads includes determining the meteorological influencing factors of multivariate loads using the Pearson correlation coefficient method; Data was collected on meteorological factors affecting multiple loads, including temperature, humidity, and wind speed data. NaN and blank values were removed from the temperature, humidity, and wind speed data. The significant impact of meteorological factors on power load was determined using the Pearson correlation coefficient method. The temperature correlation coefficient was calculated and expressed as follows: Where r1 is the temperature correlation coefficient, T o The temperature value at time point 0. For the average temperature, L o Let be the power load value at time point 0. This represents the average power load. The humidity correlation coefficient is calculated as follows: Where r2 is the humidity correlation coefficient, H o The humidity value at time point 0. This represents the average humidity level. The wind speed correlation coefficient is expressed as follows: Where r3 is the wind speed correlation coefficient, W o Let be the wind speed value at time point 0. This represents the average wind speed. By conducting statistical analysis on temperature, humidity, and wind speed, thresholds θ1, θ2, and θ3 were set for the significant impact of temperature, humidity, and wind speed. The correlation coefficients for temperature, humidity, and wind speed were evaluated separately, and the significant impact of each meteorological factor was calculated and expressed as follows: Among them, S n Let r be the significance value of the nth meteorological factor. n Let θ be the nth correlation coefficient. n γ is the threshold for the nth significant effect. n To adjust the parameters of the curve; When the significance value of the temperature correlation coefficient assessment is greater than 0.5, the temperature factor data in the meteorological factors are retained; when the significance value of the temperature correlation coefficient assessment is less than or equal to 0.5, the temperature factor data in the meteorological factors are not retained. When the significance value of the humidity correlation coefficient assessment is greater than 0.5, the humidity factor data in the meteorological factors is retained; when the significance value of the humidity correlation coefficient assessment is less than or equal to 0.5, the humidity factor data in the meteorological factors is not retained. When the significance value of the wind speed correlation coefficient assessment is greater than 0.5, the wind speed factor data in the meteorological factors is retained; when the significance value of the wind speed correlation coefficient assessment is less than or equal to 0.5, the wind speed factor data in the meteorological factors is not retained.
3. The short-term load forecasting method based on a similar day and combination model as described in claim 2, characterized in that: The construction of the historical dataset includes selecting similar days based on meteorological factors, load date types, and pricing mechanisms to construct the historical dataset. Load date types include weekdays, weekends, and public holidays; The overall similarity between each day and the predicted day is calculated as follows: S(D,d)=(w wt ·f wt (D,d)+w ld ·f ld (D,d)+w price ·f price (D,d))·P(D,d) Where S(D,d) is the comprehensive similarity between the predicted day D and the historical day d, f wt (D,d) is the similarity function of meteorological factors, f ld (D,d) is a complex similarity function for the date type, f price (D,d) is the price mechanism similarity function, w wt As the weight of meteorological factors, w ld For the load date type weight, w price Let P(D,d) be the weight of the price mechanism, and P(D,d) be the penalty term. The similarity function of meteorological factors is expressed as: Among them, THW c Let σTHW be the data value of the c-th meteorological factor. c Let c be the standard deviation of the data for the c-th meteorological factor. dynamically adjust w based on historical data wt w ld and w price The obtained S(D,d) values are selected. There are jt extreme weather or special events in the predicted day. When jt equals 0 and the meteorological factor S(D,d) value is greater than or equal to 0.6, the historical day d and the predicted day D with S(D,d) value greater than or equal to 0.6 are selected as similar days. When jt equals 0 and the meteorological factor S(D,d) value is less than 0.6, the data of sudden events and natural disasters in the collected meteorological factor data are deleted, and the significant impact of meteorological factors on power load is recalculated using the Pearson correlation coefficient method. The comprehensive similarity value between the obtained predicted day D and historical day d is recalculated as S(D,d)2. When the S(D,d)2 value is greater than or equal to 0.6, the historical day d and the predicted day D with S(D,d)2 value greater than or equal to 0.6 are selected as similar days. When the S(D,d)2 value is less than 0.6, the historical day d and the predicted day D with S(D,d)2 value less than 0.6 are not selected as similar days, and the selection of similar days is repeated. When jt is greater than or equal to 1 and the meteorological factor S(D,d) value is greater than or equal to 0.75, historical days d and predicted days D with S(D,d) values greater than or equal to 0.75 are selected as similar days. When jt is greater than or equal to 1 and the meteorological factor S(D,d) value is less than 0.75, the collected meteorological factor data is expanded by deleting data that does not include sudden events and natural disasters. The comprehensive similarity value between predicted day D and historical day d is recalculated as S(D,d)3. When the S(D,d)3 value is greater than or equal to 0.75, historical days d and predicted days D with S(D,d)3 are selected as similar days. When the S(D,d)3 value is less than 0.75, historical days d and predicted days D with S(D,d)3 values less than 0.75 are not selected as similar days, and the next similar day to predicted day D is selected. The selected similar day data are used to construct a historical dataset.
4. The short-term load forecasting method based on a similar day and combination model as described in claim 3, characterized in that: The improved variational mode decomposition includes the use of an adaptive frequency bandwidth selection mechanism based on data features; The adaptive frequency bandwidth selection mechanism dynamically adjusts the frequency bandwidth as follows: BW=β(Var(d),Skew(d),Kurt(d))·SD(f)+δ·Entropy(f) Where BW is the adaptive frequency bandwidth, Var(d) is the variance of historical day d, Skew(d) is the skewness of historical day d, Kurt(d) is the kurtosis of historical day d, Entropy(f) is the information entropy of frequency f, β is the composite function of variance, skewness and kurtosis adjustment, δ is the adjustment parameter, and SD(f) is the standard deviation of frequency sequence f. The composite function β is expressed as: β(Var(d),Skew(d),Kurt(d)) =α·exp(-γ1·Var(d)-γ2·|Skew(d)|-γ3·Kurt(d)) Among them, α, γ1, γ2 and γ3 are set parameters.
5. The short-term load forecasting method based on a similar day and combination model as described in claim 4, characterized in that: The intrinsic mode function components include those obtained by using improved variational mode decomposition to decompose the historical dataset into intrinsic mode function components at different frequencies; The characteristics of the historical dataset include variance Var(d), skewness Skew(d), kurtosis Kurt(d), and information entropy Entropy(f); The frequency bandwidth BW is dynamically calculated using an improved variational mode decomposition method on the historical dataset, and is expressed as: Where V represents a set of IMF components obtained from variational mode decomposition, u k (t) represents each IMF component in V, {u k } represents the IMF component, {ω k Let} be the center frequency of the IMF component, δ(t) be the Dirac delta function, Re{·} be the real part operation, Hilbert{·} be the Hilbert transform, K be the number of IMF components, g(BW) be the function for adaptive frequency bandwidth BW adjustment, and exp(-jω) be the frequency response of the IMF component. k t) is a complex exponential function. This is the derivative with respect to time.
6. The short-term load forecasting method based on a similar day and combination model as described in claim 5, characterized in that: The improved least squares support vector machine model includes the introduction of an adaptive learning rate; During model training, the learning rate is adaptively adjusted based on the prediction error. The improved least squares support vector machine model is represented as follows: Where J(w,b,cx; α(t)) is the objective function of the improved least squares support vector machine model, w is the model weight, b is the bias term, cx is the error term, α(t) is the adaptive learning rate, t is the number of iterations, and C(t) is the regularization parameter adjusted with the number of iterations. The sum of squared errors for all training samples; The constraints of the improved least squares support vector machine model are expressed as follows: y z =w·φ(Combine(F1,F2,…,F K ))+b+cx i y z Let φ be the target value of the z-th training sample, and let φ be the kernel function mapping. Combine(F1,F2,…,F…) K ) is a combination function that combines the features F1, F2, ..., F of V. K Combined to form a single predictor variable, F k cx is the feature set extracted from the k-th IMF component. i Let be the error term for the i-th sample.
7. The short-term load forecasting method based on a similar day and combination model as described in claim 6, characterized in that: The optimal short-term load forecast result includes using an improved least squares support vector machine model to predict the intrinsic mode function components and output the optimal short-term load forecast result. Extract time-domain and spectral features from V, and integrate the time-domain and spectral features into a comprehensive feature set F. k ; The intrinsic mode function components V are input to the improved least squares support vector machine model, and the adaptive learning mechanism and the combination function Combine(F1,F2,…,F) are used to achieve this. K Training an improved least squares support vector machine model; Inputting the historical IMF components of similar days to the predicted day into the trained improved least squares support vector machine model, the improved least squares support vector machine model outputs the short-term load forecast result, which is expressed as follows: in, For the predicted results; The output short-term load forecast results This is considered the optimal short-term load forecast result.
8. A system employing a short-term load forecasting method based on a similar day and combination model as described in any one of claims 1 to 7, characterized in that: It includes a historical dataset module, an intrinsic mode function component module, and an optimal short-term load forecasting module; The historical dataset module is used to determine the meteorological influencing factors of multivariate loads using the Pearson correlation coefficient method, and to select similar days for meteorological factors, load date types, and pricing mechanisms to construct a historical dataset. The intrinsic mode function component module is used to decompose the historical dataset into intrinsic mode function components at different frequencies using improved variational mode decomposition; The optimal short-term load forecasting module is used to predict the intrinsic mode function components using an improved least squares support vector machine model, and outputs the optimal short-term load forecasting results.
9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the short-term load forecasting method based on a similar day and combination model as described in any one of claims 1 to 7.
10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the short-term load forecasting method based on a similar day and combination model as described in any one of claims 1 to 7.