Power load prediction and optimization method based on artificial intelligence

Abstract

The invention discloses a power load prediction and optimization method based on artificial intelligence, particularly relates to the field of power load prediction and optimization, is used for solving the problem of system stability caused by a nonlinear resonance effect, and comprises the following steps: preliminarily extracting key features of a power load signal through a multi-scale decomposition and mode recognition technology; and identifying a load mode which may cause system resonance. Two key indexes of frequency resonance intensity and time phase difference are emphatically utilized, the resonance risk of a load mode is comprehensively analyzed through a Bayesian network, the weight configuration of a high-risk mode is dynamically optimized in combination with a genetic algorithm, and the sensitivity and adaptability of the model to different resonance effects are improved. And then, the classified load modes are input into a deep neural network for multi-mode prediction fusion, and the model weight is adjusted in real time through a particle swarm optimization algorithm, so that the prediction precision and the operation stability of the system under a complex load condition are ensured, and a powerful guarantee is provided for the reliability of the power system.

Description

Technical Field

[0001] The present invention relates to the field of power load prediction and optimization, and more specifically, to a power load prediction and optimization method based on artificial intelligence. Background Art

[0002] Power load forecasting is one of the core tasks in the operation and management of power systems, and its accuracy directly affects the dispatching efficiency and operational safety of power systems. In power systems, the nonlinear dynamic behavior, seasonal fluctuations, and sudden changes of loads often lead to increased complexity in system operation, especially when dealing with nonlinear resonance effects, the stability of power systems is easily challenged. With the continuous growth of power demand and the increasing complexity of load characteristics, traditional forecasting methods are unable to capture and quantify these dynamic changes, and it is difficult to meet the requirements of modern power systems for high-precision load forecasting.

[0003] Although the existing technology has introduced artificial intelligence algorithms such as support vector machines and neural networks in power load forecasting, it still has significant shortcomings. Usually, only a single time series of historical load data is relied upon for forecasting, and the energy distribution of the load mode on different frequency components is not fully considered, resulting in a significant decrease in forecast accuracy when dealing with high-frequency resonance effects. At the same time, there is a lack of in-depth analysis of the phase differences between load modes, and the impact of phase synchronization or phase offset of load modes during critical periods on system stability is ignored. In addition, most existing technologies use fixed weight configurations and lack a dynamic adjustment mechanism that flexibly optimizes model parameters based on real-time risk assessment results, making both the forecasting ability and the system response ability insufficient in high-risk scenarios. These shortcomings make it difficult for existing methods to provide high-precision forecasts in complex power load environments and cannot effectively guarantee the stable operation of the system.

[0004] In order to solve the above problems, a technical solution is now provided. Summary of the invention

[0005] In order to overcome the above-mentioned defects of the prior art, the embodiment of the present invention provides an artificial intelligence-based power load prediction and optimization method, which aims to deal with complex nonlinear dynamic behaviors and potential resonance risks, and improve the prediction accuracy, stability and operation efficiency of the power system. First, through multi-scale decomposition and feature extraction technology, the power load signal is deeply analyzed, and its high-frequency and low-frequency change characteristics are fully captured, and a multi-dimensional feature vector matrix with high precision and dynamic adaptability is generated, which provides a data basis for subsequent nonlinear relationship detection. Secondly, the support vector machine is combined with the self-organizing map technology to successfully identify the implicit mode in the load signal, especially the key mode combination that may cause system resonance, which significantly enhances the responsiveness and robustness of the model under complex load conditions. Based on this, the two key indicators of resonance intensity in the frequency direction and mode phase difference in the time direction are further used to conduct a comprehensive resonance risk assessment, and the weight of the high-risk mode is dynamically optimized through the genetic algorithm to ensure the flexibility and accuracy of the model. Finally, the classified load modes of different risk levels are input into the deep neural network for multi-mode prediction fusion, and the mode weights are adjusted in real time through the particle swarm optimization algorithm to ensure that the model can provide the best prediction effect under different load change conditions. The method of the present invention not only optimizes the accuracy of the power load forecasting model, but also greatly improves the operating stability and safety of the power system in a complex environment, providing strong technical support for the long-term reliable operation of the system to solve the problems raised in the above background technology.

[0006] To achieve the above object, the present invention provides the following technical solutions:

[0007] S1, use wavelet transform to perform multi-scale decomposition of historical power load data and extract the load pattern characteristics of each frequency level; obtain the instantaneous amplitude and instantaneous frequency of the frequency component through Hilbert-Huang transform to generate a multi-dimensional feature vector.

[0008] S2, input the multi-dimensional feature vector into the support vector machine, identify the nonlinear mode in the load through the kernel method; use the self-organizing map to cluster the mode and identify the potential resonance mode combination.

[0009] S3, through the combination of two indicators, the resonance intensity in the frequency direction and the mode phase difference in the time direction, uses the Bayesian network to comprehensively analyze the resonance risk and classify the load pattern into high, medium and low risk levels; and uses the genetic algorithm to dynamically adjust the weight of the load pattern identified as high risk.

[0010] S4, based on the risk classification results, input the classified patterns into the deep neural network for multi-pattern prediction fusion, and adjust the pattern weights in real time through particle swarm optimization to optimize the model prediction accuracy.

[0011] In a preferred embodiment, step S1 includes the following contents:

[0012] S1.1, preprocessing historical power load data;

[0013] S1.2, the preprocessed power load data is input into the wavelet transform model, and the original load data is decomposed into frequency components of different scales using the multi-scale decomposition method to generate multiple frequency band signals corresponding to the load patterns on different time scales.

[0014] S1.3, taking the signals of each frequency band after wavelet decomposition as input, the Hilbert-Huang transform is used to analyze each frequency band signal; first, the empirical mode decomposition is applied to decompose the signal into intrinsic mode functions; then, the Hilbert transform is performed on each IMF to obtain the instantaneous amplitude and instantaneous frequency of the signal.

[0015] S1.4, the instantaneous amplitude and instantaneous frequency of each frequency band are combined with the original wavelet coefficient matrix to construct a multi-dimensional feature vector.

[0016] S1.5, storing the constructed multi-dimensional feature vector matrix and associating the corresponding timestamp and power load scenario.

[0017] In a preferred embodiment, step S2 includes the following contents:

[0018] S2.1, taking the stored multi-dimensional feature vector matrix as input data and performing preprocessing.

[0019] S2.2, the preprocessed multidimensional feature vector is input into the SVM model for training to optimize the decision boundary of the model so that it can distinguish different load modes.

[0020] S2.3. After the SVM model training is completed, the multidimensional feature vector is input into the trained model again for nonlinear pattern recognition.

[0021] S2.4. Take the nonlinear patterns identified by SVM as input and perform self-organizing map cluster analysis;

[0022] S2.5, storing the pattern features after cluster analysis in combination with the identified resonance patterns, and associating their corresponding timestamps and power load scenarios.

[0023] In a preferred embodiment, step S3 includes the following contents:

[0024] S3.1, based on the nonlinear modes identified in step S2, first extract the resonance intensity of each load mode in different frequency ranges.

[0025] S3.2, based on the identified resonance intensity, the pattern phase difference in the time direction is calculated;

[0026] S3.3, taking the phase difference matrix and the phase difference matrix as input variables, a Bayesian network model is constructed to determine the risk level of the nonlinear resonance effect.

[0027] According to the posterior probability distribution, the resonance risk level of the system is determined; the load pattern with the posterior probability greater than or equal to the high risk threshold is classified as a high risk level.

[0028] Load patterns whose posterior probabilities are not less than the high risk threshold and are greater than or equal to the low risk threshold are classified as medium risk levels.

[0029] Classify patterns with posterior probabilities less than the low-risk threshold as low-risk levels;

[0030] S3.4, based on the risk analysis results of the Bayesian network, dynamically adjust the weights of the identified high-risk load pattern combinations; use genetic algorithms to prioritize the weights of high-risk patterns.

[0031] S3.5, storing the optimized high-risk load pattern weights and associating them with corresponding timestamps and power load scenarios.

[0032] In a preferred embodiment, the complete steps for obtaining the resonance strength of each load mode in different frequency ranges are:

[0033] S3.1.1, first, convert the preprocessed power load signal x(t) into the frequency domain, and use Fourier Transform (FT) to obtain the frequency domain signal X(f), where f represents the frequency; construct the frequency range F as multiple discrete intervals F = {f1, f2, ..., f n}, where f i represents the ith frequency interval; the spectrum signal X(f i ):

[0034] S3.1.2, after obtaining the spectrum signal of each frequency interval, calculate the instantaneous energy density of each frequency interval; construct the instantaneous energy density E(f i ,t) is the square form of the signal energy at time t; the nonlinear conversion form of energy density is used for calculation, using the following formula: Among them, α, β, and γ are nonlinear conversion parameters, which are used to adjust the calculation method of instantaneous energy density to capture the dynamic changes of the signal in different frequency ranges.

[0035] S3.1.3, after calculating the instantaneous energy density of each frequency interval, extract the resonance factor R (f i ): Among them, θ(f i ,t) indicates that in the frequency interval f i The instantaneous phase angle on is obtained from the time domain signal through Hilbert transform.

[0036] S3.1.4, for each frequency interval, the resonance factor is represented by the original resonance intensity of the corresponding frequency interval, and the resonance factor is normalized; the normalized resonance intensity is marked as S(f i ).

[0037] S3.1.5, finally, the normalized resonance intensity of each frequency interval is organized into a resonance intensity matrix D(M), where each row of the matrix corresponds to a frequency interval and each column corresponds to the resonance intensity of a different time window.

[0038] In a preferred embodiment, the complete steps of obtaining the mode phase difference in the time direction are:

[0039] S3.2.1, first, the instantaneous phase information is extracted from the power load signal x(t) by Hilbert transform; for a given power load signal, its corresponding analytical signal z(t) is obtained by Hilbert transform, and the analytical signal is defined as: in, represents the Hilbert transform of x(t), j is the imaginary unit; the instantaneous phase θ(t) is calculated by analyzing the phase angle of the signal: θ(t) = θ(t) represents the phase change of the load signal at each moment, laying the foundation for the subsequent phase difference analysis.

[0040] S3.2.2, after obtaining the instantaneous phase; calculate the phase synchronization index to measure the phase synchronization degree between the two load patterns x1(t) and x2(t); the calculation formula of the phase synchronization index PLV is: Where N is the number of sampling points in the time window, θ1(t k ) and θ2(t k ) are two modes at time t k The instantaneous phase.

[0041] S3.2.3, after obtaining the phase synchronization index, calculate the phase difference metric to describe the directionality of the phase offset; the calculation formula of the phase difference metric PLI is: Where sign(·) represents the sign function, which is used to preserve the directionality of the phase shift, and λ is the nonlinear expansion parameter, which is used to adjust the sensitivity of the phase difference metric.

[0042] S3.2.4, integrate the phase difference measurement values ​​in each time window to calculate the overall phase difference intensity D(T), the formula is: Where T represents the total time interval [t0, t1], t c is the time center point, and ρ is the time attenuation factor, which is used to adjust the impact of different time periods on the overall phase difference intensity.

[0043] S3.2.5, finally, the overall phase difference intensity calculated in each time window is organized into a phase difference matrix S(F), where each row of the matrix corresponds to a load mode combination, and each column corresponds to the phase difference intensity in different time periods.

[0044] In a preferred embodiment, step S4 includes the following contents:

[0045] S4.1, the load patterns classified into high, medium, and low risk levels are input into the deep neural network and preprocessed before input.

[0046] S4.2, based on the preprocessed load pattern data, a deep neural network is used to perform multi-mode prediction fusion for high, medium, and low risk level patterns.

[0047] S4.3, based on the trained deep neural network, the particle swarm optimization algorithm is used to adjust the weight of the load mode in real time.

[0048] S4.4, after completing the weight adjustment, the deep neural network outputs the fused prediction results.

[0049] Technical effects and advantages of the power load forecasting and optimization method based on artificial intelligence of the present invention:

[0050] 1. By performing multi-scale decomposition and feature extraction on the power load data in step S1, not only can the high-frequency and low-frequency change characteristics in the load signal be fully captured, but also the key feature points with strong correlation with the power load pattern can be effectively identified. The multi-dimensional feature vector matrix constructed by this process has good time-frequency accuracy and dynamic adaptability, providing a high-quality data foundation for nonlinear relationship detection and identification of complex load patterns. With the help of an efficient data management system and a dynamic update mechanism, it can be ensured that the feature vector matrix is ​​continuously optimized with the introduction of new data, so that the power load forecasting model has stronger prediction accuracy and adaptability when facing future load changes, and ultimately improves the stability and operation efficiency of the power system.

[0051] 2. By using support vector machines (SVM) for nonlinear pattern recognition in step S2 and combining self-organizing maps (SOM) for cluster analysis, the nonlinear patterns implicit in the power load signal are successfully identified, especially those key pattern combinations that may cause system resonance. This process effectively improves the ability to identify complex load patterns, allowing the prediction model to more accurately reflect the dynamic behavior of the power system. By identifying these nonlinear structures, it not only provides accurate data support for subsequent resonance risk assessment and processing, but also improves the robustness and stability of the overall prediction model, ultimately enhancing the responsiveness and optimization effect of the power system in the face of nonlinear challenges.

[0052] 3. The beneficial effect of step S3 is that through the combined analysis of the two indicators of resonance intensity in the frequency direction and mode phase difference in the time direction, a comprehensive and accurate identification of the nonlinear resonance effect of the power system is achieved. First, through the resonance intensity analysis in the frequency direction, the energy accumulation of the power load signal in a specific frequency range can be effectively detected, revealing the potential resonance risk of the system at these frequencies, especially the high-frequency resonance problem. This analysis enables the system to warn of high energy points that may cause instability at an early stage, so as to take preventive measures in advance. Secondly, through the mode phase difference analysis in the time direction, the phase synchronization and offset between different load modes in the time domain can be captured, and the dynamic behavior of the system in the critical period can be accurately identified, especially the phase offset and coupling effect between different modes that may cause system imbalance or resonance. The combined analysis of the two not only expands the depth and breadth of risk identification, but also realizes the comprehensive assessment of complex nonlinear risks through synergy, avoiding the omissions and misjudgments that may be caused by a single analysis dimension. At the same time, through the dynamic adjustment of the weights of high-risk patterns by genetic algorithms, the model can respond to potential system risks more flexibly and accurately, thereby greatly improving the adaptability, robustness and overall prediction accuracy of the power load forecasting model, providing reliable guarantee for the stable operation of the power system.

[0053] 4. By classifying the load patterns into high, medium, and low risk levels in step S4 and inputting them into a deep neural network for multi-mode prediction fusion, the model's ability to identify and predict complex power load patterns is significantly improved. Prioritizing high-risk patterns ensures the system's response speed and accuracy when facing potential major risks, while moderate adjustments to medium and low-risk patterns maintain the overall balance of the model and avoid over-allocation of resources. The introduction of the particle swarm optimization algorithm enables the model to dynamically adjust the mode weights according to real-time data in actual operation, ensuring that the contribution of each mode to the prediction results is always in the best state in different time periods. This innovative processing method not only improves the accuracy and robustness of the prediction model, but also enhances the stability and safety of the power system in the face of complex load changes, providing strong support for the reliable operation of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 It is a flow chart of the power load forecasting and optimization method based on artificial intelligence of the present invention. DETAILED DESCRIPTION

[0055] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0056] Example 1

[0057] Figure 1 The present invention provides an artificial intelligence-based power load forecasting and optimization method, including:

[0058] S1, use wavelet transform to perform multi-scale decomposition of historical power load data and extract the load pattern characteristics of each frequency level; obtain the instantaneous amplitude and instantaneous frequency of the frequency component through Hilbert-Huang transform to generate a multi-dimensional feature vector.

[0059] S2, input the multi-dimensional feature vector into the support vector machine, identify the nonlinear mode in the load through the kernel method; use the self-organizing map to cluster the mode and identify the potential resonance mode combination.

[0060] S3, through the combination of two indicators, the resonance intensity in the frequency direction and the mode phase difference in the time direction, uses the Bayesian network to comprehensively analyze the resonance risk and classify the load pattern into high, medium and low risk levels; and uses the genetic algorithm to dynamically adjust the weight of the load pattern identified as high risk.

[0061] S4, based on the risk classification results, input the classified patterns into the deep neural network for multi-pattern prediction fusion, and adjust the pattern weights in real time through particle swarm optimization to optimize the model prediction accuracy.

[0062] In modern power systems, load forecasting and optimization are one of the core tasks to ensure system stability and efficient operation. With the continuous increase in power demand and the increasing complexity of load patterns, traditional forecasting methods have been unable to cope with the frequent nonlinear dynamic behaviors and potential resonance risks in power systems. These risks may not only lead to a reduction in system operating efficiency, but also cause serious stability problems, affecting the reliability of the entire power system. In order to overcome this challenge, the present invention proposes a power load forecasting and optimization method based on multi-scale analysis, nonlinear pattern recognition and comprehensive risk assessment. By performing multi-scale decomposition of historical power load data combining wavelet transform and Hilbert-Huang transform, the load pattern characteristics of each frequency level are extracted, and then a multidimensional feature vector is generated to capture the complex dynamic behavior in the power signal. On this basis, nonlinear pattern recognition and cluster analysis are performed on these feature vectors using support vector machine and self-organizing map technology to identify key pattern combinations that may cause resonance. Subsequently, the resonance intensity in the frequency direction and the pattern phase difference in the time direction are comprehensively analyzed by the Bayesian network, the potential resonance risk of the system is classified and evaluated, and the weights of the identified high-risk load patterns are dynamically adjusted using genetic algorithms. Finally, according to the risk classification results, these classified patterns are input into the deep neural network for multi-pattern prediction fusion, and the pattern weights are adjusted in real time through particle swarm optimization to ensure the accuracy of the prediction model and the operational stability of the power system. The method of the present invention achieves accurate prediction and optimization of power loads through innovative technical combinations, significantly improves the ability of the power system to cope with complex load changes and potential risks, and ensures the reliability and efficiency of the system in a complex operating environment.

[0063] In the process of power load forecasting and optimization, accurate extraction and analysis of key features in historical load data is the basis for ensuring that the forecasting model has high accuracy and high adaptability. In complex power systems, load changes on different time scales contain rich information. The distribution and mutual relationship of this information in the frequency domain and time domain directly affect the identification of load patterns and the detection of nonlinear relationships. By multi-scale decomposition of historical power load data and extracting its multi-dimensional feature vectors, key data support can be provided for subsequent nonlinear analysis and forecasting models, thereby improving the overall forecasting accuracy and model responsiveness.

[0064] Step S1 includes the following contents:

[0065] S1.1, preprocess the historical power load data. In the data preprocessing process, first determine the time range and data granularity of the historical data to be processed according to the specific needs of power load forecasting, screen out the data related to the current analysis target, and exclude irrelevant historical data. Then, within the selected data range, use the three sigma (3σ) criterion or statistical methods such as box plots to detect abnormal data, remove obviously erroneous data points, or use linear interpolation algorithms to replace abnormal data. Prioritize the load data during special periods such as peak hours, holidays or extreme weather, and use more stringent noise filtering and high-precision interpolation algorithms to ensure that these data with important reference value are processed more carefully. After that, the screened and processed data are normalized by Min-Max or Z-score, and the data are uniformly adjusted to the same scale range to eliminate the differences between different time periods or dimensions, and ensure that the model has consistent sensitivity to different features. Finally, the preprocessed data is stored in layers and managed in layers according to the time range, load characteristics and data quality, so that it can be flexibly called in subsequent analysis, and the priority of the data can be dynamically adjusted to meet the needs of different analysis tasks.

[0066] S1.2, input the preprocessed power load data into the wavelet transform model, and select the mother wavelet function suitable for the power load characteristics, such as Daubechies wavelet (db4). Using the multi-scale decomposition method, the original load data is decomposed into frequency components of different scales, and multiple frequency band signals are generated, corresponding to the load patterns on different time scales. Through the wavelet coefficient matrix, the load pattern characteristics of each frequency band are extracted, the high-frequency and low-frequency information at each scale is recorded, and the key load change characteristics are retained.

[0067] The key load change characteristics can be defined as the dynamic behaviors of high-frequency and low-frequency components of the power load signal at different frequency levels after multi-scale decomposition, including the sharp fluctuations of instantaneous amplitude, abnormal deviations of frequency changes, and characteristic points that are strongly associated with the load pattern in a specific period of time. By analyzing the energy distribution of these high-frequency and low-frequency information, the cross-characteristics of the time domain and the frequency domain, and their phase consistency in different periods of time, the characteristics that can significantly affect the overall load pattern are extracted. These characteristics include but are not limited to the energy aggregation effect occurring at a specific frequency, the load mutation peak caused by external factors, the changing trend of periodic fluctuations, and the phase difference reflecting the system load response speed. The key load change characteristics defined in this way can fully reflect the sensitivity and response capability of the power system to various load changes at different time scales, and provide a more accurate and practical data basis for subsequent nonlinear relationship detection and load forecasting.

[0068] S1.3, take the frequency band signals after wavelet decomposition as input, and use Hilbert-Huang Transform (HHT) to further analyze each frequency band signal. First, apply Empirical Mode Decomposition (EMD) to decompose the signal into Intrinsic Mode Functions (IMFs), ensuring that each IMF meets the definition of instantaneous frequency. Then, perform Hilbert transform on each IMF to obtain the instantaneous amplitude and instantaneous frequency of the signal, forming a dynamic feature description of the frequency component.

[0069] S1.4, the instantaneous amplitude and instantaneous frequency of each frequency band are combined with the original wavelet coefficient matrix to construct a multidimensional feature vector. This feature vector not only contains the frequency characteristics of the load signal, but also includes its changing trend on the time scale. By classifying and standardizing the instantaneous characteristics of different frequency components, the final multidimensional feature vector matrix is ​​generated as the input data for subsequent nonlinear relationship detection.

[0070] S1.5, store the constructed multi-dimensional feature vector matrix and associate the corresponding timestamps and power load scenarios. By building an efficient data management system, ensure that these feature vectors can be quickly extracted and called during the nonlinear relationship detection and pattern recognition process. In order to cope with future changes in power load, set up a dynamic update mechanism so that the feature vector matrix can be updated and optimized in real time as new data is introduced.

[0071] Through the above steps, it is ensured that the features extracted from the historical power load data have high time-frequency accuracy and flexibility, providing a solid data foundation for the accurate detection of nonlinear relationships.

[0072] By performing multi-scale decomposition and feature extraction on the power load data in step S1, not only can the high-frequency and low-frequency change characteristics in the load signal be fully captured, but also the key feature points with strong correlation with the power load pattern can be effectively identified. The multi-dimensional feature vector matrix constructed by this process has good time-frequency accuracy and dynamic adaptability, providing a high-quality data foundation for nonlinear relationship detection and identification of complex load patterns. With the help of an efficient data management system and a dynamic update mechanism, it can be ensured that the feature vector matrix is ​​continuously optimized with the introduction of new data, so that the power load forecasting model has stronger prediction accuracy and adaptability when facing future load changes, and ultimately improves the stability and operation efficiency of the power system.

[0073] In the process of power load forecasting and optimization, the nonlinear patterns hidden in the power load signal have an important impact on the stability and prediction accuracy of the system. With the increasing complexity of the power system, traditional linear analysis methods are difficult to capture these complex nonlinear relationships, resulting in deviations in prediction results and reduced optimization effects. By combining support vector machines (SVM) with kernel methods for nonlinear pattern recognition and using self-organizing maps (SOM) for pattern clustering analysis, the complex dynamic characteristics in the load signal can be accurately revealed, especially those patterns that are closely related to the risk of system resonance, which provides an important basis for subsequent resonance effect prediction and response strategies.

[0074] Step S2 includes the following contents:

[0075] S2.1, taking the stored multi-dimensional feature vector matrix as input data, first preprocess the feature vector to ensure that it is suitable for training and prediction of the Support Vector Machine (SVM) model. The specific preprocessing includes standardizing the feature vector, eliminating the scale differences between different features, and using the PCA (principal component analysis) method to reduce the dimension, retain the main information in the feature vector, reduce data redundancy, and improve calculation efficiency.

[0076] S2.2, select the appropriate kernel method according to the complexity and nonlinear characteristics of the power load characteristics. Use kernel functions such as radial basis function (RBF) or polynomial kernel (Polynomial Kernel) that can capture nonlinear relationships to build the SVM model. Determine the optimal kernel parameters and regularization parameters through cross-validation to ensure that the model can accurately identify nonlinear patterns in load data. Input the preprocessed multidimensional feature vector into the SVM model for training to optimize the decision boundary of the model so that it can distinguish different load patterns.

[0077] S2.3. After the SVM model training is completed, the multidimensional feature vector is input into the trained model again for nonlinear pattern recognition. The model classifies the input data based on the decision boundary formed during training and identifies different pattern characteristics in the power load, especially those load patterns with nonlinear relationships. By identifying these patterns, the implicit nonlinear structure in the power load signal can be revealed, providing a basis for subsequent pattern analysis.

[0078] When performing nonlinear pattern recognition, the preprocessed multidimensional feature vector is input into the trained support vector machine (SVM) model, and the model classifies the input data one by one according to the optimal decision boundary constructed during the training process. During the classification process, the model maps the feature vector through the kernel method, converting the originally complex high-dimensional spatial data into a higher-dimensional feature space, so that different patterns can be more easily separated in the new space. Specifically, SVM uses its high-precision boundary division ability to identify different power load pattern characteristics hidden in the input data, especially those nonlinear patterns that are difficult to capture by linear methods. These nonlinear patterns usually reflect the complex dynamic behaviors in the power load signal, such as sudden changes in load, nonlinear enhancement of periodic fluctuations, and nonlinear resonance caused by coupling with external factors. Through the accurate identification of these patterns, the potential implicit nonlinear structures in the power system can be revealed. These structures have important reference value in subsequent pattern analysis and resonance risk assessment, providing more detailed and accurate data support for power load prediction and optimization.

[0079] S2.4. The nonlinear patterns identified by SVM are used as input for self-organizing map (SOM) clustering analysis. The SOM model performs topological mapping of pattern features through unsupervised learning and clusters similar patterns into the same category. By analyzing the clustering results, pattern combinations with potential resonance effects are identified, which may cause nonlinear resonance of the power system under certain conditions.

[0080] S2.5, store the pattern features after cluster analysis and the identified resonance pattern combination, and associate their corresponding timestamps and power load scenarios so that they can be quickly called in subsequent resonance risk prediction. At the same time, establish a dynamic update mechanism for pattern features. With the introduction of new data and retraining of the model, the definition and classification of the resonance pattern combination are updated in real time to adapt to the dynamic changes of power load.

[0081] Through the above steps, it is ensured that the nonlinear patterns in the power load are accurately identified from the multidimensional feature vector, and the potential resonance pattern combinations are revealed through cluster analysis, providing solid data support for subsequent resonance risk prediction and response strategies.

[0082] By using support vector machines (SVM) for nonlinear pattern recognition in step S2 and combining self-organizing maps (SOM) for cluster analysis, the nonlinear patterns implicit in the power load signal are successfully identified, especially those key pattern combinations that may cause system resonance. This process effectively improves the ability to identify complex load patterns, allowing the prediction model to more accurately reflect the dynamic behavior of the power system. By identifying these nonlinear structures, it not only provides accurate data support for subsequent resonance risk assessment and processing, but also improves the robustness and stability of the overall prediction model, ultimately enhancing the responsiveness and optimization effect of the power system in the face of nonlinear challenges.

[0083] Step S3 includes the following contents:

[0084] In the process of power load forecasting and optimization, identifying and quantifying the resonance intensity of load patterns in different frequency ranges is crucial to understanding the dynamic behavior and potential risks of power systems. Nonlinear resonance effects in power systems are often concentrated in specific frequency ranges. These resonance effects may cause abnormal fluctuations in the system and affect the stability of power supply. By analyzing the resonance intensity of each load pattern in different frequency ranges in detail, the key dynamic characteristics in the load signal can be deeply revealed, providing key data support for the stable operation of the power system, and providing a strong basis for the optimization of the prediction model and early warning of resonance risks.

[0085] In risk identification, the reason for selecting the combination analysis of the resonance intensity in the frequency direction and the mode phase difference in the time direction is that these two indicators can fully capture the nonlinear dynamic behavior in the power system, thereby realizing the accurate identification of potential resonance risks. The resonance intensity in the frequency direction can reveal the resonance effect that may occur in the system at a specific frequency by analyzing the energy aggregation of the power load signal in different frequency intervals. It is especially suitable for identifying high-frequency resonance problems caused by external disturbances or system instability. The mode phase difference in the time direction can effectively identify the coordination or disorder of the system in a specific period of time by measuring the phase synchronization and phase offset between different load modes. Especially when there is a complex coupling relationship between load modes, this indicator can reflect the dynamic instability of the system in the time domain and the potential nonlinear resonance risk. The combined analysis of the two indicators can not only fully capture the dynamic behavior of the system from the two dimensions of frequency domain and time domain, but also more accurately identify those complex risk patterns that are difficult to detect with a single indicator through their synergy, thereby providing more powerful support for the stability analysis and risk warning of the system.

[0086] S3.1, based on the nonlinear mode identified in step S2, first extract the resonance intensity of each load mode in different frequency ranges. The resonance intensity is calculated based on the change in energy density of each frequency component within a specific time window. The load signal is converted to the frequency domain through Fourier transform, the energy distribution of different frequency bands is analyzed, and high energy points where resonance effects may occur in these frequency bands are identified. These high energy points are quantitatively analyzed to generate a resonance intensity index. The specific calculation includes the time-varying characteristics of energy density and its average and standard deviation over multiple time periods, so as to accurately reflect the resonance risk at different frequencies.

[0087] Complete steps to obtain the resonance strength of each load mode in different frequency ranges:

[0088] S3.1.1, first, convert the preprocessed power load signal x(t) into the frequency domain, and use Fourier Transform (FT) to obtain the frequency domain signal X(f), where f represents the frequency. Construct the frequency range F as multiple discrete intervals F = {f1, f2, ..., f n}, where f i Represents the ith frequency interval. By performing Fourier transform on the power load signal, the spectrum signal X(f i ):

[0089] S3.1.2, after obtaining the spectrum signal of each frequency interval, calculate the instantaneous energy density of each frequency interval. Construct the instantaneous energy density E(f i ,t) is the square form of the signal energy at time t. Considering the nonlinear characteristics of the power load signal, the nonlinear conversion form of energy density is used for calculation, using the following formula: Among them, α, β, and γ are nonlinear conversion parameters, which are used to adjust the calculation method of instantaneous energy density to capture the dynamic changes of the signal in different frequency ranges. i ) represents the frequency interval f i The corresponding signal amplitude.

[0090] S3.1.3, after calculating the instantaneous energy density of each frequency interval, further extract the resonance factor R (f i ). The resonance factor reflects the degree of signal energy concentration in the frequency range and its impact on the load mode. The formula for calculating the resonance factor is: Among them, θ(f i ,t) indicates that in the frequency interval f iThe instantaneous phase angle on the signal is obtained from the time domain signal through Hilbert transform. The integration range [t0, t1] is a specific time period. The purpose of selecting this period is to focus on the key time window where the resonance effect occurs. By introducing the phase angle, the resonance factor can better reflect the resonance characteristics of the signal in this frequency range.

[0091] S3.1.4, for each frequency interval, the original resonance intensity of the corresponding frequency interval is represented by the resonance factor. In order to unify the resonance intensity representation between different frequency intervals, the resonance factor needs to be normalized. The normalized resonance intensity S(f i ) is calculated as: Among them, δ is a normalization parameter, which is used to adjust the relative weights between the resonance factors in different frequency intervals to ensure that the normalized resonance intensity is in the range of 0 to 1. In this way, the energy differences in different frequency intervals are eliminated, which facilitates the unified processing of subsequent analysis.

[0092] S3.1.5, finally, the normalized resonance intensity of each frequency interval is organized into a resonance intensity matrix D(M), where each row of the matrix corresponds to a frequency interval and each column corresponds to the resonance intensity of a different time window. This matrix constitutes a comprehensive description of the resonance effect of the power load signal in different frequency ranges, providing basic data for subsequent Bayesian network risk analysis and dynamic adjustment of load patterns.

[0093] Through the above steps, the resonance intensity of the power load mode can be accurately obtained in different frequency ranges, thereby providing reliable data support and analysis basis for the stable operation of the power system.

[0094] By accurately calculating and analyzing the resonance intensity of each load mode in different frequency ranges, the potential resonance effect in the power system and the dynamic instability it may cause can be effectively identified. This process not only helps to understand the complex dynamic behavior in the power load signal, but also provides high-precision input data for subsequent risk assessment and prediction models through the normalized resonance intensity matrix. Ultimately, using these resonance intensity data, the adaptability and responsiveness of the power load prediction model to nonlinear resonance effects can be significantly improved, ensuring that the system can maintain a stable operating state in a complex and changing power environment, and improving the reliability of overall prediction and optimization.

[0095] In the process of power load forecasting and optimization, the phase synchronization and phase difference between load modes are important factors affecting system stability. The phase offset of different modes in the power load signal, especially the phase synchronization or misalignment in a specific time period, may cause nonlinear resonance effects in the system, thereby causing abnormal fluctuations or failures. Therefore, accurately obtaining and quantifying the phase difference of power load patterns in the time direction is of great significance for identifying potential system risks, improving the adaptability of prediction models, and optimizing system operation. By accurately analyzing the phase difference of load patterns in the time domain, we can deeply understand the dynamic behavior of the system and provide strong support for the safe and stable operation of the power system.

[0096] S3.2, based on the identification of resonance intensity, further calculate the mode phase difference in the time direction. Mode phase difference refers to the phase synchronization or difference between different load modes in the time domain, especially focusing on those mode combinations that have phase offsets at critical moments. The instantaneous phase information of the load signal is obtained by Hilbert transform, and the phase difference between different modes is quantitatively evaluated through indicators such as Phase Locking Value (PLV) or Phase Lag Index (PLI). This indicator reflects the synergy or opposition relationship of load modes in different time periods and is an important basis for measuring the stability of the system in the time dimension.

[0097] Complete steps to obtain the mode phase difference in the time direction:

[0098] S3.2.1, first, extract the instantaneous phase information from the power load signal x(t) through Hilbert Transform. For a given power load signal, obtain its corresponding analytical signal z(t) through Hilbert Transform. The analytical signal is defined as: in, represents the Hilbert transform of x(t), j is the imaginary unit. The instantaneous phase θ(t) is calculated by analyzing the phase angle of the signal: θ(t) represents the phase change of the load signal at each moment, laying the foundation for the subsequent phase difference analysis.

[0099] S3.2.2, after obtaining the instantaneous phase, consider the phase synchronization between different load modes. Calculate the phase locking value (PLV) to measure the phase synchronization degree between the two load modes x1(t and x2(t). The calculation formula of the phase locking value PLV is: Where N is the number of sampling points in the time window, θ1(t k ) and θ2(tk ) are two modes at time t k The instantaneous phase of the two modes. Phase synchronization refers to a value range of 0 to 1. The closer the value is to 1, the higher the phase synchronization between the two modes, and vice versa.

[0100] S3.2.3, after obtaining the phase synchronization index, in order to further quantify the phase difference between different modes, the phase difference metric (Phase Lag Index, PLI) is calculated to describe the directionality of the phase offset. The calculation formula of the phase difference metric PLI is: Where sign(·) represents the sign function, which is used to preserve the directionality of the phase shift, and λ is a nonlinear expansion parameter, which is used to adjust the sensitivity of the phase difference metric. The phase difference metric can reflect the frequency of the phase shift between modes and the degree of its aggregation in a specific direction.

[0101] S3.2.4, in order to quantify the overall phase difference of the mode in the time direction, the phase difference measurement value in each time window is integrated to calculate the overall phase difference intensity D(T), the formula is: Where T represents the total time interval [t0, t1], t c is the time center point, and ρ is the time decay factor, which is used to adjust the influence of different time periods on the overall phase difference intensity. This makes the phase difference measurement value close to the center time contribute more to the overall phase difference intensity, and conversely, the contribution away from the center time is reduced.

[0102] S3.2.5, finally, the overall phase difference intensity calculated in each time window is organized into a phase difference matrix S(F), where each row of the matrix corresponds to a load pattern combination, and each column corresponds to the phase difference intensity in different time periods. The phase difference matrix provides a comprehensive description of the phase difference of different load patterns in the time direction, providing key data support for subsequent Bayesian network analysis and dynamic adjustment of load pattern weights.

[0103] Through the above steps, the phase difference of the power load mode in the time direction can be accurately obtained, the phase synchronization and offset between different modes can be quantified, and a scientific basis can be provided for the stability analysis and prediction of the system.

[0104] By analyzing the phase differences of power load patterns in the time direction in detail, the phase synchronization and offset between different patterns can be effectively identified, revealing the potential dynamic instability factors in the power system. This process not only helps to identify high-risk pattern combinations that are prone to resonance effects, but also provides accurate data support for subsequent Bayesian network analysis and load pattern weight adjustment. By constructing a phase difference matrix, the phase dynamics of the system in different time periods can be fully reflected, providing more detailed and accurate input data for the prediction model, ultimately improving the model's response to nonlinear resonance effects and improving the operational stability and safety of the power system under complex load conditions.

[0105] In power systems, nonlinear resonance effects are often hidden risks that lead to system instability and failure. With the increasing complexity and dynamic changes of power loads, traditional linear analysis methods are difficult to fully capture the potential nonlinear correlations and risk factors in the system. As a powerful probabilistic graphical model, Bayesian networks can achieve comprehensive risk analysis of complex systems by modeling the causal relationship between different variables. Using the phase difference matrix and the resonance intensity matrix as input variables to construct a Bayesian network model, the dynamic characteristics of power load patterns can be deeply analyzed from the two dimensions of time and frequency, and the risks that may cause system resonance can be quantified and warned. This analysis method provides new perspectives and tools for the operation optimization and risk management of power systems, helping to improve the stability and safety of the system.

[0106] S3.3, taking the phase difference matrix and the phase difference matrix as input variables, constructing a Bayesian Network model, and conducting a comprehensive risk analysis of the nonlinear resonance effect. The Bayesian Network constructs a probability relationship diagram between variables, combines historical data and real-time data, and calculates the probability of occurrence of different resonance mode combinations and their impact on system stability. According to the resonance intensity in the frequency direction and the mode phase difference in the time direction, the Bayesian network can dynamically adjust the conditional probability distribution of each node, output the risk level of each resonance mode combination, and identify the potential resonance effect with the greatest impact on the power system. When the posterior probability exceeds the corresponding threshold, it can be judged that the current load mode combination has a large resonance risk and outputs a high-risk signal.

[0107] S3.3.1, the phase difference matrix and the resonance intensity matrix are used as input variables for constructing the Bayesian network model. These two matrices reflect the dynamic characteristics of different load pattern combinations in the time and frequency directions. For each load pattern combination, its phase difference D in a specific time period is extracted. i and the resonance intensity S j , as the value of the input node. To ensure the modeling effect of the Bayesian network, the data in these two matrices need to be preprocessed.

[0108] Data discretization: Since Bayesian networks usually process discrete data, the continuous values ​​of phase difference and resonance intensity need to be discretized. They can be divided into several intervals according to the range of values, such as low, medium, and high risk intervals. The number of discretized intervals and the division method can be adjusted according to actual needs and data distribution.

[0109] Variable combination: For each load pattern combination, form the input variable V k , which is in the form of V k =(D i ,S j ). The joint variable contains the dynamic characteristics of the load pattern in both time and frequency directions and is the input node of the Bayesian network.

[0110] S3.3.2, based on the physical meaning of the power load pattern and known system characteristics, construct the structure of the Bayesian network. The nodes in the network include the input variables V k and system status nodes (such as system stability, resonance risk level, etc.). The directed edges between nodes represent the causal relationship between variables, and the structural design can refer to expert knowledge or be automatically generated through structural learning algorithms.

[0111] Node definition: Construct a system status node to represent the resonance risk level of the power system. Possible values ​​include low risk, medium risk, and high risk. The input node is the parent node of the system status node and affects the resonance risk level of the system.

[0112] Causality: In a Bayesian network, input nodes are connected to system state nodes through directed edges of causal relationships. Different combinations of input variables will affect the value of the system state node with a certain probability.

[0113] S3.3.3, define a conditional probability table (CPT) for each Bayesian network node to describe the probability of different values ​​of the input node affecting the system state node. The conditional probability table can be determined by statistical analysis of historical data.

[0114] Historical data training: Use the historical operation data of the power system to count the frequency of the impact of different phase differences and resonance strengths on the system resonance risk. Determine the conditional probability table of each node through the maximum likelihood estimation (MLE) or Bayesian estimation method.

[0115] S3.3.4, after completing the construction of the conditional probability table, the Bayesian network can perform reasoning analysis. After inputting the phase difference and resonance strength of different load mode combinations, the posterior probability distribution of the system state node is calculated using a Bayesian reasoning algorithm (such as a belief propagation algorithm or a variational reasoning algorithm).

[0116] Risk level determination: Determine the resonance risk level of the system based on the posterior probability distribution. Classify load patterns with a posterior probability greater than or equal to the high risk threshold as high risk levels, indicating that these patterns may cause serious system instability or resonance effects in the current or future time period. Classify load patterns with a posterior probability not less than the high risk threshold and greater than or equal to the low risk threshold as medium risk levels, indicating that these patterns may cause a certain degree of fluctuation in the system, but not enough to cause major failures or instability. Finally, patterns with a posterior probability less than the low risk threshold are classified as low risk levels. Such patterns have little impact on the system under current conditions and are within a safe and controllable range. Through this classification method, it is ensured that targeted response strategies can be adopted for different load patterns according to the risk level in the subsequent processing process, thereby optimizing the operational safety and stability of the power system.

[0117] Risk analysis report: Outputs comprehensive risk analysis results for each load pattern combination, including the probability distribution of each risk level and its possible impact. These results can be used for further decision support, such as dynamic adjustment of system operating parameters or early warning systems.

[0118] Through the comprehensive risk analysis of Bayesian networks, it is not only possible to quantify and identify the potential resonance risks of various load mode combinations in the power system, but also to dynamically adjust and optimize the system's operating strategy. The Bayesian network model can effectively integrate multi-dimensional information from the phase difference matrix and the resonance intensity matrix, and use conditional probability reasoning to accurately predict the risk level under different load modes. This process significantly improves the system's responsiveness to complex dynamic behaviors, helping managers identify high-risk modes in advance and take preventive measures to avoid potential system instability or failures. Ultimately, the comprehensive risk analysis of Bayesian networks helps to improve the accuracy and adaptability of power load forecasting models, enhance the operational stability of power systems in complex environments, and provide important guarantees for the long-term reliable operation of power systems.

[0119] S3.4, based on the risk analysis results of the Bayesian network, focus on dynamically adjusting the weights of the identified high-risk load pattern combinations. Use the Genetic Algorithm (GA) to prioritize the weight optimization of high-risk patterns. The genetic algorithm increases the weight sensitivity of high-risk patterns through operations such as selection, crossover, and mutation, so that the model has a higher response speed and prediction accuracy when dealing with these potential resonance effects. The weight adjustment of medium and low-risk patterns will remain basically stable, with only fine-tuning to maintain the overall balance and robustness of the model. In this way, it is ensured that the model can concentrate resources and computing power to prioritize high-risk patterns that may have a significant impact on system stability.

[0120] The specific processing process of dynamic adjustment of load mode weights includes the following technical features: First, the high-risk load mode combination identified by the Bayesian network is used as input to optimize the weight configuration of the model through the genetic algorithm. The selection operation in the genetic algorithm selects the best performing weight combination to enter the next generation based on the performance of the high-risk mode in the fitness function; the crossover operation exchanges some genes of the two weight combinations to generate new weight combinations and increase the diversity of the search space; the mutation operation randomly adjusts some weight values ​​to prevent falling into the local optimum and ensure global search capabilities. The fitness function is evaluated based on the prediction accuracy and response speed of the model when dealing with high-risk modes, and optimized generation by generation, so that the weight of the high-risk mode is reasonably improved. The optimized weight parameters are directly updated to the model and applied to the system's load forecasting and resonance risk assessment in real time to ensure that the model has stronger response capabilities and prediction accuracy in high-risk scenarios. At the same time, by introducing an adaptive mutation rate, the intensity of the mutation operation is dynamically adjusted to balance the sensitivity and robustness of the model, so that it can maintain stability and flexibility when dealing with complex load changes.

[0121] S3.5, the optimized high-risk load mode weights are stored and associated with the corresponding timestamps and power load scenarios for quick call in future predictions and real-time applications. At the same time, a feedback mechanism for high-risk modes is established to compare and analyze the actual operation data with the high-risk prediction results of the Bayesian network. If it is confirmed in actual operation that a certain high-risk mode does indeed cause a resonance effect or other unstable phenomena, the model will further strengthen the weight adjustment of the mode to ensure a more sensitive and accurate response to similar situations in the future. Conversely, if some high-risk predictions do not show the expected severity in practice, the model will moderately call back the weights of these modes to prevent false alarms or waste of resources caused by excessive sensitivity. Through this feedback mechanism, the model can continuously optimize its prediction capabilities in actual operation to ensure the long-term stability and reliability of the power system.

[0122] The beneficial effect of step S3 is that through the combined analysis of the two indicators of resonance intensity in the frequency direction and mode phase difference in the time direction, a comprehensive and accurate identification of the nonlinear resonance effect of the power system is achieved. First, through the resonance intensity analysis in the frequency direction, the energy accumulation of the power load signal in a specific frequency range can be effectively detected, revealing the potential resonance risk of the system at these frequencies, especially the high-frequency resonance problem. This analysis enables the system to warn of high energy points that may cause instability at an early stage, so as to take preventive measures in advance. Secondly, through the mode phase difference analysis in the time direction, the phase synchronization and offset between different load modes in the time domain can be captured, and the dynamic behavior of the system in the critical period can be accurately identified, especially the phase offset and coupling effect between different modes that may cause system imbalance or resonance. The combined analysis of the two not only expands the depth and breadth of risk identification, but also realizes the comprehensive assessment of complex nonlinear risks through synergy, avoiding the omissions and misjudgments that may be caused by a single analysis dimension. At the same time, through the dynamic adjustment of the weights of high-risk patterns by genetic algorithms, the model can respond to potential system risks more flexibly and accurately, thereby greatly improving the adaptability, robustness and overall prediction accuracy of the power load forecasting model, providing reliable guarantee for the stable operation of the power system.

[0123] In the process of power load forecasting and optimization, the risk level of different load modes directly affects the stability and responsiveness of the system. The traditional single forecasting model is difficult to fully capture the complex dynamic behaviors under different risk levels, resulting in a decrease in forecasting accuracy and an increase in system risk. By dividing the load modes into high, medium, and low risk levels according to the resonance risk assessment results, and using deep neural networks for multi-mode forecasting fusion, the characteristic information of various modes can be effectively integrated to improve the adaptability of the forecasting model to complex load changes. Further combined with the particle swarm optimization algorithm to adjust the mode weights in real time, the model can flexibly respond to risk changes in different time periods, ensuring high accuracy of the forecast results and high stability of system operation.

[0124] Step S4 includes the following contents:

[0125] S4.1, input the load patterns classified into high, medium and low risk levels into the Deep Neural Network (DNN), and perform necessary preprocessing before input. Preprocessing includes normalizing the input data to eliminate the dimensional differences between different patterns, and screening out the key features that have the greatest impact on the prediction results through feature selection methods, reducing the redundancy of input data and improving the training efficiency of the network. For high-risk patterns, higher input weights are used to increase their influence in the prediction, ensuring that the model is more sensitive to these patterns.

[0126] S4.2, based on the preprocessed load pattern data, a deep neural network is used to perform multi-mode prediction fusion for high, medium and low risk level patterns. During the training process, a multi-layer network structure is set up, and the backpropagation algorithm is used to optimize the network weights and bias parameters to minimize the prediction error. Due to its significant impact on system stability, high-risk patterns are given a higher learning rate in training to accelerate the optimization process of their weights. The network generates a unified prediction output covering all possible system states by fusing the diverse features of patterns at different risk levels.

[0127] S4.3, based on the trained deep neural network, the particle swarm optimization algorithm (PSO) is used to adjust the weights of the load patterns in real time. The particle swarm optimization algorithm simulates the movement of particles in the search space to find the optimal weight configuration. Each particle represents a possible weight combination, and the fitness of the particle is determined by the prediction accuracy of the deep neural network. During the search process, the particles update their speed and position based on their own experience and the optimal position of the group to gradually approach the optimal solution. For high-risk modes, the particle swarm optimization algorithm dynamically adjusts the weights to ensure that the prediction accuracy of these modes in different time periods is optimal; for medium and low-risk modes, the particle swarm optimization maintains its basic accuracy while moderately lowering its weight to maintain the overall balance of the model.

[0128] S4.4, after completing the weight adjustment, the deep neural network outputs the fused prediction results. To verify the accuracy of the model, the prediction results are compared and analyzed with the actual load data to identify possible sources of error, and the particle swarm optimization process is fine-tuned based on the feedback of the results to further improve the accuracy and reliability of the model. Special attention will be paid to the prediction results of high-risk modes to ensure that the system can respond and warn in time under these modes. Finally, through this series of steps, an optimized, multi-mode fusion load forecasting model is generated with high accuracy, high adaptability and flexible processing capabilities for modes with different risk levels.

[0129] By classifying the load patterns into high, medium, and low risk levels in step S4 and inputting them into a deep neural network for multi-mode prediction fusion, the model's ability to identify and predict complex power load patterns is significantly improved. The priority treatment of high-risk patterns ensures the system's response speed and accuracy when facing potential major risks, while the moderate adjustment of medium and low-risk patterns maintains the overall balance of the model and avoids over-allocation of resources. The introduction of the particle swarm optimization algorithm enables the model to dynamically adjust the mode weights according to real-time data in actual operation, ensuring that the contribution of each mode to the prediction results is always in the best state in different time periods. It not only improves the accuracy and robustness of the prediction model, but also enhances the stability and safety of the power system in the face of complex load changes, providing strong support for the reliable operation of the system.

[0130] The above formulas are all dimensionless and numerical calculations. The formula is a formula for the most recent real situation obtained by collecting a large amount of data and performing software simulation. The preset parameters in the formula are set by technicians in this field according to actual conditions.

[0131] The above description is only by way of illustration of certain exemplary embodiments of the present invention. It is undoubted that those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the above drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.

[0132] It should be noted that, in this article, if there are relational terms such as first and second, etc., they are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "including a..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.

[0133] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. The power load forecasting and optimization method based on artificial intelligence is characterized by: Includes steps: S1, using wavelet transform to perform multi-scale decomposition of historical power load data, extracting the load pattern characteristics of each frequency level; obtaining the instantaneous amplitude and instantaneous frequency of the frequency component through Hilbert-Huang transform, and generating a multi-dimensional feature vector; S2, input the multi-dimensional feature vector into the support vector machine, identify the nonlinear mode in the load through the kernel method; cluster the mode using self-organizing map to identify the potential resonance mode combination; S3, through the combination of two indicators, the resonance intensity in the frequency direction and the mode phase difference in the time direction, the resonance risk is comprehensively analyzed using the Bayesian network, and the load patterns are classified into high, medium and low risk levels; and the weights of the load patterns identified as high-risk are dynamically adjusted using the genetic algorithm; S4, based on the risk classification results, input the classified patterns into the deep neural network for multi-pattern prediction fusion, and adjust the pattern weights in real time through particle swarm optimization to optimize the model prediction accuracy.

2. The method for power load forecasting and optimization based on artificial intelligence according to claim 1, characterized in that: Step S1 includes the following contents: S1.1, preprocessing historical power load data; S1.2, input the preprocessed power load data into the wavelet transform model, and use the multi-scale decomposition method to decompose the original load data into frequency components of different scales to generate multiple frequency band signals corresponding to the load patterns on different time scales; S1.3, taking the signals of each frequency band after wavelet decomposition as input, the Hilbert-Huang transform is used to analyze the signals of each frequency band; first, the empirical mode decomposition is applied to decompose the signal into intrinsic mode functions; then, the Hilbert transform is performed on each IMF to obtain the instantaneous amplitude and instantaneous frequency of the signal; S1.4, combining the instantaneous amplitude and instantaneous frequency of each frequency band with the original wavelet coefficient matrix to construct a multidimensional feature vector; S1.5, storing the constructed multi-dimensional feature vector matrix and associating the corresponding timestamp and power load scenario.

3. The method for predicting and optimizing power load based on artificial intelligence according to claim 2, characterized in that: Step S2 includes the following contents: S2.1, taking the stored multi-dimensional feature vector matrix as input data and performing preprocessing; S2.2, input the preprocessed multidimensional feature vector into the SVM model for training, and optimize the decision boundary of the model so that it can distinguish different load modes; S2.

3. After the SVM model training is completed, the multidimensional feature vector is input into the trained model again to perform nonlinear pattern recognition; S2.

4. Take the nonlinear patterns identified by SVM as input and perform self-organizing map cluster analysis; S2.5, storing the pattern features after cluster analysis in combination with the identified resonance patterns, and associating their corresponding timestamps and power load scenarios.

4. The method for predicting and optimizing power load based on artificial intelligence according to claim 3 is characterized in that: Step S3 includes the following contents: S3.1, based on the nonlinear modes identified in step S2, first extract the resonance intensity of each load mode in different frequency ranges; S3.2, based on the identified resonance intensity, the pattern phase difference in the time direction is calculated; S3.3, using the phase difference matrix and the phase difference matrix as input variables, a Bayesian network model is constructed to determine the risk level of the nonlinear resonance effect; Determine the resonance risk level of the system based on the posterior probability distribution; Classifying load patterns with posterior probabilities greater than or equal to a high risk threshold as high risk levels; Classify the load patterns whose posterior probability is not less than the high risk threshold and greater than or equal to the low risk threshold as medium risk level; Classify patterns with posterior probabilities less than the low-risk threshold as low-risk levels; S3.4, dynamically adjust the weights of the identified high-risk load pattern combinations based on the risk analysis results of the Bayesian network; Using genetic algorithms, the weights of high-risk patterns are optimized first; S3.5, storing the optimized high-risk load pattern weights and associating them with corresponding timestamps and power load scenarios.

5. The method for power load forecasting and optimization based on artificial intelligence according to claim 4 is characterized in that: Complete steps to obtain the resonance strength of each load mode in different frequency ranges: S3.1.1, first, convert the preprocessed power load signal x(t) into the frequency domain, and use Fourier Transform (FT) to obtain the frequency domain signal X(f), where f represents the frequency; construct the frequency range F as multiple discrete intervals F = {f1, f2, ..., f n }, where f i represents the ith frequency interval; the spectrum signal X(f i ): S3.1.2, after obtaining the spectrum signal of each frequency interval, calculate the instantaneous energy density of each frequency interval; construct the instantaneous energy density E(f i ,t) is the square form of the signal energy at time t; the nonlinear conversion form of energy density is used for calculation, using the following formula: Among them, α, β, and γ are nonlinear conversion parameters, which are used to adjust the calculation method of instantaneous energy density to capture the dynamic changes of the signal in different frequency ranges; S3.1.3, after calculating the instantaneous energy density of each frequency interval, extract the resonance factor R (f i ): Among them, θ(f i ,t) indicates that in the frequency interval f i The instantaneous phase angle on is obtained from the time domain signal through Hilbert transform; S3.1.4, for each frequency interval, the resonance factor is represented by the original resonance intensity of the corresponding frequency interval, and the resonance factor is normalized; the normalized resonance intensity is marked as S(f i ); S3.1.5, finally, the normalized resonance intensity of each frequency interval is organized into a resonance intensity matrix D(M), where each row of the matrix corresponds to a frequency interval and each column corresponds to the resonance intensity of a different time window.

6. The method for predicting and optimizing power load based on artificial intelligence according to claim 4, characterized in that: Complete steps to obtain the mode phase difference in the time direction: S3.2.1, first, the instantaneous phase information is extracted from the power load signal x(t) by Hilbert transform; for a given power load signal, its corresponding analytical signal z(t) is obtained by Hilbert transform, and the analytical signal is defined as: in, represents the Hilbert transform of x(t), j is the imaginary unit; the instantaneous phase θ(t) is calculated by analyzing the phase angle of the signal: θ(t) represents the phase change of the load signal at each moment, which lays the foundation for the subsequent phase difference analysis; S3.2.2, after obtaining the instantaneous phase; calculate the phase synchronization index to measure the phase synchronization degree between the two load patterns x1(t) and x2(t); the calculation formula of the phase synchronization index PLV is: Where N is the number of sampling points in the time window, λ1(t k ) and θ2(t k ) are two modes at time t k The instantaneous phase of S3.2.3, after obtaining the phase synchronization index, calculate the phase difference metric to describe the directionality of the phase offset; the calculation formula of the phase difference metric PLI is: Where sign(·) represents the sign function, which is used to preserve the directionality of the phase shift, and λ is the nonlinear expansion parameter, which is used to adjust the sensitivity of the phase difference metric; S3.2.4, integrate the phase difference measurement values ​​in each time window to calculate the overall phase difference intensity D(T), the formula is: Where T represents the total time interval [t0, t1], t c is the time center point, ρ is the time attenuation factor, which is used to adjust the influence of different time periods on the overall phase difference intensity; S3.2.5, finally, the overall phase difference intensity calculated in each time window is organized into a phase difference matrix S(F), where each row of the matrix corresponds to a load mode combination, and each column corresponds to the phase difference intensity in different time periods.

7. The method for predicting and optimizing power load based on artificial intelligence according to claim 6, characterized in that: Step S4 includes the following contents: S4.1, the load patterns classified into high, medium, and low risk levels are input into the deep neural network and preprocessed before input; S4.2, based on the preprocessed load pattern data, use a deep neural network to perform multi-mode prediction fusion for high, medium, and low risk level patterns; S4.3, based on the trained deep neural network, the particle swarm optimization algorithm is used to adjust the weight of the load pattern in real time; S4.4, after completing the weight adjustment, the deep neural network outputs the fused prediction results.

Images