Cable load prediction method based on time sequence analysis

Abstract

The invention relates to the field of cable load prediction, in particular to a cable load prediction method based on time sequence analysis. The method comprises the following steps: firstly, based on collected key variable data, constructing a state vector, and forming a time sequence matrix; extracting time domain features of the time sequence matrix to obtain a time domain feature vector; then, based on the time sequence matrix, introducing fast Fourier transform to obtain a frequency domain feature matrix; obtaining a composite feature matrix based on the time domain feature vector and the frequency domain feature matrix; based on the composite feature matrix, introducing an attention mechanism, and generating a comprehensive feature vector; and finally, performing load prediction based on the comprehensive feature vector, and outputting a prediction result based on historical residual correction. The technical problems that non-linear features in cable load changes cannot be effectively captured, and particularly the prediction precision is low under complex working conditions or emergency situations are solved.

Description

Technical Field

[0001] This invention relates to the field of cable load prediction, and more particularly to a cable load prediction method based on time series analysis. Background Technology

[0002] Cable load management is a crucial component in ensuring the safe, stable, and efficient operation of the power grid. Cables, as an essential part of the power system, bear vital electrical energy transmission tasks, and their load status directly impacts the stability and operational safety of the power system. To prevent cable overload and damage, power companies need to monitor and predict cable loads in real time. However, cable load prediction is highly complex due to the influence of various factors, including current, voltage, temperature, and ambient temperature. Current traditional cable load prediction methods typically employ linear prediction models based on physical models or simple statistical methods. These methods rely on assumptions or simplified models, neglecting the complex nonlinear relationships between cable load and multiple factors. For example, linear regression models based on historical data, while simple to implement, have low prediction accuracy when faced with the volatility, suddenness, and nonlinear changes in cable load. Furthermore, traditional methods lack sufficient adaptability to abnormal operating conditions in the power system, such as sudden load changes and equipment failures, easily leading to prediction errors and affecting optimized power system scheduling and fault early warning. With the increasing complexity and intelligence of power systems, higher demands are placed on the accuracy and real-time performance of cable load prediction. Traditional linear forecasting methods have significant limitations in practical applications because they cannot fully explore the complex relationship between cable load and environmental factors. To address this issue, we propose a cable load forecasting method based on time series analysis. Summary of the Invention

[0003] This invention provides a cable load prediction method based on time series analysis to solve the technical problem that existing methods usually rely on linear regression or statistical models, which often fail to effectively capture the nonlinear characteristics of cable load changes, especially in complex or sudden situations where prediction accuracy is low.

[0004] The present invention provides a cable load prediction method based on time series analysis, comprising the following steps: S1. Based on the collected key variable data, construct a state vector to form a time series matrix; extract time-domain features from the time series matrix to obtain a time-domain feature vector; S2. Based on the time series matrix, a fast Fourier transform is introduced to obtain the frequency domain feature matrix; based on the time domain feature vector and the frequency domain feature matrix, a composite feature matrix is ​​obtained; based on the composite feature matrix, an attention mechanism is introduced to generate a comprehensive feature vector. S3. Based on the comprehensive feature vector, perform load prediction and output the prediction results based on historical residual correction.

[0005] Preferably, S1 specifically includes: Based on the time series matrix, the time-domain feature vector of each key variable is calculated. After the dimensionless processing of each time-domain feature vector, they are combined into a time-domain feature vector.

[0006] Preferably, S2 specifically includes: For the variables in the time series matrix, perform a fast Fourier transform to obtain the frequency domain signal of each variable; extract the frequency domain features of each variable's frequency domain signal, and concatenate the frequency domain features of each variable row by row to form a frequency domain feature matrix.

[0007] Preferably, S2 specifically includes: The frequency domain feature matrix and the time domain feature vector are horizontally concatenated to construct a composite feature matrix.

[0008] Preferably, S2 specifically includes: An attention mechanism is introduced to perform weighted calculations on the time-domain and frequency-domain features in the composite feature matrix, and the weighted features are summed to obtain a comprehensive feature vector.

[0009] Preferably, S3 specifically includes: Based on the comprehensive feature vector, a residual enhancement mechanism is introduced. After each cable load prediction, the error between the predicted cable load value and the actual load value at the previous moment is calculated, which is the residual.

[0010] Preferably, S3 specifically includes: The residual is incorporated into the cable load prediction calculation at the current moment to obtain the cable load prediction value at the next moment.

[0011] Preferably, S3 specifically includes: Based on the predicted cable load value at the next time step, a nonlinear error correction mechanism is introduced to obtain the prediction result based on historical residual correction.

[0012] The beneficial effects of the technical solution of the present invention are: 1. By introducing residuals as a self-feedback term into cable load prediction calculations, dynamic correction of short-term errors is achieved. The application of the residual enhancement mechanism endows the load prediction method with a good ability to cope with cable load fluctuations, making it particularly suitable for scenarios with frequent or sudden cable load fluctuations, significantly improving the flexibility and robustness of prediction.

[0013] 2. By extracting the time-domain and frequency-domain features of the time series matrix, the regular fluctuations of cables under factors such as day-night cycles and power grid load changes are reflected. An attention mechanism is used to dynamically allocate the weights of each feature, strengthening the influence of key variables on prediction and effectively improving feature utilization efficiency and the discriminative ability of the load forecasting method.

[0014] 3. While maintaining prediction accuracy, the load prediction method abandons redundant structural design and adopts lightweight neural networks and mathematical functions. It has a small number of parameters and low computational cost, and can be efficiently deployed in field monitoring systems or embedded devices to meet real-time prediction requirements and has good engineering feasibility. Attached Figure Description

[0015] Figure 1 This is a flowchart of a cable load prediction method based on time series analysis as described in this invention. Detailed Implementation

[0016] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0017] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0018] The following description, in conjunction with the accompanying drawings, details a specific scheme for a cable load prediction method based on time series analysis provided by the present invention.

[0019] See attached document Figure 1 The diagram illustrates a flowchart of a cable load prediction method based on time series analysis provided by an embodiment of the present invention. The method includes the following steps: S1. Based on the collected key variable data, construct a state vector to form a time series matrix; extract time-domain features from the time series matrix to obtain a time-domain feature vector.

[0020] First, sensors installed on the cable itself are used to periodically collect key variable data during cable operation. These key variables include cable current, cable voltage, conductor temperature, and ambient temperature. The collected data on cable current, cable voltage, conductor temperature, and ambient temperature each time are combined to form a state vector representing the cable's operating state at a given moment. The mathematical expression is as follows: ,in This represents the corresponding data collection time. represent Cable current at any given time represent Cable voltage at any given time represent The conductor temperature at time [time] represent The ambient temperature at any given time. To ensure data integrity and validity when collecting key variable data, it is necessary to guarantee that data at each time step is continuous and without gaps. If data loss occurs, linear interpolation based on data from adjacent time points can be used to supplement the lost data. Simultaneously, the sampling period should be set based on the core principle of accurately capturing cable load fluctuations. As a specific example, a standard time step can be set to collect key variable data every 10 minutes.

[0021] The state vectors are further processed using a sliding window method, with the window size set according to the periodic characteristics of power system load fluctuations. As a specific example, a time series matrix is ​​constructed by collecting key variable data from the cable's operation over a 60-minute period. Using the previously established sampling period of 10 minutes per time step, the 60-minute span will contain data from six consecutive moments, i.e., six time steps, each corresponding to an independent state vector. These six state vectors collected over the 60 minutes are then horizontally concatenated in chronological order to form a time series matrix containing the historical operating states of the cable. The mathematical expression is ,in, express The state vector of the cable's operating state at time 1. express The state vector of the cable's operating state at time 1. express The state vector of the cable's operating state at time 1. express The state vector of the cable's operating state at time 1. express The state vector of the cable's operating status at any given time. The time series matrix fully covers the cable's operating status information over the past six time steps, clearly reflecting the dynamic process of cable load changes over time and providing comprehensive historical data support for subsequent analysis.

[0022] Finally, for the time series matrix For each type of physical variable in the data, such as cable current, cable voltage, conductor temperature, and ambient temperature, time-domain features of the cable's operating status are extracted over a 60-minute time span. When extracting time-domain features, the time series matrix... For each key variable in the data, such as cable current and cable voltage, feature extraction is performed. The feature extraction process includes calculating the time-domain feature vector of each key variable, such as statistical analysis. Cable current at any given time Temporal feature vectors , ,in, for The mean value of the cable current at any given time is obtained by calculating the arithmetic mean of all cable currents over a 60-minute time span. for The variance of the cable current at any given time; for The maximum value of the cable current at any given moment, that is, the maximum value of all cable currents within a 60-minute time span; for The minimum value of the cable current at a given moment, that is, the minimum value of all cable currents within a 60-minute time span; for The kurtosis of the cable current at a given time. Time-domain features such as mean, variance, maximum, minimum, and kurtosis are calculated using methods well-known to those skilled in the art. After extracting the time-domain features, the following is obtained: Time-domain eigenvector of cable current at time t Time-domain eigenvectors of cable voltage Time-domain eigenvectors of conductor temperature Time-domain feature vector of ambient temperature The extracted time-domain feature vectors are then dedimensionalized, and the Z-score standardization method is used to unify the dimensions of each time-domain feature vector. Finally, the dedimensionalized time-domain feature vectors are combined into a single time-domain feature vector. The time-domain feature vector can effectively reflect the law of cable load change over time. For example, the trend of current and voltage change can reflect the normal or abnormal fluctuation of the load. Temperature data (conductor temperature and ambient temperature) helps to assess the thermal stress of the cable during operation. The calculated time-domain feature vector reflects the dynamic change of the cable load over a period of time, providing important input data for subsequent load prediction methods.

[0023] S2. Based on the time series matrix, a fast Fourier transform is introduced to obtain the frequency domain feature matrix; based on the time domain feature vector and the frequency domain feature matrix, a composite feature matrix is ​​obtained; based on the composite feature matrix, an attention mechanism is introduced to generate a comprehensive feature vector.

[0024] First, for the time series matrix, a Fast Fourier Transform (FFT) is introduced for frequency domain analysis, achieving accurate conversion from the time domain signal to the frequency domain signal. The previously constructed time series matrix... Using the time series matrix as input, a Fast Fourier Transform (FFT) is performed on each variable represented by a row (cable current, cable voltage, conductor temperature, ambient temperature, etc.) to obtain the frequency domain signal for each variable. Key features, i.e., frequency domain features, are extracted from the frequency domain signal of each variable: the three frequency components with the largest amplitudes and their corresponding amplitudes are selected; based on the specific time-scale patterns of cable load changes in the power system and signal processing theory, the energy integral within the 0.1-1Hz frequency band is calculated; and statistical quantities such as spectral entropy are obtained based on information entropy theory. Standardization methods such as Z-score are used to ensure uniform feature scale. Finally, the frequency domain features of each variable are concatenated row by row to form a frequency domain feature matrix that comprehensively characterizes the cable's operating status. This can help subsequent steps calculate and identify the daily fluctuation patterns, peak values, and periodic changes in the load.

[0025] Then, the frequency domain feature matrix obtained after frequency domain feature extraction is... With time-domain feature vectors Perform horizontal stitching to construct a composite feature matrix containing multidimensional information. Before data splicing, pay attention to data standardization, ensuring that frequency and time domain features have undergone appropriate standardization or normalization, such as Z-score standardization, to avoid the problem of directly splicing frequency and time domain features having an excessive impact on the load prediction method. The composite feature matrix contains multi-dimensional information of time and frequency domain features, which can provide a more comprehensive description of cable load for the load prediction method.

[0026] Finally, existing attention mechanisms, such as soft attention mechanisms, are introduced to address the complex feature matrix. The time-domain and frequency-domain features are weighted and calculated. The attention mechanism dynamically calculates the weight of each feature based on its contribution to the prediction of cable load changes, and then sums the weighted features to obtain a comprehensive feature vector. , It integrates time-domain and frequency-domain information, and dynamically adjusts the influence of each feature based on the attention mechanism.

[0027] S3. Based on the comprehensive feature vector, perform load prediction and output the prediction results based on historical residual correction.

[0028] After obtaining the comprehensive eigenvector, the load prediction stage begins, where the cable load at the next time step is predicted. This process involves three important stages: load prediction, residual enhancement mechanism, and nonlinear error correction mechanism. The first step is to perform load prediction; input the integrated feature vector and historical data into an existing nonlinear prediction model, such as XGBoost, to predict the load for the next time step. Forecast of cable load at any given time. Historical data consists of key variable data from the cable operation process collected at 10-minute sampling intervals, used to construct a time series matrix. This data includes cable current, cable voltage, conductor temperature, and ambient temperature data for specific time periods in the past (60 minutes when constructing the time series matrix).

[0029] The second step involves employing a residual enhancement mechanism. After each cable load prediction, the error between the predicted cable load value and the actual load value at the previous moment, i.e., the residual, is calculated using the following formula: , in, yes The actual load value at any given time is collected by the monitoring system; yes The basic load forecast value is obtained based on the comprehensive feature vector and historical data, that is, the cable load forecast value obtained by using the nonlinear prediction model; express The residual at any given time represents the difference between the predicted cable load and the actual load.

[0030] The calculated residuals are then used as a self-feedback term in the current cable load prediction calculation, resulting in an optimized cable load prediction algorithm. The purpose is to incorporate historical residuals to adjust the current cable load prediction value, enabling the prediction results to adapt to fluctuations or sudden changes in cable load. By introducing residuals, the cable load prediction calculation can adjust the current prediction based on historical residuals, thereby improving the flexibility of short-term predictions. The formula for the optimized cable load prediction algorithm is as follows: , in, The predicted cable load for the next moment is a combined prediction that integrates the predicted trend (inferred from the comprehensive eigenvector) and the historical correction feedback (residual). for The base load forecast value at any given time; and This is the residual adjustment coefficient. Used to control the impact of the current time residual on the predicted cable load value. Used to control for historical residuals, i.e., the combined impact of errors from multiple past moments on current cable load forecasting, especially suitable for handling periodic load changes or long-term trends, determined based on expert experience. and The range of values ​​is ; For the front The sum of residuals within a historical moment represents the past. Prediction error feedback at each time point For a historic moment The residual; Indicates the number of historical moments.

[0031] The third step involves employing a nonlinear error correction mechanism. After performing the cable load prediction described above, the error between the current predicted cable load value and the actual load value is recalculated using the same method as the residual calculation method in the second step. The residual is then calculated. Meanwhile, to avoid abrupt expansion of residuals and improve the stability of cable load prediction, the residuals... Input to hyperbolic tangent function In this process, it can compress the residuals to a reasonable range, reducing the amplification of errors. The corrected final prediction result is output using the following formula: , in, This indicates a forecast result based on historical residual correction. The forecast takes into account the impact of historical residuals on future predictions, and can better adapt to load fluctuations and sudden changes; error correction coefficient. and These are the learning parameters, obtained by gradually adjusting them using the existing gradient descent method. Used to control the current time. The magnitude of the residual correction, Used to control the smoothness of historical residual correction. and The range of values ​​is limited to [0,1], introducing and This is to control the impact of current and historical residuals on the final prediction results, ensure that prediction errors are reasonably corrected, and improve the stability and accuracy of prediction results; Used to transfer residuals Mapped to The range is used to reduce the impact of extreme errors; Used for history Step residuals are smoothed to reduce the impact of excessive historical residuals; For a historic moment The residual.

[0032] The above three steps improve the accuracy and robustness of cable load forecasting. The combination of residual adjustment and error correction mechanisms enables the load forecasting method to maintain high-precision load forecasting capability even when there are large fluctuations or sudden changes in power load, thereby enhancing the applicability and reliability of the load forecasting method in complex power systems.

[0033] In summary, a cable load prediction method based on time series analysis has been developed.

[0034] The order of the embodiments is for illustrative purposes only and does not represent the superiority or inferiority of the embodiments. The processes depicted in the drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are possible or may be advantageous.

[0035] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

[0036] 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 of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A method for cable load prediction based on time series analysis, characterized in that, The method comprises the following steps: S1. Based on the collected key variable data, a state vector is constructed to form a time series matrix; time domain features are extracted from the time series matrix to obtain a time domain feature vector; S2. Based on the time series matrix, a fast Fourier transform is introduced to obtain a frequency domain feature matrix; Based on the time domain feature vector and the frequency domain feature matrix, a composite feature matrix is obtained; based on the composite feature matrix, an attention mechanism is introduced to generate a comprehensive feature vector; S3. Based on the comprehensive feature vector, load prediction is performed, and a prediction result based on historical residual correction is output.

2. The cable load prediction method based on time series analysis according to claim 1, characterized in that, The S1 specifically comprises: Based on the time series matrix, the time domain feature vector of each key variable is calculated, and after the dimensionless processing of each time domain feature vector, the time domain feature vectors are combined into a time domain feature vector.

3. The method of claim 1, wherein, The S2 specifically comprises: The fast Fourier transform is performed on the variables in the time series matrix to obtain the frequency domain signal of each variable; the frequency domain features of each variable are extracted, and the frequency domain features of each variable are spliced by rows to form a frequency domain feature matrix.

4. The method of claim 3, wherein, The S2 specifically comprises: The frequency domain feature matrix and the time domain feature vector are horizontally spliced to construct a composite feature matrix.

5. The method of claim 4, wherein, The S2 specifically comprises: An attention mechanism is introduced to weight the time domain features and the frequency domain features in the composite feature matrix, and the weighted features are added to obtain a comprehensive feature vector.

6. The method of claim 1, wherein, The S3 specifically comprises: Based on the comprehensive feature vector, a residual enhancement mechanism is introduced to calculate the error between the cable load prediction value at the last time and the actual load value, i.e., the residual, after each cable load prediction.

7. The method of claim 6, wherein, The S3 specifically comprises: The residual is introduced into the cable load prediction calculation at the current time to obtain the cable load prediction value at the next time.

8. The method of claim 7, wherein, The S3 specifically comprises: Based on the cable load prediction value at the next time, a nonlinear error correction mechanism is introduced to obtain a prediction result based on historical residual correction.

Images