Intelligent multi-channel control method and system for power supply power

Abstract

The invention relates to the technical field of power supply power control, and discloses an intelligent multi-channel control method and system for power supply power, and the method comprises the steps: collecting voltage and current signals of a multi-channel power supply system in real time through SCADA, carrying out the frequency domain conversion of the signals through Fourier transform, and carrying out the frequency domain conversion of the signals; obtaining a power characteristic vector containing power amplitude, phase deviation and harmonic content; using a convolutional neural network to extract amplitude and phase offset features in the vectors, comparing with a preset load dynamic rule base to determine potential anomalies, and obtaining an abnormal feature set; and comparing the multi-channel harmonic content similarity to position an interference source, determining a fault type in combination with load dynamic historical data and a fault classification knowledge base, fusing features to generate an enhanced control strategy template, optimizing power distribution parameters through a convolutional neural network, and outputting a dynamic regulation and control instruction to realize power balance. According to the method, the data acquisition and control efficiency is improved based on the SCADA system, the problem of power imbalance of the multi-path power supply system is effectively solved, and stable and efficient operation of the system is guaranteed.

Description

Technical Field

[0001] This invention relates to the field of multi-channel power supply system control technology, and in particular to an intelligent multi-channel control method and system for power supply. Background Technology

[0002] Currently, in the field of modern power systems and intelligent devices, stable control of multiple power sources is the key to efficient system operation. SCADA (Supervisory Control and Data Acquisition System), as a mainstream industrial control technology, is often used for remote monitoring and data acquisition of power systems. It obtains parameters such as voltage and current through sensor networks to provide a data foundation for power control.

[0003] In existing technologies, static power thresholds or simple monitoring mechanisms are primarily relied upon. For example, when judging power anomalies in a multi-channel system, a single power amplitude is used for assessment, ignoring key characteristics such as phase shift and harmonic content. However, these characteristics exhibit nonlinear changes under complex loads, significantly affecting power distribution stability, and existing technologies cannot fully capture their dynamic patterns. Furthermore, due to insufficient depth in power characteristic analysis, existing technologies struggle to comprehensively capture the dynamic changes in amplitude, phase, and harmonics, making it difficult to pinpoint the source of anomalies. Moreover, existing technologies suffer from low anomaly response and control efficiency, failing to handle the cascading effects of harmonic interference between multiple channels and hindering the generation of targeted control strategies.

[0004] Therefore, existing technologies have the problem of being unable to quickly identify the source of anomalies and generate targeted multi-path control strategies. Summary of the Invention

[0005] This invention provides an intelligent multi-channel control method and system for power supply, to solve the problem that existing multi-channel power control cannot quickly identify the source of anomalies and generate targeted multi-channel control strategies.

[0006] In a first aspect, to solve the above-mentioned technical problems, the present invention provides an intelligent multiplexing control method for power supply, comprising: Voltage and current signals are acquired in real time and converted in the frequency domain to obtain a power characteristic vector containing power amplitude, phase shift and harmonic content; Features are extracted based on the power characteristic vector and compared with a preset load dynamic rule base to obtain an abnormal feature set; Based on the abnormal feature set, the similarity of the harmonic content is calculated to identify multi-channel cooperative interference and determine the interference source identifier; From the interference source identifier, obtain the load dynamic real-time data, analyze the phase offset trend, locate the anomaly to the specific control loop, and obtain the location result vector; Based on the location result vector, it is compared with a preset fusion fault classification knowledge base to obtain classification labels and fault types; Compare the phase shift in the abnormal feature set with the matching degree of the classification label. If the matching degree threshold is met, then fuse the abnormal feature set and the preset control strategy template to obtain an enhanced template set. Based on the enhanced template set, multi-path power allocation adjustment parameters are generated and optimized to obtain a control strategy set; Execute the dynamic adjustment instructions of the control strategy set to obtain a stable power state.

[0007] Secondly, the present invention provides an intelligent multi-channel control system for power supply, comprising: The data acquisition and conversion module is used to acquire the original voltage and current signals of each power source in a multi-power system, generate the power characteristic vector of each power source, and form a multi-power characteristic set. The data acquisition and conversion module is used to acquire voltage and current signals in real time, and obtain a power characteristic vector containing power amplitude, phase shift and harmonic content through frequency domain conversion; An anomaly identification module is used to extract features based on the power characteristic vector and compare them with a preset load dynamic rule base to obtain an anomaly feature set; The interference source localization module is used to calculate the similarity of the harmonic content based on the abnormal feature set, identify multi-channel coordinated interference, and determine the interference source identifier. The fault location module is used to obtain real-time dynamic load data from the interference source identifier, analyze the phase offset trend, locate the anomaly to the specific control loop, and obtain the location result vector. The fault classification module is used to compare the location result vector with a preset fusion fault classification knowledge base to obtain classification labels and fault types. The template enhancement module is used to compare the phase shift in the abnormal feature set with the matching degree of the classification label. If the matching degree threshold is met, the abnormal feature set and the preset control strategy template are fused to obtain an enhanced template set. The strategy optimization module is used to generate and optimize multi-path power allocation adjustment parameters based on the enhanced template set to obtain a control strategy set. The control execution module is used to execute the dynamic control instructions of the control strategy set to obtain a stable power state.

[0008] Thirdly, the present invention also provides an electronic device, including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein the processor, when executing the computer program, implements the intelligent multiplexing control method for power supply described in any of the preceding claims.

[0009] Fourthly, the present invention also provides a computer-readable storage medium comprising a stored computer program, wherein, when the computer program is executed, it controls the device where the computer-readable storage medium is located to perform the intelligent multiplexing method for power supply described in any one of the preceding claims.

[0010] Compared with the prior art, the present invention has the following beneficial effects: (1) This invention collects the voltage and current signals of the power supply from multiple dimensions, and extracts power characteristics and abnormal features by combining Fourier transform and convolutional neural network, thus solving the problem that traditional methods cannot quickly identify the source of abnormality and generate targeted multi-path control strategies.

[0011] (2) By analyzing the harmonic correlation between multiple channels, fitting the phase shift trend and matching the control loop template, this invention realizes the rapid location of multiple cooperative interference sources and abnormal control loops, avoids the location deviation caused by single-channel isolated analysis, and shortens the fault diagnosis time.

[0012] (3) The present invention constructs a closed-loop control mechanism of “feature extraction-anomaly location-strategy fusion-parameter optimization”, which enables the control strategy to be adjusted in real time according to the load dynamics and equipment status, effectively improving the power balance rate and energy utilization efficiency of the multi-power supply system, reducing the equipment failure rate, and extending the service life of the system. Attached Figure Description

[0013] Figure 1 This is a flowchart illustrating an intelligent multiplexing method for power supply control provided in the first embodiment of the present invention. Figure 2 This is a schematic diagram of the structure of an intelligent multi-channel control system for power supply provided in the second embodiment of the present invention; Detailed Implementation The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. 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.

[0014] Reference Figure 1 The first embodiment of the present invention provides an intelligent multiplexing control method for power supply, comprising the following steps: S1 collects the voltage and current signals of the multi-power system in real time, and after Fourier transform frequency domain conversion, obtains a power characteristic vector containing power amplitude, phase shift and harmonic content. S2, Based on the power characteristic vector, a convolutional neural network is used to extract amplitude and phase shift features, and a preset load dynamic model is compared to determine potential anomalies and generate an anomaly feature set; S3, For the abnormal feature set, calculate the similarity of multi-channel harmonic content to identify collaborative interference and determine the interference source identifier; S4. Obtain load dynamic historical data based on the interference source identifier, analyze the phase offset trend to locate the anomaly to a specific control loop, and obtain the location result vector. S5. Compare the positioning result vector with the fault classification knowledge base to determine the fault type and generate a classification label. S6. Compare the matching degree between the phase shift in the abnormal feature set and the classification label. If the threshold is met, fuse the feature set and the control strategy template to obtain the enhanced template set. S7. Power allocation adjustment parameters are generated based on the enhanced template set, and the optimized control strategy set is determined by optimizing the parameters through a convolutional neural network. S8 executes the dynamic control instructions in the control strategy set, monitors the system operating status, and finally obtains the stable power state.

[0015] In step S1, the voltage and current signals of the multi-power supply system are acquired in real time, and Fourier transform is used for frequency domain conversion to obtain a power characteristic vector containing power amplitude, phase shift, and harmonic content, including: S11: The original voltage signal and the original current signal are acquired through the sensor array, filtered and denoised to obtain the filtered signal, which is then screened to obtain the preliminary power data pair. S12, Perform frequency domain conversion on the preliminary power data pair, determine the power amplitude and phase shift, and correct the phase shift anomaly to obtain the corrected spectrum data; S13, evaluate the harmonic content in the corrected spectrum data, correlate it with the power amplitude through multi-channel data synchronization and optimize the abnormal features to generate a power characteristic vector.

[0016] In step S11, the original voltage signal and the original current signal are acquired by the sensor array, filtered and denoised to obtain the filtered signal, which is then screened to obtain the preliminary power data pair.

[0017] It should be noted that the sensor array consists of voltage sensors (accuracy ±0.1V) and current sensors (accuracy ±0.05A), and needs to cover all power supply circuits of the multi-power system (such as 3 power supply circuits for industrial equipment). The original voltage signals collected are as follows: circuit 1 fluctuates within a range of 2.2-2.5V, circuit 2 fluctuates within a range of 2.0-2.3V, and the original current signals are as follows: circuit 1 fluctuates within a range of 8-12A, circuit 2 fluctuates within a range of 7.5-11A (all of which are the characterization range of the secondary side low-level analog signal).

[0018] The preprocessing involves using a low-pass filter (cutoff frequency 1kHz) to remove high-frequency noise. For example, the original voltage signal of channel 1 has high-frequency fluctuations of 2.2-2.5V (such as superimposed kHz-level interference ripple). After filtering, the high-frequency ripple of the voltage signal is suppressed, and the fluctuation amplitude is narrowed to 2.3-2.4V. After synchronization processing, the high-frequency noise of the current signal of channel 1 is also removed, changing from a high-frequency fluctuation state of 8-12A to a smoother range of 8.5-11.5A, thereby avoiding the impact of high-frequency interference on the accuracy of subsequent frequency domain analysis.

[0019] It should be noted that a voltage effective threshold (corresponding to the effective state of the secondary side of the industrial power supply) is set. After filtering, loops whose voltage meets the threshold are paired with the corresponding current signal, and preliminary power is calculated to form preliminary power data pairs containing power. For example, loop 1 is "power 20W, secondary side voltage 2.0V, secondary side current 10A", and loop 2 is "power 19W, secondary side voltage 1.9V, secondary side current 9.5A". The power calculated here is the equivalent power at the secondary side signal level, used for subsequent relative trend analysis. Its absolute value needs to be converted through a fixed transformation ratio coefficient to reflect the true power of the primary side.

[0020] In step S12, frequency domain conversion is performed on the preliminary power data pair to determine the power amplitude and phase shift, and the phase shift anomaly is corrected to obtain the corrected spectrum data.

[0021] It should be noted that the Fast Fourier Transform uses a 10kHz sampling frequency to convert the time-domain signal (such as the waveform of voltage and current changing with time) of the initial power data pair into a frequency-domain signal, thus obtaining the converted spectrum data.

[0022] The extraction of the fundamental frequency voltage amplitude, current amplitude, and phase shift between them involves reading key parameters at the fundamental frequency (50Hz, the standard fundamental frequency for industrial power) from the converted spectrum data. Based on the voltage amplitude, current amplitude, and phase shift, the power value is calculated. For example, the calculated power at the fundamental frequency of 50Hz for channel 1 is 20W with a phase shift of 5 degrees, while the calculated power for channel 2 is 19W with a phase shift of 12 degrees.

[0023] The phase offset correction involves setting a phase offset threshold (e.g., 10 degrees). For phase offsets exceeding this threshold (e.g., the original 12 degrees for loop 2), the historical time-series data of the recent fundamental frequency phase offset of that loop is used as the object. The least squares method is employed to fit the trend of this trend, obtaining a predicted trend value for the current moment (e.g., 4 degrees). This predicted value is then used to replace the abnormal measured phase offset value. Finally, a corrected feature vector is formed, containing parameters such as the corrected power value and phase offset.

[0024] In step S13, the harmonic content in the corrected spectrum data is evaluated, and multi-channel data synchronization is performed with the power amplitude to optimize the abnormal features and generate a power characteristic vector.

[0025] It should be noted that the amplitudes of the 3rd to 21st harmonics are calculated from the corrected frequency domain signal (spectral data); where the 3rd to 21st harmonics are common interference harmonics in multi-channel power supply systems; then the proportion of each harmonic to the total power, i.e., the harmonic content, is calculated. For example, the 3rd harmonic of channel 1 accounts for 8% and the 5th harmonic accounts for 5%, while the 3rd harmonic of channel 2 accounts for 7% and the 5th harmonic accounts for 4.5%. The synchronization association is based on timestamps, binding the harmonic content data of each channel with parameters such as power amplitude and phase offset of the corresponding circuit with millisecond-level precision, ensuring that all data are completely matched in the time dimension, forming a power characteristic group after synchronization; for example, the synchronization characteristic group of channel 1 is [20W, 5 degrees, 8%, 5%]).

[0026] The process of marking abnormal features involves setting a harmonic content standard (e.g., 3rd harmonic ≤ 5%), marking harmonic content exceeding the standard, recording the magnitude of the exceedance, and generating an optimized power feature set. This feature set clearly identifies the harmonic components that need to be suppressed and the target suppression amount, laying the foundation for generating a precise control strategy. For example, if the 3rd harmonic of path 1 is detected to be 8%, exceeding the standard value of 5%, an abnormal feature is marked.

[0027] In step S2, features are extracted based on the power characteristic vector and compared with a preset load dynamic rule base to obtain an abnormal feature set, including: S21, A convolutional neural network is used to perform a convolution operation on the power characteristic vector to generate preliminary extracted features; S22, calculate the energy distribution deviation between the preliminary extracted features and the preset load dynamic rule base. If the energy distribution deviation exceeds the preset deviation threshold, then filter the harmonic components and instantaneous fluctuation values ​​to obtain the filtered features. S23, compare the frequency domain peak deviation of the filtered features with the preset load dynamic rule base, analyze the correlation between amplitude deviation and phase offset, obtain abnormal patterns through pattern fusion, and generate an abnormal feature set if the overall deviation exceeds the threshold.

[0028] In step S21, a convolutional neural network is used to perform convolution operations on the power characteristic vector to generate preliminary extracted features.

[0029] It should be noted that the power characteristic vector is a single-path vector sequence generated in step S1; for example, the vector sequence of path 1 under continuous timestamps is: [t1 time 20W, 3 degrees, 4%], [t2 time 21W, 3.2 degrees, 4.1%], ..., [t256 time 22W, 2.8 degrees, 3.8%]; by continuously caching the latest N sampling periods (for example, N can be set to 256) of the power characteristic vector of each loop, an amplitude time series and a phase offset time series are formed for each loop; it should be noted that the harmonic content does not participate in the feature extraction in this step.

[0030] The one-dimensional convolutional neural network is designed for time-series data. Its input layer receives a three-dimensional tensor with dimensions of 1 (height) × N (sequence length) × 2 (number of channels). This tensor consists of two time-series channels: channel 0 is the amplitude time-series, and channel 1 is the phase-shift time-series. The network structure can be configured as follows: one-dimensional convolutional layer 1 contains 32 convolutional kernels of length 5 (stride 1, causal padding, ReLU activation) for extracting short-term dynamic patterns; one-dimensional convolutional layer 2 contains 64 convolutional kernels of length 3 (stride 1, causal padding, ReLU activation) for further feature abstraction; finally, a global average pooling layer is connected to convert the variable-length sequence features into a fixed-length feature vector.

[0031] The aim is to automatically learn and extract local correlation patterns (such as specific fluctuations and trend changes) of amplitude and phase shifts in temporal variations by sliding a one-dimensional convolutional kernel along the time dimension. These patterns are key to identifying dynamic anomalies in the load. The initial extracted features are the fixed-length abstract feature vector (e.g., a 64-dimensional vector) output by the network. This feature vector encodes deep pattern information of power characteristics within a recent time window. While its numerical value itself has no direct physical meaning, it provides a high-dimensional, nonlinear feature foundation for subsequent anomaly detection based on energy distribution deviations.

[0032] In step S22, the energy distribution deviation between the preliminary extracted features and the preset load dynamic rule base is calculated. If the energy distribution deviation exceeds the preset deviation threshold, harmonic components and instantaneous fluctuation values ​​are screened and filtered to obtain the filtered features.

[0033] It should be noted that the preset load dynamic rule base is trained from feature vectors extracted by the CNN network in S21 from a large amount of historical data under normal operating conditions. For example, the boundaries of normal features are established through a support vector machine model. For instance, based on the preliminary extracted features, a support vector machine model is compared with the preset load dynamic rule base. If the abnormal score exceeds a preset threshold, an abnormality is determined, and the associated harmonic and fluctuation parameters are obtained to generate a feature set to be analyzed. The fixed-length abstract feature vector (e.g., a 64-dimensional vector) output from step S21 is used as the input features. Each dimension is an abstract numerical value extracted from the original power time series data by the CNN, without direct physical meaning. The kernel function is set to a radial basis function (RBF). The allowed outlier ratio is set to a range of (0,1], representing the upper bound of the training error and the lower bound of the support vector ratio. For example, setting it to 0.05 means that about 5% of the training samples are allowed to be treated as outliers. The parameters of the RBF kernel function affect the complexity of the model. The larger the gamma value, the more complex the model, the more tortuous the decision boundary, and the easier it is to overfit; the smaller the value, the smoother the boundary. It is usually selected through cross-validation.

[0034] The model's decision function f(x) returns a sign (+1 or -1) and an anomaly score for a new input feature vector x (i.e., the output of S21). The sign can be represented as f(x) = +1, indicating the sample is identified as normal (falling within the decision boundary); f(x) = -1 indicates an anomaly (falling outside the decision boundary). The anomaly score is a real number representing the sign distance from the sample point to the decision boundary; a more negative score indicates a higher degree of anomaly; a positive score or close to zero indicates normality. The preset load dynamic rule base is essentially a pre-trained One-Class SVM model. The training data uses historical data collected over a long period under various known normal load conditions (such as motor no-load, half-load, stable operation of resistive load, etc.). This historical data is processed according to step S21 to obtain tens of thousands of abstract feature vectors under normal conditions, which serve as the training set.

[0035] During training, the SVM model is trained using the aforementioned normal feature vectors. The goal is to find a minimum volume decision boundary in the 64-dimensional abstract feature space that can encompass the vast majority of normal samples. The specific operation process includes: inputting the 64-dimensional abstract feature vector generated for the current loop in step S21 into the pre-trained One-Class SVM model; the model calculates and returns the anomaly score of this vector. For example, for the feature vector of loop 1, the model returns an anomaly score of -0.85; a preset anomaly score threshold is set, for example, -0.5. This threshold is determined based on the detection results of the validation set (containing known normal and slightly anomalous data), aiming to balance the false positive rate and the false negative rate. The anomaly score is compared with the threshold: if the score < the threshold, it is determined to be an anomaly. For example, the score of loop 1, -0.85 < -0.5, therefore loop 1 is determined to be in an abnormal state. Once an anomaly is determined for a loop, the system immediately extracts the original physical parameters of that loop from the power characteristic vector generated in step S1 at the current time.

[0036] The parameters include: harmonic content, such as the current 3rd harmonic accounting for 8% and the 5th harmonic accounting for 5% of the circuit. Instantaneous fluctuation value, which is the standard deviation or peak-to-peak value of the circuit power within the most recent short time window (such as 1 second), as an indicator of fluctuation quantification, for example, 0.5W.

[0037] These parameters are packaged with the loop's anomaly score to form the filtered features. For example, the filtered features of loop 1 are: {Loop ID: 1, Anomaly score: -0.85, 3rd harmonic: 8%, 5th harmonic: 5%, Power fluctuation: 0.5W}.

[0038] In step S23, the frequency domain peak deviation of the filtered features is compared with that of the preset load dynamic rule base, and the correlation between amplitude deviation and phase offset is analyzed. If the correlation exceeds the preset correlation threshold, the abnormal pattern is obtained and fused to generate an abnormal feature set.

[0039] It should be noted that the filtered features are the output of step S22, including the anomaly scores of each loop and their associated physical parameters (such as harmonic content and power fluctuation values). The preset load dynamic rule base stores different anomaly modes, such as "uneven load," "harmonic interference," and "phase loss," and mapping rules for anomaly score ranges and physical parameter characteristics. For example, the rule corresponding to the "uneven load" mode is "anomaly score < -0.5 and power fluctuation value > 0.3W." The data in the filtered features is matched with the preset load dynamic rule base. If the features of a loop meet the conditions of a certain rule, it is determined to be an anomaly mode of that type. Finally, the judgment results, anomaly scores, and physical parameters of all loops are integrated to generate an anomaly feature set, providing a clear classification basis for subsequent interference source localization. For example, the filtered features of loop 1 are {anomaly score: -0.85, 3rd harmonic: 8%, power fluctuation: 0.5W}. After matching with the preset load dynamic rule base, because its power fluctuation (0.5W > 0.3W) and score (-0.85 < -0.5) are both determined, its dominant anomaly mode is determined to be "uneven load," and an anomaly feature set is generated.

[0040] In step S3, based on the anomalous feature set, the similarity of the harmonic content is calculated to identify multi-path cooperative interference and determine the interference source identifier, including: S31, Perform spectral decomposition on the harmonic components in the abnormal feature set, calculate the harmonic peak difference, and obtain the inter-path harmonic difference set; S32, Based on the inter-path harmonic difference set, calculate the correlation with the power characteristic vector to identify preliminary cooperative interference; S33, perform convolution operation on the preliminary coordinated interference, identify the dominant interference path, match the preset interference source template, and determine the interference source identifier.

[0041] In step S31, the harmonic components in the abnormal feature set are subjected to spectral decomposition, the harmonic peak differences are calculated, and the inter-path harmonic difference set is obtained.

[0042] It should be noted that the harmonic content data in the abnormal feature set refers to the percentage data of each loop and each harmonic contained in the abnormal features generated in step S2 (e.g., the 5th harmonic percentage of loop 1 is 15%, and the 3rd harmonic percentage is 8%; the 5th harmonic percentage of loop 2 is 12%, and the 3rd harmonic percentage is 7%). The percentage of each harmonic needs to be extracted separately for each loop. The harmonic peak value difference is the difference in the harmonic percentage values ​​of different loops at the same harmonic order. It is calculated in the format of "harmonic order - loop A peak value - loop B peak value - difference", for example: For the 5th harmonic, the peak value difference between path 1 and path 2 is 15% - 12% = 3%; for the 3rd harmonic, the peak value difference between path 1 and path 2 is 8% - 7% = 1%. The inter-path harmonic difference set is a structured set formed by integrating peak value difference data for all harmonic orders and all loops. For example, this set can be recorded as: {"5th_harmonic":{"circuit1-circuit2":3%,"circuit2-circuit3":2%},"3rd_harmonic":{"circuit1-circuit2":1%,"circuit2-circuit3":0.8%}}.

[0043] In step S32, the correlation with the power characteristic vector is calculated based on the inter-path harmonic difference set to identify preliminary cooperative interference.

[0044] It should be noted that the system continuously records the following data over a period of time (e.g., the past 5 minutes): the inter-path difference time series data of each harmonic, for example, recording the difference between the 5th harmonic and path 1 and path 2 once per second, forming a time series Diff_5th_harmonic(t)=[d1,d2,...,d300]; the phase offset difference time series data of the corresponding loop, for example, recording the phase offset difference between path 1 and path 2 once per second, forming a time series Diff_phase(t)=[p1,p2,...,p300].

[0045] The correlation analysis involves calculating the Pearson correlation coefficient between two time series, Diff_harmonic(t) and Diff_phase(t), within the same time window. At this point, there are 300 pairs of observations (d_i, p_i).

[0046] The preset correlation threshold is set based on the system's anti-interference capability and significance level (e.g., 0.6). If the absolute value of the correlation coefficient exceeds the threshold (e.g., 0.7 > 0.6), it indicates that there is a significant linear correlation between the change in harmonic differences and the change in phase differences within that time period, and it is determined that there is preliminary cooperative interference at that harmonic order. The above time series correlation analysis was performed on all harmonic orders and loop pairs one by one to form a preliminary list of cooperative interferences. For example, "5th harmonic: the harmonic difference and phase difference of loop 1-loop 2 are correlated with 0.7, which is determined to be cooperative interference."

[0047] In step S33, a convolution operation is performed on the preliminary coordinated interference to identify the dominant interference path, match the preset interference source template, and determine the interference source identifier.

[0048] It should be noted that the convolution operation adopts the same 3-layer convolutional neural network as step S2 (this network is trained with training data from a multi-power system, the data source is the actual test records under different load types and different interference sources, including the cooperative interference correlation coefficient matrix of normal / abnormal working conditions and the corresponding real interference propagation direction label; the input layer matches the dimension of the preliminary cooperative interference data, convolutional layer 1 contains 32 3×3 convolutional kernels, and convolutional layer 2 contains 64 2×2 convolutional kernels), and performs feature extraction on the correlation coefficient matrix of the preliminary cooperative interference (such as the 5th harmonic inter-path correlation coefficient matrix [[1,0.85],[0.85,1]]). The interference propagation direction is determined by the feature weight value output by the network. For example, the feature weight value of path 1-path 2 is 0.9 (higher than 0.7 of path 2-path 3), then "path 1→path 2" is determined to be the dominant interference path of the 5th harmonic.

[0049] The training data for the convolutional neural network can be obtained by constructing a multi-power system simulation platform in a laboratory environment, and by using a controllable interference injection method (such as connecting a nonlinear load or faulty component to a specific circuit) to accurately simulate different interference sources and propagation paths, thereby obtaining a large number of cooperative interference correlation coefficient matrix samples with accurate annotations.

[0050] The preset interference source template is a database that stores common interference types (such as load coupling interference and line crosstalk interference) and their corresponding harmonic characteristics (such as the 5th harmonic peak value exceeding 12% and the correlation coefficient exceeding 0.8). For example, the characteristics of the "load coupling interference" template are "5th harmonic peak value 12%-18% and correlation coefficient 0.8-0.9".

[0051] The matching process involves comparing the harmonic characteristics of the dominant interference path (e.g., the peak value of the 15th harmonic from path 1 to path 25 is 15%, and the correlation coefficient is 0.85) with a preset interference source template one by one, and calculating the matching degree (e.g., a matching degree of 92% with the "load-coupled interference" template). The specific calculation process is as follows: first, determine the comparison dimensions (harmonic peak value percentage, inter-path correlation coefficient), and set weights (peak value percentage 0.6, correlation coefficient 0.4, allocated according to the influence of the characteristics on the interference type); then, take 5 from the "load-coupled interference" template... The standard range for subharmonics (peak value 13%-15%, correlation coefficient 0.78-0.83) is calculated. The deviation deduction for each feature is calculated: a peak value within the range of 15% earns 60 points (out of 60), and a correlation coefficient of 0.85 exceeding the standard upper limit by 0.02 deducts 2 points for every 0.01 exceeding the limit, earning 36 points (out of 40). The final total score is 60 + 36 = 96 points. After adjusting for template matching accuracy (such as additional deductions for template deviations in the correlation coefficient), the final score is converted to a 92% matching degree.

[0052] The preset matching degree threshold is set to 90%. The basis for this setting is based on the statistical analysis of 1000+ sets of measured interference case data. The minimum effective matching degree of 85% is taken as the matching degree between various interference source templates and real interference types, and a 5% safety margin is reserved to reduce the false judgment rate. If the matching degree exceeds the threshold (e.g., 92% > 90%), the interference type corresponding to the dominant interference path is integrated with the source loop (e.g., "load-coupling interference - path 1") to generate an interference source identifier, such as "Interference_Type:Load_Coupling;Source_Circuit:path 1;Harmonic_Order:5 times").

[0053] In step S4, real-time load dynamic data is obtained from the interference source identifier, and the phase offset trend is analyzed to locate the anomaly to a specific control loop, resulting in a location result vector, including: S41, Obtain real-time load dynamic data from the interference source identifier and perform time series segmentation to generate a load dynamic time series set; S42, Perform spectral decomposition on the time series set, extract the phase offset vector and analyze the offset trend to generate a phase offset trend curve; S43, match the phase offset trend curve with the preset control loop template, determine the control loop where the anomaly is located, and generate a positioning result vector.

[0054] In step S41, load dynamic real-time data is obtained from the interference source identifier and time series segmentation is performed to generate a load dynamic time series set.

[0055] It should be noted that the interference source identifier includes the interference type, source circuit, and associated harmonic order (e.g., "Interference_Type:Load_Coupling;Source_Circuit:Circuit 1;Harmonic_Order:5th"). The acquired real-time load dynamic data is the real-time power, current, and voltage changes over time for the circuit (Circuit 1) and associated circuit (Circuit 2) corresponding to the identifier. The sampling interval is set to 1 second, and the data duration is 30 minutes before and after the interference occurs (e.g., 14:00-14:30).

[0056] The time series segmentation adopts the fixed time window method, dividing the 30-minute data into 1 second / segment to obtain 1800 time series segments; the load dynamic time series set is a set that is structured according to "circuit number-time segment-power value-current value-voltage value".

[0057] For example, the time series set of Route 1 is recorded as "Route 1, 14:00:01, 10.2kW, 10.1A, 1.82V; Route 1, 14:00:02, 10.3kW, 10.2A, 1.83V", ensuring that the data is arranged in an orderly manner according to the time dimension, providing a basis for subsequent trend analysis.

[0058] In step S42, spectral decomposition is performed on the time series set to extract the phase offset vector and analyze the offset trend, generating a phase offset trend curve.

[0059] It should be noted that the original voltage and current waveform signals are instantaneous values ​​acquired by the sensor, with a sampling frequency of 10kHz. Spectral decomposition uses Fast Fourier Transform (FFT) to process the waveform signal, converting it into a frequency domain spectrum.

[0060] From the converted spectrum data, extract the phase values ​​of the fundamental frequency (50Hz) and associated harmonics (such as the 5th harmonic at 250Hz). It is important to note that the phases of voltage and current must be extracted separately, and their difference must be calculated to obtain the phase shift (power factor angle). For example, in segment 114:00:01, if the 50Hz voltage phase is 10 degrees and the current phase is 7 degrees, then the phase shift is 3 degrees; if the 250Hz voltage phase is 25 degrees and the current phase is 17 degrees, then the phase shift is 8 degrees.

[0061] The phase offset vector is a vector generated in the format of "time segment - 50Hz phase - 250Hz phase" (e.g., [14:00:01, 3 degrees, 8 degrees]), which needs to cover all time series segments; the offset trend analysis adopts a linear regression algorithm, with time as the horizontal axis (unit: seconds) and phase offset value as the vertical axis, to fit the phase offset vector of the same frequency (e.g., 250Hz) to obtain a trend fitting line. For example, if the phase offset of 1250Hz increases from 8 degrees to 15 degrees in 10 seconds, the slope of the fitting line is 0.7 degrees / second.

[0062] The phase shift trend curve is a visual curve formed by integrating all frequency trend fitting lines, and the shift values ​​at key time points are marked (such as 15 degrees shift at 14:00:10 and 22 degrees shift at 14:00:20), which intuitively reflects the change law of phase shift over time.

[0063] In step S43, the phase offset trend curve is matched with a preset control loop template to determine the control loop where the anomaly is located and generate a positioning result vector.

[0064] It should be noted that the preset control loop template stores the phase offset standard curves of different control loops (such as loop 1 and loop 2) under normal operating conditions. For example, the 250Hz phase offset standard curve of loop 2 is "stable at 5-7 degrees from 14:00 to 14:30", and the 250Hz standard curve of loop 1 is "stable at 3-5 degrees from 14:00 to 14:30".

[0065] The matching uses a curve similarity algorithm, such as dynamic time warping algorithm or mean-based overlap, to calculate the overlap between the phase offset trend curve (such as the offset curve of loop 1250Hz) and the template curve of each loop. For example, the similarity between the loop 1250Hz curve and the template curve of loop 2 is 92%, while the similarity with the template of loop 1 is only 65%.

[0066] The preset matching threshold is set to 90% (based on and sourced from: statistical analysis of 1000+ sets of "abnormal features-loop correspondence" test case data, extracting the minimum effective similarity between various abnormalities and the actual corresponding loops as 85%, and reserving an additional 5% safety margin to reduce the false positive and false negative rates). If the similarity exceeds the threshold (92% > 90%), then the abnormal corresponding loop 2 is determined.

[0067] The location result vector is a vector generated in the format of "abnormal loop-correlated frequency-matching degree", such as [loop 2, 250Hz, 92%], which clarifies the specific location and associated parameters of the abnormality and provides a basis for subsequent fault classification.

[0068] In step S5, the location result vector is compared with a preset fusion fault classification knowledge base to obtain classification labels and fault types, including: S51, extract features from the positioning result vector to generate a harmonic content dataset; S52, compare the harmonic content dataset with the preset fusion fault classification knowledge base. If the preset comparison threshold is met, the fault type is determined. S53, the fault type is encoded to generate a classification label.

[0069] In step S51, features are extracted from the positioning result vector to generate a harmonic content dataset.

[0070] It should be noted that the features extracted from the positioning result vector (such as [loop 2, 250Hz, 92%]) include the associated frequency (250Hz, i.e., the 5th harmonic) of the abnormal loop (loop 2) and the harmonic content data of the loop at that frequency. The data comes from the power characteristic vector of step S1 and the real-time monitoring data of step S4. For example, the 5th harmonic of loop 2 accounts for 15%, the 3rd harmonic accounts for 8%, and the fundamental frequency (50Hz) accounts for 77%.

[0071] The harmonic content dataset is a structured dataset organized according to "loop number-frequency-harmonic percentage-acquisition time", for example, recorded as "loop 2, 50Hz (fundamental frequency), 77%, 14:00:30; loop 2, 150Hz (3rd harmonic), 8%, 14:00:30; loop 2, 250Hz (5th harmonic), 15%, 14:00:30".

[0072] In step S52, the harmonic content dataset is compared with a preset fusion fault classification knowledge base. If the preset comparison threshold is met, the fault type is determined.

[0073] It should be noted that the preset fusion fault classification knowledge base stores common fault types (such as equipment aging, line crosstalk, and load overload) and corresponding harmonic characteristic rules. For example, the rule for "equipment aging fault" is "5th harmonic proportion > 12% and 3rd harmonic proportion > 7%, fundamental frequency proportion < 80%", and the rule for "line crosstalk fault" is "5th harmonic proportion > 10% and the difference in harmonic proportion between adjacent circuits < 5%".

[0074] The comparison involves matching dataset features with knowledge base rules one by one. The overlap calculation method is as follows: determine the key feature items of the rule (such as the rule "equipment aging failure" which includes the 5th harmonic, 3rd harmonic, and fundamental frequency) and their weights (each of the three items accounts for 1 / 3), and determine whether each feature item in the dataset is within the rule standard range. The overlap is obtained by accumulating the weights of the matching items (for example, loop 2's "5th harmonic 15%, 3rd harmonic 8%, fundamental frequency 77%" all meet the rule range, with only the fundamental frequency being close to the edge and deducting a small number of points, resulting in a final overlap of 95%). The preset comparison threshold is 85% (based on statistics of 1000+ sets of failure cases, taking the minimum effective overlap of 80% for each type of failure and rule matching, with a 5% safety margin). If the overlap exceeds the threshold (95% > 85%), the failure type is determined to be "equipment aging failure", and the essential cause of the anomaly is clarified.

[0075] In step S53, the fault type is encoded to generate a classification label.

[0076] It should be noted that the encoding process adopts a fixed format of "fault type_abnormal loop_associated frequency_matching degree", where the fault type is abbreviated in English (e.g., equipment aging fault is abbreviated as Motor_Aging), the abnormal loop is identified by a number (e.g., loop 2 is Circuit2), the associated frequency is a specific value (e.g., 250Hz), and the matching degree is kept as an integer.

[0077] The classification label is an encoded string, such as "Motor_Aging_Circuit2_250Hz_95%", which contains core fault information and can be quickly called during subsequent template fusion and strategy generation.

[0078] In step S6, the phase shift in the abnormal feature set is compared with the matching degree of the classification label. If the matching degree threshold is met, the abnormal feature set and the preset control strategy template are fused to obtain an enhanced template set, including: S61, preprocess the phase offset of the abnormal feature set to generate a phase offset feature dataset; S62, calculate the matching degree between the phase offset feature dataset and the classification label. If the matching degree meets the preset phase offset matching degree threshold, a phase offset feature set that meets the conditions is obtained. S63, integrate the phase offset feature set that meets the conditions and the preset control strategy template to obtain an enhanced template set.

[0079] In step S61, the phase offset of the abnormal feature set is preprocessed to generate a phase offset feature dataset.

[0080] It should be noted that, based on the fault classification result obtained in step S5 (such as "equipment aging fault - loop 2"), the phase offset data of the loop (loop 2) related to the fault characteristic frequency (such as 250Hz) during the fault occurrence period is extracted from the record. After the preprocessing, the generated phase offset feature dataset is a dataset that is structured according to "loop number-frequency-phase offset value-time stamp", for example, recorded as "loop 2, 250Hz, 12 degrees, 14:00:30". This dataset will be used as the direct input for template fusion. In step S62, the matching degree between the phase offset feature dataset and the classification label is calculated. If the matching degree meets the preset matching degree threshold, a phase offset feature set that meets the conditions is obtained.

[0081] It should be noted that the classification labels include the abnormal circuit (Circuit2), the associated frequency (250Hz), and the fault type (Motor_Aging). When calculating the matching degree, the data of the corresponding circuit and frequency in the phase offset dataset (e.g., circuit with a phase offset of 12 degrees at 2250Hz) are filtered and compared with the standard phase offset range of this fault type (e.g., equipment aging fault: 10-15 degrees). The overlap ratio is calculated as the matching degree (e.g., if 12 degrees is within the range, the matching degree is 100%). The preset matching degree threshold is set to 85%. If the matching degree exceeds the threshold, the associated abnormal characteristics of the circuit at that frequency are directly queried from the harmonic content dataset generated in step S51 (e.g., the 5th harmonic of circuit 2 accounts for 15%). Finally, the matched phase offset data is integrated with the associated anomaly features to generate a phase offset feature set that meets the conditions (e.g., {loop:2, frequency:250Hz, phase offset:12°, 5th harmonic:15%}), providing input for subsequent template fusion. In step S63, the phase offset feature set that meets the conditions and the preset control strategy template are integrated to obtain an enhanced template set.

[0082] It should be noted that the preset control strategy template is a standardized adjustment parameter template for different fault types. For example, the template for the "Motor_Aging" fault is "250Hz harmonic suppression: reduce voltage by 5%, power distribution: increase fundamental frequency power by 10%, and increase 3rd harmonic filtering intensity by 20%". The integration is to associate the phase offset characteristics that meet the conditions (such as loop 2250Hz 12 degrees) with the adjustment parameters in the template and supplement the "abnormal loop - control target value" information. For example, the template parameters are adjusted to "Loop 2: 250Hz harmonic suppression (voltage drop of 5%), increase fundamental frequency power by 10% (from 10kW to 11kW), and increase 3rd harmonic filtering intensity by 20%".

[0083] The enhanced template set is a collection of multiple fault-strategy association records. The core data of each record consists of the following: 1. A clear fault type (such as equipment aging fault, load coupling interference); 2. Corresponding specific abnormal features (such as 5th harmonic proportion of 12%-18%, 250Hz phase offset of 10-15 degrees); 3. Matching parameter optimization strategies (such as adjusting the filter cutoff frequency, optimizing the load distribution ratio), ensuring that each strategy corresponds precisely to the specific abnormal features, forming a structured association record that can be directly used for subsequent parameter optimization.

[0084] In step S7, based on the enhanced template set, multi-path power allocation adjustment parameters are generated and optimized to obtain a control strategy set, including: S71, perform structured processing on the initial power allocation data in the enhanced template set to generate a structured dataset; S72, if the structured dataset meets the preset system efficiency threshold, then a convolutional neural network is used to process and optimize the multi-path power allocation adjustment parameters; S73, based on the operating status, dynamically adjust the multi-channel power allocation adjustment parameters, and integrate the preset control strategy template to generate a control strategy set.

[0085] In step S71, the initial power allocation data in the enhanced template set is processed into a structured dataset.

[0086] It should be noted that the enhanced template set includes initial power allocation data (e.g., the baseband power of loop 2 is increased by 10%, the current baseband power is 10kW, the initial adjustment target is 11kW, the baseband power of loop 1 is maintained at 10.2kW, and the baseband power of loop 3 is maintained at 10.5kW).

[0087] The structured processing involves organizing data in the format of "loop number - current power - adjustment direction - initial target power - associated frequency", for example, recorded as "loop 2, 10kW (base frequency), boost, 11kW, 50Hz; loop 1, 10.2kW (base frequency), maintain, 10.2kW, 50Hz; loop 3, 10.5kW (base frequency), maintain, 10.5kW, 50Hz".

[0088] The structured dataset must include harmonic suppression-related parameters (such as a 5% voltage drop at 2250Hz in the circuit) to ensure unified management of power distribution and harmonic control data.

[0089] In step S72, if the structured dataset meets the preset system efficiency threshold, a convolutional neural network is used to process and optimize the multi-path power allocation adjustment parameters.

[0090] It should be noted that the preset system efficiency threshold is the overall system power utilization standard (e.g., ≥90%), which is determined by calculating the ratio of the sum of the power of each loop in the structured data to the rated total power of the system (e.g., 30kW). For example, if the adjusted total power is 31.7kW (11+10.2+10.5) and the rated total power is 32kW, the utilization rate is 99.06% ≥90%, which meets the threshold.

[0091] The convolutional neural network employs a 3-layer structure (input layer dimension matches the number of dataset parameters, convolutional layer 1 contains 32 3×3 convolutional kernels, convolutional layer 2 contains 64 2×2 convolutional kernels, and the output layer outputs optimized power values). The input is a structured dataset. First, the "loop-power-frequency" tabular data (e.g., 3 loops, 5 frequency points, each row containing "loop ID, frequency value, power value") is converted into tensor format: grouped by loop, the "frequency-power" data of each loop is organized into a 1×M vector (M is the number of frequency points), N loops constitute an N×M feature matrix, and then a channel dimension (set to 1) and a batch dimension are added to form [1,1,N,M]. The CNN training data comes from measured "structured data-optimal power parameter" sample pairs under different loads and fault conditions (the input is normalized and the label is the high-efficiency power value verified by simulation). Iterative optimization is based on the backpropagation algorithm. The difference between the output and the optimal label is minimized by the mean square error loss function. The output layer inversely scales the normalized result and then fine-tunes it in combination with the "system total power constraint rule". Finally, the optimized power allocation parameters are output. For example, the base frequency power of loop 2 is fine-tuned from 11kW to 11.2kW (improving the system utilization rate to 99.5%), loop 1 is fine-tuned to 10.1kW, and loop 3 is fine-tuned to 10.6kW.

[0092] The optimization objective is to minimize power fluctuations in each loop (e.g., power fluctuations in each loop ≤ 0.2kW after optimization) while meeting the efficiency threshold, ensuring stable system operation. The efficiency threshold is set based on: statistically analyzing over 1000 sets of measured operating data under various normal operating conditions such as full load, half load, and different load types, to determine the minimum efficiency value (e.g., 95%) when the system can output stably and the power quality (e.g., harmonic ratio, phase offset) meets the standards. To address potential impacts such as temporary load fluctuations and slight equipment aging, an additional 3% safety margin is reserved, ultimately determining the efficiency threshold (e.g., 92%) to prevent the system efficiency from falling below the critical level required for stable operation due to sudden interference.

[0093] In step S73, the multi-channel power allocation adjustment parameters are dynamically adjusted based on the operating status, and a control strategy set is generated by integrating a preset control strategy template.

[0094] It should be noted that the operating status refers to real-time monitoring of system current fluctuations (e.g., ≤5%), voltage stability (e.g., voltage deviation ≤ ±0.05V), and harmonic content (e.g., 5th harmonic ≤ 10%). For example, if the monitored current fluctuation of loop 2 is 3%, voltage deviation is 0.03V, and 5th harmonic is 9.5%, it meets the stability requirements. The dynamic adjustment refers to fine-tuning and optimizing parameters based on the operating status. The specific fine-tuning rules are as follows: 1. First, preset the normal deviation range of the operating status parameters (e.g., voltage) (e.g., ±0.02V). 1. Calculate the voltage drop parameter by 0.1% for every 0.01V deviation from the normal range and the corresponding adjustment coefficient; 2. Monitor the operating parameters in real time and calculate the actual deviation value (e.g., if the voltage deviation of loop 2 is 0.03V, it exceeds the normal range by 0.01V); 3. Determine the adjustment amount according to the principle of "the direction of deviation is opposite to the direction of parameter adjustment" and the adjustment coefficient (if the deviation exceeds 0.01V, the voltage drop parameter is reduced by 0.1%×1=0.2%); 4. Perform fine-tuning (fine-tune the voltage drop parameter of loop 2 from 5% to 4.8%).

[0095] The integrated preset control strategy template is used to supplement the execution timing (e.g., "execute voltage adjustment first, then execute power distribution") and priority (e.g., "loop 2 control priority 1, loop 1 and loop 3 priority 2").

[0096] The control strategy set is a structured collection organized according to "control object - power parameter - voltage parameter - execution timing - priority", for example, recorded as "Loop 2: base frequency power 11.2kW, 250Hz voltage drop 4.8%, execution timing 1, priority 1; Loop 1: base frequency power 10.1kW, no voltage adjustment, execution timing 2, priority 2; Loop 3: base frequency power 10.6kW, no voltage adjustment, execution timing 2, priority 2", ensuring that instructions can be directly issued and executed.

[0097] In step S8, the dynamic adjustment instructions of the control strategy set are executed to obtain a power stable state, including: S81, obtain initial control instructions from the control strategy set, classify and sort the initial control instructions according to their timing, and generate a classified instruction set; S82, if the response delay of the classification instruction set exceeds the preset response delay threshold, then calibrate the power adjustment parameters in the initial control instruction and generate a dynamic control instruction. S83, execute the dynamic control command, monitor the real-time operating status and obtain feedback data. If the feedback data meets the preset power operating threshold, a power stable state is obtained.

[0098] In step S81, initial control instructions are obtained from the control strategy set, and the initial control instructions are classified and sorted in time to generate a classified instruction set.

[0099] It should be noted that the initial control command is a specific operation command in the control strategy set (such as "Loop 2 voltage drop of 4.8%", "Loop 2 base frequency power increased to 11.2kW", "Loop 1 base frequency power adjusted to 10.1kW"); the classification processing is divided into "voltage adjustment type" and "power distribution type" according to the control type. For example, "Loop 2 voltage drop of 4.8%" belongs to the voltage adjustment type, and "Loop 2 base frequency power of 11.2kW" belongs to the power distribution type.

[0100] The timing sequence is arranged according to the execution timing of the control strategy set (timing 1 takes precedence over timing 2) and the loop numbering order of the same type of instruction (e.g., loop 1 takes precedence over loop 3). For example, the sorting result is "Timing 1: Voltage regulation type - loop 2 (4.8%); Timing 2: Power distribution type - loop 1 (10.1kW), power distribution type - loop 3 (10.6kW), power distribution type - loop 2 (11.2kW)".

[0101] The classification instruction set is a structured list of instructions that integrates classification and sorting results, ensuring that the instruction execution logic is clear and conflict-free.

[0102] In step S82, if the response delay of the classification instruction set exceeds a preset response delay threshold, the power adjustment parameters in the initial control instruction are calibrated to generate a dynamic control instruction.

[0103] It should be noted that the response delay refers to the actual time it takes for the system to execute and provide feedback after the command is issued (for example, after issuing the command "Circuit 2 voltage drop 4.8%", the adjustment and feedback are completed within 100ms, and the response delay is 100ms). The preset response delay threshold is set to 150ms (based on the real-time requirements of industrial power systems). If the response delay exceeds the threshold (e.g., a command delay of 160ms), it is determined that the power adjustment parameter corresponding to the command needs to be calibrated. The calibration is based on a dynamic fine-tuning strategy using the correlation between "system execution load and response delay," specifically as follows: First, a calibration trigger threshold is preset (e.g., execution load exceeding 80% and response delay exceeding 50ms), and loop operation data is monitored in real time. If the trigger threshold is reached (e.g., the execution load of the system containing loop 2 reaches 82% and the response delay is 55ms), the adjustment amount is calculated according to the correlation rule of "a reduction of 0.1kW power adjustment range for every 5% increase in execution load" (a 2% increase in load requires a reduction of 0.04kW, rounded to 0.1kW). Finally, the power adjustment range is fine-tuned (reducing the baseband power of loop 2 from 11.2kW to 11.1kW) to achieve the goal of reducing system execution load and shortening response delay. The dynamic control command is the calibrated command, such as "Loop 2: baseband power 11.1kW, 250Hz voltage drop 4.8%, response delay 140ms," ensuring that the command execution efficiency meets the requirements.

[0104] In step S83, the dynamic control command is executed, the real-time operating status is monitored and feedback data is obtained. If the feedback data meets the preset power operating threshold, a power stable state is obtained.

[0105] It should be noted that the dynamic control commands are issued to each loop controller via the SCADA system. The controllers adjust the voltage and power parameters according to the commands (e.g., the loop 2 controller reduces the voltage by 4.8% and increases the power to 11.1kW). The real-time operating status monitoring includes the power values ​​of each loop (e.g., loop 2 11.1kW, loop 1 10.1kW, loop 3 10.6kW), harmonic content (e.g., 5th harmonic 8.2%, 3rd harmonic 7.1%), and load rate difference (the difference between the load rate of each loop and the average load rate, e.g., ≤5%). The preset power operating thresholds are "power fluctuation ≤0.2kW, 5th harmonic ≤10%, 3rd harmonic ≤8%, and load rate difference ≤5%". The feedback data are the monitored real-time parameters, such as "power fluctuation 0.1kW, 5th harmonic 8.2%, 3rd harmonic 7.1%, load rate difference 3%", all of which meet the thresholds. If the feedback data meets all the thresholds, the system is determined to have reached a power stable state, and the stable state parameters are recorded (such as "stable time: 14:05:00, power of each loop: loop 211.1kW, loop 110.1kW, loop 310.6kW, harmonic content: 5th harmonic 8.2%, 3rd harmonic 7.1%), thus completing this control process.

[0106] In summary, this invention achieves closed-loop management of the entire process from anomaly identification and interference location to precise control through multi-dimensional analysis of the power characteristics of multi-power supply systems and intelligent algorithm regulation. This effectively improves the power balance accuracy and operational stability of multi-power supply systems and provides technical support for efficient collaboration of multiple power supplies in scenarios such as industrial automation and energy management.

[0107] Reference Figure 2 The second embodiment of the present invention provides an intelligent multi-channel control system for power supply, comprising: The data acquisition and conversion module is used to acquire voltage and current signals in real time, and obtain a power characteristic vector containing power amplitude, phase shift and harmonic content through frequency domain conversion; An anomaly identification module is used to extract features based on the power characteristic vector and compare them with a preset load dynamic rule base to obtain an anomaly feature set; The interference source localization module is used to calculate the similarity of the harmonic content based on the abnormal feature set, identify multi-channel coordinated interference, and determine the interference source identifier. The fault location module is used to obtain real-time dynamic load data from the interference source identifier, analyze the phase offset trend, locate the anomaly to the specific control loop, and obtain the location result vector. The fault classification module is used to compare the location result vector with a preset fusion fault classification knowledge base to obtain classification labels and fault types. The template enhancement module is used to compare the phase shift in the abnormal feature set with the matching degree of the classification label. If the matching degree threshold is met, the abnormal feature set and the preset control strategy template are fused to obtain an enhanced template set. The strategy optimization module is used to generate and optimize multi-path power allocation adjustment parameters based on the enhanced template set to obtain a control strategy set. The control execution module is used to execute the dynamic control instructions of the control strategy set to obtain a stable power state.

[0108] It should be noted that the intelligent multiplexing control system for power supply provided in this embodiment of the invention is used to execute all the process steps of the intelligent multiplexing control method for power supply in the above embodiment. The working principle and beneficial effects of the two are one-to-one, so they will not be described again.

[0109] It should be noted that the device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Furthermore, in the accompanying drawings of the device embodiments provided by this invention, the connection relationships between modules indicate that they have communication connections, which can be specifically implemented as one or more communication buses or signal lines. Those skilled in the art can understand and implement this without any creative effort.

[0110] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. In particular, it should be noted that any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention for those skilled in the art.

Claims

1. A method for intelligent multiplexing control of power supply, characterized in that, Includes the following steps: Voltage and current signals are acquired in real time and converted in the frequency domain to obtain a power characteristic vector containing power amplitude, phase shift and harmonic content; Features are extracted based on the power characteristic vector and compared with a preset load dynamic rule base to obtain an abnormal feature set; Based on the abnormal feature set, the similarity of the harmonic content is calculated to identify multi-channel cooperative interference and determine the interference source identifier; From the interference source identifier, obtain the load dynamic real-time data, analyze the phase offset trend, locate the anomaly to the specific control loop, and obtain the location result vector; Based on the location result vector, it is compared with a preset fusion fault classification knowledge base to obtain classification labels and fault types; Compare the phase shift in the abnormal feature set with the matching degree of the classification label. If the matching degree threshold is met, then fuse the abnormal feature set and the preset control strategy template to obtain an enhanced template set. Based on the enhanced template set, multi-path power allocation adjustment parameters are generated and optimized to obtain a control strategy set; Execute the dynamic adjustment instructions of the control strategy set to obtain a stable power state.

2. The intelligent multi-channel control method for power supply according to claim 1, characterized in that, The real-time acquisition of voltage and current signals from the multi-source power supply system, after frequency domain conversion, yields a power characteristic vector containing power amplitude, phase shift, and harmonic content, including: The raw voltage and current signals are acquired by a sensor array, filtered and denoised to obtain the filtered signals, which are then screened to obtain preliminary power data pairs. The initial power data is subjected to frequency domain transformation to determine the power amplitude and phase shift, and the phase shift anomaly is corrected to obtain the corrected spectrum data; The harmonic content in the corrected spectrum data is evaluated, and multi-channel data synchronous correlation is performed with the power amplitude to optimize the abnormal features and generate a power characteristic vector.

3. The intelligent multi-channel control method for power supply according to claim 1, characterized in that, The step of extracting features based on the power characteristic vector, comparing them with a preset load dynamic rule base to determine potential anomalies, and obtaining an anomaly feature set includes: A convolutional neural network is used to perform convolution operations on the power characteristic vector to generate preliminary extracted features; Calculate the energy distribution deviation between the preliminary extracted features and the preset load dynamic rule base. If the energy distribution deviation exceeds the preset deviation threshold, then filter the harmonic components and instantaneous fluctuation values ​​to obtain the filtered features. The frequency domain peak deviation of the filtered features is compared with that of the preset load dynamic rule base, and the correlation between amplitude deviation and phase offset is analyzed. If the correlation exceeds the preset correlation threshold, the abnormal pattern is obtained and fused to generate an abnormal feature set.

4. The intelligent multi-channel control method for power supply according to claim 1, characterized in that, The step of calculating the similarity of the harmonic content based on the abnormal feature set, identifying multi-channel coordinated interference, and determining the interference source identifier includes: The harmonic components in the abnormal feature set are subjected to spectral decomposition, and the harmonic peak differences are calculated to obtain the inter-path harmonic difference set. Based on the inter-path harmonic difference set, the correlation with the power characteristic vector is calculated to identify preliminary cooperative interference; Perform convolution operations on the initial coordinated interference, identify the dominant interference path, match the preset interference source template, and determine the interference source identifier.

5. The intelligent multi-channel control method for power supply according to claim 1, characterized in that, The step of obtaining real-time load dynamic data from the interference source identifier, analyzing the phase shift trend, locating the anomaly to a specific control loop, and obtaining a location result vector includes: From the interference source identifier, obtain the load dynamic real-time data and perform time series segmentation to generate a load dynamic time series set; Perform spectral decomposition on the time series set, extract the phase shift vector and analyze the shift trend to generate a phase shift trend curve; The phase offset trend curve is matched with a preset control loop template to determine the control loop where the anomaly is located and generate a positioning result vector.

6. The intelligent multiplexing control method for power supply according to claim 1, characterized in that, The step involves comparing the location result vector with a preset fusion fault classification knowledge base to obtain classification labels and fault types, including: Features are extracted from the location result vector to generate a harmonic content dataset; The harmonic content dataset is compared with a preset fusion fault classification knowledge base. If the preset comparison threshold is met, the fault type is determined. The fault types are encoded to generate classification labels.

7. The intelligent multi-channel control method for power supply according to claim 1, characterized in that, The process involves comparing the phase shift in the abnormal feature set with the matching degree of the classification label. If a preset matching degree threshold is met, the abnormal feature set and the preset control strategy template are fused to obtain an enhanced template set, including: The phase shift of the abnormal feature set is preprocessed to generate a phase shift feature dataset; Calculate the matching degree between the phase shift feature dataset and the classification label. If the matching degree meets the preset matching degree threshold, a phase shift feature set that meets the conditions is obtained. By integrating the phase offset feature set that meets the conditions and the preset control strategy template, an enhanced template set is obtained.

8. The intelligent multi-channel control method for power supply according to claim 1, characterized in that, The step involves generating and optimizing multi-path power allocation adjustment parameters based on the enhanced template set to obtain a control strategy set, including: The initial power allocation data in the enhanced template set is processed in a structured manner to generate a structured dataset; If the structured dataset meets the preset system efficiency threshold, then a convolutional neural network is used to process and optimize the multi-path power allocation adjustment parameters; Based on the operating status, the multi-channel power allocation adjustment parameters are dynamically adjusted, and a control strategy set is generated by integrating preset control strategy templates.

9. The intelligent multi-channel control method for power supply according to claim 1, characterized in that, The process of executing the dynamic control instructions of the control strategy set to obtain a power stable state includes: Initial control instructions are obtained from the control strategy set, and the initial control instructions are classified and sorted in time sequence to generate a classified instruction set. If the response delay of the classification instruction set exceeds a preset response delay threshold, then the power adjustment parameters of the initial control instruction are calibrated to generate a dynamic control instruction. The dynamic control command is executed to monitor the real-time operating status and obtain feedback data. If the feedback data meets the preset power operating threshold, a power stable state is obtained.

10. An intelligent multi-channel control system for power supply, characterized in that, include: The data acquisition and conversion module is used to acquire voltage and current signals in real time, and obtain a power characteristic vector containing power amplitude, phase shift and harmonic content through frequency domain conversion; An anomaly identification module is used to extract features based on the power characteristic vector and compare them with a preset load dynamic rule base to obtain an anomaly feature set; The interference source localization module is used to calculate the similarity of the harmonic content based on the abnormal feature set, identify multi-channel coordinated interference, and determine the interference source identifier. The fault location module is used to obtain real-time dynamic load data from the interference source identifier, analyze the phase offset trend, locate the anomaly to the specific control loop, and obtain the location result vector. The fault classification module is used to compare the location result vector with a preset fusion fault classification knowledge base to obtain classification labels and fault types. The template enhancement module is used to compare the phase shift in the abnormal feature set with the matching degree of the classification label. If the matching degree threshold is met, the abnormal feature set and the preset control strategy template are fused to obtain an enhanced template set. The strategy optimization module is used to generate and optimize multi-path power allocation adjustment parameters based on the enhanced template set to obtain a control strategy set. The control execution module is used to execute the dynamic control instructions of the control strategy set to obtain a stable power state.

Images