Power load prediction method and system based on frequency domain adaptive filter

Abstract

The application discloses a power load prediction method and system based on a frequency domain adaptive filter, which generates available time series data by cleaning power load related data; converts time domain data into frequency domain data by using discrete Fourier transform on the time series data; introduces a frequency domain adaptive normalization layer to filter the amplitude of the frequency domain data, distinguish stationary and non-stationary frequency domain data, and restore time domain data by using inverse discrete Fourier transform; normalizes the stationary and non-stationary time domain data respectively; extracts features by using a frequency domain adaptive prediction model to generate corresponding stationary and non-stationary time domain prediction data; and weights and fuses the results of the stationary and non-stationary time domain prediction data to generate complete time series prediction data. The application significantly improves the prediction ability of stationary time domain data, can more accurately capture the change rule of non-stationary data, and thus enhances the overall prediction accuracy.

Description

Technical Field

[0001] This invention belongs to the field of power system operation and dispatching technology, specifically relating to a power load prediction method and system based on frequency domain adaptive filters. Background Technology

[0002] Electricity load forecasting is fundamental to the optimal scheduling and safe operation of power systems, directly impacting power supply planning, energy resource allocation, and demand response strategies. However, electricity load data exhibits complex characteristics such as nonlinearity, volatility, and intermittency, and contains significant stationary and non-stationary information. Traditional electricity load forecasting methods often fail to effectively separate and model these complex characteristics when processing such data, resulting in shortcomings in model accuracy and robustness. Existing load forecasting models generally face the following problems: First, the multi-scale nature of electricity load data makes it difficult for traditional models to effectively integrate different time dimensions and data features; second, due to the non-stationarity of load data, traditional models struggle to capture dynamic changes within the data, particularly performing poorly in understanding the relationship between high-frequency fluctuations and low-frequency trends; finally, the multivariate characteristics in electricity load data are often intertwined, making it difficult for models to effectively extract the correlations between variables.

[0003] Existing time series modeling methods, such as regression models and support vector machines, while possessing certain predictive capabilities, typically assume that load data is stationary and linear, making it difficult to handle complex non-stationary characteristics and frequency domain features. Deep learning models, such as Long Short-Term Memory (LSTM) networks and Transformers, while exhibiting significant advantages in nonlinear modeling, still have limitations in handling non-stationarity, noise, and frequency domain feature extraction. In particular, the prevalent non-stationarity, periodic fluctuations, and random noise in electricity load data severely impact the performance of prediction models. Therefore, there is an urgent need for a prediction model that can effectively separate stationary and non-stationary characteristics and fully utilize frequency domain information. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention combines frequency domain adaptive filtering techniques with time series modeling methods to effectively separate and process non-stationary characteristics in data, thereby improving the accuracy and robustness of power load forecasting. The technical solution adopted by this invention is as follows:

[0005] The power load forecasting method based on frequency domain adaptive filters includes the following steps:

[0006] Step 1: Obtain time series data of power load from relevant power load data;

[0007] In step 1, the power load-related data includes historical power load data, meteorological information, special event information, and user behavior characteristic data. The user behavior characteristic data includes the electricity consumption characteristics of different categories of users. The power load-related data is cleaned, including handling missing and outlier values, and smoothed using a moving average method to remove random noise. The smoothed data is then used as the input time series data. The smoothing formula is as follows:

[0008]

[0009] Among them, y t x represents the smoothed data. i This represents the original data points, N represents the window size, and t represents time.

[0010] Step 2: Perform a time-domain to frequency-domain conversion on the time series data to obtain a frequency-domain signal;

[0011] In step 2, the discrete Fourier transform is used to convert the time domain to the frequency domain, which can greatly reduce the computational complexity. Assume there is a time-domain signal x(t) of length N = {x0, x1, ..., x...}. N-1 The time series data is processed using the Discrete Fourier Transform algorithm to obtain the corresponding frequency domain signal X(f), as shown in the formula:

[0012]

[0013] Where X(f) represents the frequency domain signal, which contains amplitude and phase information of different frequency domains in the signal.

[0014] Step 3: Introduce a frequency domain adaptive normalization layer (L-FAN) to perform amplitude filtering on the frequency domain signal, distinguish the frequency domain signal into stationary frequency domain data and non-stationary frequency domain data, then convert the frequency domain data back to time domain data, and perform normalization processing on the stationary time domain data and non-stationary time domain data respectively.

[0015] Step 4: Construct and utilize the frequency domain adaptive prediction model to extract features from the normalized stationary and non-stationary time domain data and generate corresponding stationary and non-stationary time domain prediction data, respectively.

[0016] Step 5: Fuse the stationary time-domain prediction data and the non-stationary time-domain prediction data to generate a complete time series target output, as shown in the following formula:

[0017] Y final =α·Y stationary +(1-α)·Y non-stationary

[0018] Here, α represents the weighting factor, which is usually determined based on the quality or prediction accuracy of stationary and non-stationary time-domain data.

[0019] Furthermore, in step 3, the amplitude of each frequency component is first extracted from the frequency domain signal. The frequency components with large amplitude in the low-frequency part are taken as stationary frequency domain data, and the frequency components with small amplitude in the high-frequency part are taken as non-stationary frequency domain data.

[0020] The frequency domain signal is a complex frequency domain signal, and the formula is as follows:

[0021]

[0022] Where |X(f)| represents the amplitude of the frequency component extracted from the frequency domain signal X(f), and Re(X(f)) and Im(X(f)) represent the real and imaginary parts of the frequency domain signal X(f), respectively.

[0023] In step 3, the formula for performing an inverse discrete Fourier transform on the frequency domain data and converting it back to a time domain signal is as follows:

[0024]

[0025] Among them, Y k Represents frequency domain data, X n The time-domain signal after inverse transformation is represented by k, which represents the index of the time series corresponding to the frequency-domain data, and n, which represents the index of the time series corresponding to the time-series signal. k and n are not necessarily equal because some time-domain data may be Fourier transformed into the same frequency-domain data. N represents the length of the time series.

[0026] Furthermore, in step 3, the stationary time-domain data is normalized as follows:

[0027]

[0028] Among them, X stable This represents stationary time-domain data, and Norm represents the normalization operation. Mean represents the time-domain data of the first L time steps of the i-th time-domain data point. L Std represents the operation of calculating the average value along the time dimension. L This indicates the operation of calculating the standard deviation along the time dimension, where N represents the length of the stationary time-domain data and L represents the length of the backtracking window.

[0029] The non-stationary time-domain data is normalized as follows:

[0030]

[0031] For each non-stationary time-domain data X unstableThe affine transformation is achieved through the weighting parameter affine. weight and bias parameter affine bias A linear transformation is performed to enhance the expressive power of the model, resulting in normalized non-stationary time-domain data. σ represents the standard deviation, N′ represents the length of the non-stationary time-domain data, and μ represents the mean.

[0032] Furthermore, in step 4, a Discrete Fourier Transform is first performed on the first and second dimensions of the stationary time-domain data to transform it to the frequency domain, and orthogonalization is then performed to ensure that the energy before and after the transformation is consistent. The transformation considers the symmetry of the input sequence, thereby avoiding redundant frequency domain components. The formula is as follows:

[0033]

[0034] Among them, X k Let x represent the k-th frequency component in the frequency domain. n Represents normalized stationary time-domain data The nth value, where N represents the length of the input time-domain data and j represents the imaginary unit, ultimately yields the converted time-series data in the frequency domain. L represents the backtracking window length. Represents a collection of time series data;

[0035] Then, based on the transformed time series data Through learnable filters Generate the output of a learnable filter The formula is as follows:

[0036]

[0037] in, Indicates Fourier transform, Indicates the inverse Fourier transform, ⊙ L This represents the element-wise product along dimension L.

[0038] Furthermore, in step 4, the frequency-domain adaptive prediction model addresses the limitations of existing time-series prediction models that widely employ linear and attention mapping methods when processing stationary frequency-domain data. These methods struggle to efficiently handle frequency-domain characteristics and non-stationarity issues. This invention utilizes a learnable filter module to extract features from the stationary time-domain data, dynamically capturing key frequency-domain feature patterns to obtain the prediction result Y for the stationary time-domain data. stationary For the non-stationary time-domain data, a multilayer perceptron (MLP) is used to model the first K time-domain data, extract the evolution law of the non-stationary frequency domain mode, and obtain the prediction result Y of the non-stationary time-domain data.non-stationary Finally, the prediction result Y of the stationary time-domain data is weighted and then... stationary The prediction result Y of the non-stationary time-domain data non-stationary Aggregate the results to obtain the final prediction result Y. final The formula is as follows:

[0039] Y final =α·Y stationary +(1-α)·Y non-stationary

[0040] Y stationary =Learning_Filter(X stationary )

[0041] Y non-stationary =MLP(X non-stationary )

[0042] Where α represents the weighting factor, Learning_Filter represents the learnable filter, and MLP represents the multilayer perceptron.

[0043] Furthermore, the learnable filter module generates an initial learnable filter by randomly initializing learnable parameters and multiplying them with the input stationary time-domain data. It employs a channel-independent strategy to model the channels of the stationary time-domain data, where each channel is modeled separately to improve the filter's modeling accuracy and capture the main features between different channels. The learnable filter dynamically adjusts the weights of different frequency components based on the frequency domain features in the time-domain data. The filter can adjust for high-frequency changes and low-frequency trends to extract key frequency domain features during the modeling process, thereby improving the modeling capability of stationary time-domain data.

[0044] Furthermore, the output of the learnable filter module, after passing through a feedforward neural network consisting of two linear transformations and a Leaky ReLU activation function, yields the prediction result for stationary time-domain data. The formula for the Leaky ReLU activation function is as follows:

[0045]

[0046] When x≥0, the output is x, the same as standard ReLU; when x<0, the output is αx, where α represents a small positive number, usually between 0.01 and 0.1, used to control the slope of the negative input part, avoiding the "dead neuron" problem in standard ReLU where negative values ​​are directly output as 0; the Leaky ReLU activation function improves the model's learning ability by allowing small negative slopes, so that neurons remain activated even when receiving negative input.

[0047] Furthermore, the multilayer perceptron extracts feature patterns from non-stationary time-domain data, including multiple fully connected layers, each containing a nonlinear activation function, to mine deeper features in the data; for the frequency characteristics of non-stationary time-domain data, the time-domain components of the top K frequencies are selected, and the multilayer perceptron is used to model and predict these frequency time-domain components to capture key change patterns in non-stationary time-domain data.

[0048] Although normalization eliminates the impact of non-stationary data on prediction results, in the real world, the non-stationarity of data often reveals potential volatility and intermittency in future data. Therefore, it is necessary to model non-stationary data separately. To effectively model the changing patterns of non-stationary information, a multilayer perceptron is used to perform a linear transformation on the non-stationary time-domain data. Through matrix operations, the data is mapped from one space to another to achieve feature extraction, dimensionality adjustment, and data representation. It is an indispensable tool in neural network modeling and dimensionality reduction, providing a foundation for nonlinear activation and simplifying large-scale data computation. The linear transformation formula is as follows:

[0049] y = W·x + b

[0050] Where W represents the weight matrix, x represents the input data, b represents the bias term, and y represents the output of the linear transformation. The ReLU activation function is applied to the result of the linear transformation, with the formula: ReLU(x) = max(0,x). This operation can introduce nonlinear characteristics for feature extraction to enhance the expressive power of the model.

[0051] Furthermore, the output of the multilayer perceptron is passed through a linear projection layer consisting of two linear transformations and a ReLU activation function to map the intermediate prediction results of the multilayer perceptron to generate prediction results. Then, through normalization processing, the prediction results of non-stationary time domain data are obtained.

[0052] The power load forecasting system based on frequency domain adaptive filters includes a data acquisition module, a frequency domain conversion module, a frequency domain normalization module, a frequency domain adaptive prediction module, and a weighted fusion module. According to the power load forecasting method based on frequency domain adaptive filters, the system sequentially acquires time series data of power load, converts and normalizes frequency domain signals, extracts and predicts features, and weights and fuses the predicted data to generate a complete time series target output.

[0053] The advantages and beneficial effects of this invention are as follows:

[0054] This invention significantly improves the processing capability for stationary frequency domain data by introducing a learnable filter to replace traditional linear and attention mappings, especially in capturing high-frequency variations and low-frequency trends. It can more accurately identify high-frequency fluctuations and low-frequency trends in signals, thereby improving prediction accuracy and adaptability. Furthermore, by introducing a frequency-domain adaptive normalization layer, this invention effectively distinguishes between stationary and non-stationary data, eliminating the interference of non-stationarity on model predictions. Through adaptive processing, the frequency-domain adaptive normalization layer ensures the adaptability of the prediction model to different data types, significantly improving the accuracy and information completeness of the prediction results. Finally, by using a multilayer perceptron to model non-stationary patterns, especially optimizing the first K frequency domain signals, this invention enhances the model's dynamic response capability to non-stationary data. This mechanism enables the prediction model to more accurately capture the changing patterns in non-stationary data, thereby enhancing the overall prediction capability. Attached Figure Description

[0055] Figure 1 This is a flowchart of a method according to an embodiment of the present invention.

[0056] Figure 2 This is a framework diagram of a prediction model based on a frequency domain adaptive filter in an embodiment of the present invention.

[0057] Figure 3 This is a schematic diagram of the structure of the stationary frequency domain data processing module in an embodiment of the present invention.

[0058] Figure 4 This is a schematic diagram of the structure of the non-stationary frequency domain data processing module in an embodiment of the present invention. Detailed Implementation

[0059] The specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0060] This invention relates to a power load forecasting method based on frequency domain adaptive filters. This method improves the accuracy and robustness of power load forecasting, providing support for optimized power system scheduling, such as... Figure 1 , Figure 2 As shown, the specific implementation steps include the following:

[0061] Step 1: Obtain power load related data and clean the data to obtain usable time series data; specifically, the obtained power load related data is time series data, and missing values ​​and outliers are processed in the data.

[0062] To remove random noise, power load-related data can be smoothed using a moving average method, the formula of which is shown below:

[0063]

[0064] Among them, y t It is the smoothed data, x i These are the original data points, and N is the window size;

[0065] The smoothed data is then used as input time series data.

[0066] Step 2: Perform frequency domain conversion on the time series data using Discrete Fourier Transform to convert the time domain data into frequency domain data and obtain the frequency domain signal;

[0067] The Discrete Fourier Transform can significantly reduce computational complexity. Suppose we have a time-domain signal x(t) of length N = {x0, x1, ..., xt}. N-1 The time series data is processed using the Discrete Fourier Transform algorithm to obtain the corresponding frequency domain signal X(f), as shown in the formula:

[0068]

[0069] Where X(f) is a frequency domain signal, containing amplitude and phase information of different frequency domains in the signal.

[0070] Step 3: Introduce a frequency domain adaptive normalization layer to perform amplitude filtering on the frequency domain signal, distinguishing it into stationary and non-stationary frequency domain data, and using inverse Discrete Fourier transform to convert the frequency domain data into time domain data; and then, perform normalization processing on the stationary and non-stationary frequency domain data respectively, specifically including the following steps:

[0071] First, extract the amplitude of each frequency component from the frequency domain signal. Given a frequency domain signal X(f), the amplitude of each frequency component |X(f)| can be calculated, where X(f) is a complex frequency domain signal, as shown in the following formula:

[0072]

[0073] Where Re(X(f)) and Im(X(f)) are the real and imaginary parts of the frequency domain signal, respectively.

[0074] The key to the frequency domain adaptive normalization layer is to distinguish between stationary and non-stationary data. This distinction is based on the size of the frequency domain or the amplitude of the frequency domain components. In practical applications, a threshold is used to distinguish between stationary and non-stationary data.

[0075] Assuming the amplitude of each frequency component is |X(f)|, the following criteria can be used to distinguish between stationary and non-stationary data:

[0076] Stable data: Low-frequency components, typically those with larger amplitudes.

[0077] Non-stationary data: High-frequency components, typically frequency domain components with small amplitudes.

[0078] Specifically, the frequency domain component f of stationary data stable If a certain amplitude threshold condition is met, assuming the threshold is T, then the condition for stationary data can be defined as follows:

[0079] |X(f)|≥T

[0080] The frequency domain component f of non-stationary data unstable satisfy:

[0081] |X(f)| <T

[0082] After distinguishing between stationary and non-stationary frequency domain data, for all frequency domain Y... k Performing the inverse discrete Fourier transform to convert it back to a time-domain signal, the formula is as follows:

[0083]

[0084] Among them, Y k It is a frequency domain signal, X n It is the time-domain signal after inverse transformation.

[0085] Next, we will normalize these two types of data respectively. The normalization process for stationary data is as follows:

[0086] Input X for time series stable Using the instance normalization method, represented as Norm, it can be expressed as:

[0087]

[0088] Among them, Mean L Std represents the operation of calculating the average value along the time dimension. L This represents the operation of calculating the standard deviation along the time dimension, where N represents the length of the input signal and L represents the length of the backtracking window.

[0089] For a non-stationary data normalization process, each data point X unstable The mean μ will be subtracted from each dataset X, where the mean μ is the value of the dataset X. unstable The mean of the data points is N, where N′ is the total number of data points. The formula for calculating N′ is:

[0090]

[0091] Assume the standard deviation σ is:

[0092]

[0093] Then, the standardization operation will process each data point X stable Divide by the standard deviation σ:

[0094]

[0095] For each data point X unstable Affine transformation through weighted affine weight and bias affine bias Perform a linear transformation:

[0096]

[0097] affine here weight and affine bias These are learnable parameters that can be updated via gradients during training. This transformation adjusts the scale and offset of the data, enhancing the model's expressive power.

[0098] Step 4: Construct and utilize a frequency domain adaptive prediction model to extract features from the normalized stationary and non-stationary time-domain data, respectively, and generate corresponding stationary and non-stationary time-domain prediction data. Specifically, this includes:

[0099] By using learnable filters to replace traditional linear mapping and attention mapping, feature extraction of stationary time-domain data is performed, and the main time-domain feature patterns are dynamically captured. In particular, the modeling ability of stationary time-domain data is improved, especially in handling high-frequency changes and low-frequency trends. At the same time, for non-stationary time-domain data separated from the input data, multilayer perceptron is used for modeling to predict the first K frequency domain signals and capture the evolution law of non-stationary patterns.

[0100] Considering that power load data usually contains obvious stationary and non-stationary signals, the linear mapping and attention mapping widely used in existing time series forecasting models have limitations when processing stationary time-domain data, and are difficult to efficiently deal with time-domain characteristics and non-stationarity issues.

[0101] To address the shortcomings of existing prediction models, this invention proposes a power load prediction method based on a frequency domain adaptive filter. The learnable filter is used to overcome the aforementioned problems. The specific steps are as follows:

[0102] like Figure 3 As shown, a Discrete Fourier Transform is first performed on the first and second dimensions of the stationary time-domain data to transform x into the frequency domain, and orthogonalization is then applied to ensure consistent energy before and after the transformation. The transformation takes into account the symmetry of the input sequence, thus avoiding redundant frequency domain components. The formula is as follows:

[0103]

[0104] Among them, X k It is the k-th frequency component in the frequency domain, x n It is a time-domain signal The nth value, where N is the length of the input signal and j is the imaginary unit, ultimately yields the converted time series data. L is the length of the backtracking window.

[0105] Next, given the input time series and learnable filters The formula is as follows:

[0106]

[0107] in, It is a Fourier transform. It is the inverse Fourier transform, ⊙ L This represents the element-wise product along dimension L. Ind represents a channel-independent learnable filter, and Ind represents the channel-independent learnable filter strategy. Indicates the total number of channels. It is the output of the learnable filter.

[0108] The above components form a learnable filter prediction model, and the overall formula is as follows:

[0109] Y stationary =Learning_Filter(X stationary )

[0110] After obtaining the intermediate prediction results, the data are processed through a feedforward neural network consisting of two linear transformations and a Leaky ReLU activation function. The formula for the Leaky ReLU activation function is as follows:

[0111]

[0112] When x≥0, the output is x, the same as standard ReLU; when x<0, the output is αx, where α is a small positive number, usually between 0.01 and 0.1, used to control the slope of the negative input part, avoiding the "dead neuron" problem in standard ReLU where negative values ​​are directly output as 0; Leaky ReLU can improve the model's learning ability by allowing small negative slopes, so that neurons remain active even when receiving negative inputs.

[0113] The intermediate prediction results are mapped to generate the final prediction results. After normalization, the final prediction results are used to obtain the target output of stable prediction data.

[0114] The core of the learnable filter lies in dynamically adjusting the weights of different frequency domain components based on the characteristics of the frequency domain signal. Specifically, the filter can adjust for high-frequency changes and low-frequency trends, extracting key frequency domain features during the modeling process to improve the modeling capability of stationary time domain data.

[0115] Although the normalization process eliminates the impact of non-stationary data on the prediction results, in the real world, the non-stationarity of data often reveals the volatility and intermittency that future data may contain. Therefore, it is particularly important to model non-stationary data separately.

[0116] To effectively model the changing patterns of non-stationary information, this invention proposes a power load forecasting method based on a frequency domain adaptive filter, employing the aforementioned multilayer perceptron to overcome the problems. The specific steps are as follows:

[0117] like Figure 4 As shown, for non-stationary time-domain data, firstly, linear transformation maps the data from one space to another through matrix operations, achieving feature extraction, dimensionality adjustment, and data representation. It is an indispensable tool in neural network modeling and dimensionality reduction, providing a foundation for nonlinear activation and simplifying large-scale data computation. The linear transformation formula is as follows:

[0118] y = W·x + b

[0119] Where W is the weight matrix, x is the input data, b is the bias term, and y is the output.

[0120] Applying the ReLU activation function to the result of the linear transformation, with the formula: ReLU(x) = max(0,x), can introduce nonlinear characteristics and enhance the expressive power of the model.

[0121] The overall formula for the multilayer perceptron prediction model is as follows:

[0122] Y non-stationary =MLP(X non-stationary )

[0123] After obtaining the intermediate prediction results, the results are passed through a linear projection layer consisting of two linear transformations and a ReLU activation function to map the intermediate prediction results to generate the final prediction results. The final prediction results are then normalized to obtain the target output of the non-stationary prediction data.

[0124] The core of the multilayer perceptron lies in extracting feature patterns from non-stationary time-domain data through a multilayer neural network. The multilayer perceptron includes multiple fully connected layers, each containing a nonlinear activation function to mine deep-level features in the data. For the frequency characteristics of non-stationary time-domain data, the top K frequency components are selected, and the multilayer perceptron is used to model and predict these frequency components to capture the key change patterns of non-stationary data.

[0125] Step 5: Weight and fuse the results of the stationary time-domain prediction data and the non-stationary time-domain prediction data described in Step 4 to generate a complete time series target output.

[0126] By combining the prediction results of stationary and non-stationary data through weighted summation, the interference of non-stationary data on the prediction of stationary frequency domain data can be reduced. The formula is as follows:

[0127] Y final =α·Y stationary +(1-α)·Y non-stationary

[0128] Here, α is a weighting factor, which is usually determined based on the quality of stationary and non-stationary data or the accuracy of prediction.

[0129] Ultimately, the model of this invention can not only effectively handle the non-stationarity in power load data, but also significantly improve the frequency domain modeling capability of complex time series, and has strong engineering applicability and promotion value.

[0130] The power load forecasting system based on frequency domain adaptive filters includes: a data acquisition and cleaning module, a frequency domain conversion module, a frequency domain normalization module, a feature extraction and prediction module, and a weighted fusion module.

[0131] The data acquisition and cleaning module is used to acquire power load-related data and clean the data to generate usable time-series data.

[0132] The frequency domain conversion module is used to perform a discrete Fourier transform on time series data, converting the time domain data into frequency domain data to obtain a frequency domain signal;

[0133] The frequency domain normalization module is used to perform amplitude filtering on the frequency domain signal, distinguish the frequency domain signal into stationary frequency domain data and non-stationary frequency domain data, convert the frequency domain data back to time domain data through inverse discrete Fourier transform, and perform normalization processing on the stationary time domain data and non-stationary time domain data respectively.

[0134] The feature extraction and prediction module is used to extract features from normalized stationary time-domain data to generate stationary time-domain prediction data; and to extract features from normalized non-stationary time-domain data to generate non-stationary time-domain prediction data.

[0135] The weighted fusion module is used to weight and fuse the results of stationary and non-stationary time-domain prediction data to generate a complete time series target output.

[0136] On the other hand, the present invention provides a computer-readable storage medium storing program code for execution by a device, the program code including steps for performing any of the implementation methods such as a power load forecasting method based on a frequency domain adaptive filter.

[0137] On the other hand, the present invention provides an electronic device including a processor, a memory, and a program or instructions stored in the memory and executable on the processor, wherein the program or instructions, when executed by the processor, implement the steps of any of the implementation methods of the power load forecasting method based on a frequency domain adaptive filter.

[0138] 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 the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A power load forecasting method based on a frequency domain adaptive filter, characterized in that... Includes the following steps: Step 1: Obtain time series data of power load from relevant power load data; Step 2: Perform a time-domain to frequency-domain conversion on the time series data to obtain a frequency-domain signal; Step 3: Introduce a frequency domain adaptive normalization layer to perform amplitude filtering on the frequency domain signal, distinguish the frequency domain signal into stationary frequency domain data and non-stationary frequency domain data, then convert the frequency domain data back to time domain data, and perform normalization processing on the stationary time domain data and non-stationary time domain data respectively. First, extract the amplitude of each frequency component from the frequency domain signal. Then, take the frequency components with large amplitude in the low-frequency part as stationary frequency domain data and the frequency components with small amplitude in the high-frequency part as non-stationary frequency domain data. The stationary time-domain data is normalized as follows: Among them, X stable This represents stationary time-domain data, and Norm represents the normalization operation. Mean represents the time-domain data of the first L time steps of the i-th time-domain data point. L Std represents the operation of calculating the average value along the time dimension. L This indicates the operation of calculating the standard deviation along the time dimension, where N represents the length of the stationary time-domain data and L represents the length of the backtracking window. The non-stationary time-domain data is normalized as follows: For each non-stationary time-domain data X unstable The affine transformation is achieved through the weighting parameter affine. weight and bias parameter affine bias Perform a linear transformation to obtain normalized non-stationary time-domain data. σ represents the standard deviation, N′ represents the length of the non-stationary time-domain data, and μ represents the mean; Step 4: Construct and utilize the frequency domain adaptive prediction model to extract features from the normalized stationary and non-stationary time domain data and generate corresponding stationary and non-stationary time domain prediction data, respectively. First, the stationary time-domain data is subjected to a Discrete Fourier Transform to transform it to the frequency domain, and then orthogonalized. The symmetry of the input sequence is considered during the transformation. The formula is as follows: Among them, X k Let x represent the k-th frequency component in the frequency domain. n Represents normalized stationary time-domain data The nth value, where N represents the length of the input time-domain data and j represents the imaginary unit, ultimately yields the converted time-series data in the frequency domain. L represents the backtracking window length. Represents a collection of time series data; Then, based on the transformed time series data Through learnable filters Generate the output of a learnable filter The formula is as follows: in, Indicates Fourier transform, Indicates the inverse Fourier transform, ⊙ L Represents the element-wise product along dimension L; The frequency domain adaptive prediction model uses a learnable filter module to extract features from the stationary time domain data, dynamically capturing the main frequency domain feature patterns to obtain the prediction results for the stationary time domain data. For the non-stationary time domain data, a multilayer perceptron is used to model the first K time domain data, extracting the evolution law of the non-stationary frequency domain patterns to obtain the prediction results for the non-stationary time domain data. Finally, the prediction results for the stationary time domain data and the prediction results for the non-stationary time domain data are aggregated by weighting to obtain the final prediction result. Step 5: Fuse the stationary time-domain prediction data and the non-stationary time-domain prediction data to generate a complete time series target output.

2. The power load forecasting method based on a frequency domain adaptive filter according to claim 1, characterized in that: The learnable filter module generates an initial learnable filter by randomly initializing learnable parameters and multiplying them with the input stationary time-domain data; it then models the channels of the stationary time-domain data using a channel-independent strategy, where each channel is modeled separately to capture the main features between different channels. Learnable filters dynamically adjust the weights of different frequency components based on the frequency domain characteristics in the time domain data in order to extract key features.

3. The power load forecasting method based on a frequency domain adaptive filter according to claim 1, characterized in that: The output of the learnable filter module is passed through a feedforward neural network consisting of two linear transformations and a Leaky ReLU activation function to obtain the prediction result of stationary time-domain data. The formula for the Leaky ReLU activation function is as follows: When x≥0, the output is x, the same as standard ReLU; when x<0, the output is αx, where α represents a small positive number.

4. The power load forecasting method based on a frequency domain adaptive filter according to claim 1, characterized in that: The multilayer perceptron includes multiple fully connected layers, each containing a nonlinear activation function, used to mine deep features in the data; for the frequency characteristics of non-stationary time-domain data, the time-domain components of the top K frequencies are selected, and the multilayer perceptron is used to model and predict these frequency time-domain components to capture the key change patterns of non-stationary time-domain data. The multilayer perceptron performs linear transformations on non-stationary time-domain data, mapping the data from one space to another through matrix operations to achieve feature extraction, dimensionality adjustment, and data representation. An activation function is applied to the result of the linear transformation to introduce nonlinear characteristics for feature extraction.

5. The power load forecasting method based on a frequency domain adaptive filter according to claim 1, characterized in that: The output of the multilayer perceptron is passed through a linear projection layer consisting of two linear transformations and a ReLU activation function to map the intermediate prediction results of the multilayer perceptron to generate the prediction results. Then, through normalization processing, the prediction results of non-stationary time domain data are obtained.

6. A power load forecasting system based on a frequency domain adaptive filter, comprising a data acquisition module, a frequency domain conversion module, a frequency domain normalization module, a frequency domain adaptive prediction module, and a weighted fusion module, wherein the power load forecasting method based on a frequency domain adaptive filter according to any one of claims 1 to 5 sequentially performs the acquisition of time series data of power load, conversion and normalization of frequency domain signals, feature extraction and prediction, and weighted fusion of prediction data to generate a complete time series target output.

Images