Intelligent control method of three-phase load based on electric meter data

Abstract

The present application relates to the field of power supply and distribution technology, and in particular to a method for intelligent control of three-phase loads based on meter data, comprising: performing empirical mode decomposition on each phase current time series in three-phase current time series data to obtain multiple intrinsic mode function components; obtaining frequency domain information of each intrinsic mode function component through fast Fourier transform, calculating the power spectrum density and extracting the peak to obtain a characteristic period set; setting a basic priori period, and generating a priori period set based on different multiples, calculating the matching confidence of each characteristic period in each intrinsic mode function component and each prior period in the priori period set and constructing a corresponding normalized characteristic period matrix; constructing a global period confidence vector; using the global period confidence vector as the input of a long short-term memory network to predict the changing trend of the target signal, thereby improving the prediction accuracy of the LSTM model for load changes and realizing intelligent control of the three-phase load.

Description

Technical Field

[0001] The present application relates to the field of power supply and distribution technology, and in particular to a three-phase load intelligent control method based on electricity meter data. Background Art

[0002] In modern industrial power distribution systems, optimizing the three-phase load distribution of production equipment is key to improving production efficiency and reducing operating costs. Balancing three-phase loads is crucial to the stability of industrial power distribution systems and the efficiency of equipment operation. Unbalanced three-phase loads can lead to equipment overload, energy waste, and power system instability, which in turn affects production efficiency and equipment life.

[0003] Three-phase current time series data is a crucial resource for monitoring and optimizing factory energy management. This data not only reflects the factory's production status and equipment operation, but can also be used to forecast energy demand, optimize load distribution, and proactively detect potential equipment failures. Analyzing this data can achieve efficient energy utilization, reduce operating costs, and improve production efficiency. The operating cycles of different equipment may overlap within the same time series, causing the original current curve to exhibit complex nonlinear and nonstationary characteristics. Long Short-Term Memory (LSTM) networks require high-quality feature input when processing complex nonlinear and nonstationary time series data; otherwise, their predictive performance will be limited.

[0004] In industrial park power distribution systems, load variations exhibit distinct periodic characteristics, primarily driven by factory production plans and equipment operating patterns. Traditional LSTM models have limitations in processing these periodic characteristics. While LSTMs can capture long-term dependencies in time series, they are limited in their ability to extract complex periodic features, particularly nonlinear and nonstationary periodic variations. This can result in the model failing to fully utilize this periodic information when predicting three-phase load variations, impacting both accuracy and reliability. Summary of the Invention

[0005] In order to accurately extract and integrate the periodic features in three-phase current time series data, improve the LSTM model's prediction accuracy for load changes, realize intelligent regulation of three-phase loads, reduce three-phase imbalance, and enhance power system stability and equipment operation efficiency, this application provides a three-phase load intelligent regulation method based on meter data.

[0006] In a first aspect, the present application provides a three-phase load intelligent control method based on meter data, which adopts the following technical solutions:

[0007] A three-phase load intelligent control method based on electric meter data includes the following steps:

[0008] Acquire three-phase current time series data, and perform empirical mode decomposition on each phase current time series in the three-phase current time series data to obtain multiple intrinsic mode function components;

[0009] The frequency domain information of each intrinsic mode function component is obtained by fast Fourier transform, the power spectrum density is calculated and the peak value is extracted to obtain the characteristic period set;

[0010] By defining a priori period set and using a confidence index to evaluate the degree of matching between each period in the characteristic period set and the priori period set, a normalized characteristic period matrix is ​​constructed, including: setting a basic priori period and generating a priori period set based on different multiples, calculating the matching confidence of each characteristic period in the i-th intrinsic mode function component and each prior period in the priori period set, selecting the priori period with the highest confidence as the corresponding target matching period for each characteristic period, and statistically constructing a corresponding normalized characteristic period matrix based on each intrinsic mode function component;

[0011] fusing the energy weights of the intrinsic mode function components and the normalized characteristic periodic matrix to construct a global periodic confidence vector;

[0012] Based on the periodic feature construction of the sliding window, the global periodic confidence vector is used as the input of the long short-term memory network to predict the change trend of the target signal;

[0013] Based on the change trend of the target signal, the three-phase load is regulated.

[0014] Optionally, performing empirical mode decomposition on each phase current time series in the three-phase current time series data to obtain multiple intrinsic mode function components includes:

[0015] Collect the a-phase current data I in the three-phase current time series data a ;

[0016] The a-phase current data I a Empirical mode decomposition is performed to obtain a corresponding plurality of intrinsic mode function components, wherein the plurality of intrinsic mode function components respectively represent energy fluctuations at different frequency levels.

[0017] Optionally, obtaining frequency domain information of each intrinsic mode function component by fast Fourier transform, calculating power spectrum density, and extracting peak values ​​to obtain a characteristic period set includes:

[0018] Perform fast Fourier transform on each intrinsic mode function component to obtain the corresponding spectrum expression;

[0019] Performing a square operation on the frequency spectrum expression to obtain a corresponding power spectrum density to evaluate the energy distribution of each intrinsic mode function component at each frequency;

[0020] Identifying a peak point in the energy distribution curve, where the peak point is the frequency corresponding to a local maximum of the power spectrum;

[0021] Setting a power spectrum density threshold and screening out peak frequency points where the energy distribution is greater than the power spectrum density threshold;

[0022] Based on each peak frequency point, calculate and obtain the target period value that meets the preset conditions;

[0023] Based on the target period value, a characteristic period set corresponding to the intrinsic mode function component is constructed.

[0024] Optionally, the statistical construction corresponding to the normalized characteristic periodic matrix includes:

[0025] Construct an M×K characteristic periodic matrix for the M intrinsic mode function components and initialize all elements to 0;

[0026] Perform statistics on the prior period and matching confidence of each characteristic period in the characteristic period set of the i-th intrinsic mode function component;

[0027] Obtain the column in the characteristic period matrix that corresponds to the prior period with the maximum matching confidence , and the The value of is set to the value corresponding to the maximum matching confidence, and the corresponding periodic feature matrix is ​​constructed;

[0028] The periodic characteristic matrix is ​​normalized to obtain a normalized characteristic periodic matrix.

[0029] Optionally, if the prior period has no corresponding characteristic period matching, the confidence level corresponding to the prior period is defined as 0.

[0030] Optionally, a confidence model is constructed based on an exponential decay function, and the matching confidence of each characteristic period in the intrinsic mode function component and each prior period in the prior period set is calculated.

[0031] Optionally, fusing the energy weights of the intrinsic mode function components and the normalized characteristic periodic matrix to construct a global periodicity confidence vector includes:

[0032] Calculating the energy weight of each intrinsic mode function component based on the energy proportion;

[0033] A weighted sum is performed on each priori period in the normalized characteristic period matrix based on the energy weight to obtain a corresponding global period confidence vector.

[0034] Optionally, the periodic feature construction based on the sliding window, using the global periodic confidence vector as the input of the long short-term memory network, and predicting the change trend of the target signal includes:

[0035] For each time point t n Use the first N time points to calculate the corresponding global period confidence vector to obtain the corresponding time series global period confidence vector;

[0036] The global period confidence vector of the time series is used as the input of the long short-term memory network to predict the changing trend of the target signal.

[0037] This application has the following technical effects:

[0038] Using empirical mode decomposition (EMD), the system decomposes complex time series data layer by layer and weights different IMF components based on their energy proportions. This ensures that the periodic feature array in each component matches its contribution to the overall signal periodicity, enabling the LSTM model to more accurately capture periodic variations in the data. Furthermore, it enables adaptive matching of prior periods. By calculating the confidence level between the detected feature period and the prior period, it dynamically selects the optimal matching k value, improving the accuracy and robustness of periodic feature extraction. It can automatically match the prior period under different operating conditions, better meeting actual needs. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 This is a flow chart of steps S1-S6 in a three-phase load intelligent control method based on meter data in the present application. DETAILED DESCRIPTION

[0040] The specific scenario targeted by the present invention is: when processing the three-phase power system in an industrial park, the extraction of the periodic characteristics of the current data is not accurate enough, and it is impossible to fully adapt to and utilize these periodic characteristics for effective load regulation, thereby affecting the accuracy and efficiency of three-phase load regulation, and it is difficult to fully utilize the potential advantages of periodic characteristics in optimizing power distribution.

[0041] The present application embodiment discloses a three-phase load intelligent control method based on meter data, referring to Figure 1 , including the following steps:

[0042] S1: Acquire three-phase current time series data, and perform empirical mode decomposition on each phase current time series in the three-phase current time series data to obtain multiple intrinsic mode function components.

[0043] It's important to note that empirical mode decomposition (EMD) is an adaptive decomposition technique based on the local characteristics of a signal. It decomposes the original signal into a series of intrinsic mode functions (IMFs) representing characteristics at different local time scales (i.e., different frequency distribution levels) and a residual term. Each IMF component reflects the energy fluctuations of the signal within a specific frequency range, effectively isolating local periodic characteristics and trend changes during device operation.

[0044] In the embodiment of the present application, the first Phase current data is , obtain the time series measurement data with a time series length of N. Perform empirical mode decomposition to obtain M IMF components. , representing energy fluctuations at different frequency levels.

[0045] S2: For each intrinsic mode function component, the frequency domain information is obtained by fast Fourier transform, the power spectrum density is calculated, and the peak value is extracted to obtain the characteristic period set.

[0046] It should be noted that the IMF components obtained after EMD decomposition of the three-phase current time series reflect the oscillation characteristics of the signal at different time scales. To extract the periodic information in these components, this step S2 converts the IMFs from the time domain to the frequency domain using a fast Fourier transform and calculates the power spectral density (PSD). By setting a threshold, one or more energy peaks in the frequency domain are extracted and the corresponding characteristic period is inferred.

[0047] In one embodiment of the present application, the process of extracting the characteristic period of each IMF component is as follows:

[0048] a. Perform fast Fourier transform on each IMF component to obtain its frequency domain information; b. Calculate the power spectrum density to obtain the energy distribution corresponding to each frequency point; c. Extract the peak point; d. Derivation and construction of the period.

[0049] The details are as follows:

[0050] a. Perform fast Fourier transform on each IMF component to obtain its spectrum information: IMF components Perform fast Fourier transform to convert it from time domain to frequency domain and obtain frequency domain expression ,in is the frequency variable. It describes the complex amplitude distribution of the component at each frequency point and reflects the structure of its main frequency components.

[0051] b. Calculate the power spectrum density (PSD): Get the energy distribution corresponding to each frequency point. Perform a square operation to obtain the power spectrum density, which is used to evaluate the energy distribution of the component at each frequency, where .

[0052] c. Peak point extraction: Identify the peak point in the curve, that is, the frequency corresponding to the local maximum of the power spectrum; set the power spectrum density threshold , filter out The peak frequency point .

[0053] d. Period derivation and construction: For each peak frequency , calculate the corresponding period: , all period value sets that meet the conditions constitute the first The characteristic period set of IMF components : .

[0054] Therefore, through the above steps, the IMF components can be effectively extracted from the three-phase current time series, and the corresponding characteristic period set can be constructed to obtain the characteristic period set corresponding to each IMF component, providing a basis for further period analysis.

[0055] S3: By defining a priori period set and using a confidence index to evaluate the degree of matching between each period in the characteristic period set and the prior period set, a normalized characteristic period matrix is ​​constructed, including: setting a basic priori period and generating a priori period set based on different multiples, calculating the matching confidence of each characteristic period in the i-th intrinsic mode function component and each prior period in the prior period set, for each characteristic period, selecting the prior period with the highest confidence as the corresponding target matching period, and based on each intrinsic mode function component, statistically constructing the corresponding normalized characteristic period matrix.

[0056] It's important to note that in industrial parks, different production equipment and processes often have relatively fixed operating cycles, such as 24-hour shifts, eight-hour work shifts, or automated control tasks with specific frequencies. These cyclical production behaviors directly affect the fluctuation patterns of the power load, which in turn manifests as certain stable periodic signals in the current time series.

[0057] Therefore, in order to further match the typical operating cycle structure in industrial production and improve the interpretability and structural consistency of the cycle characteristics, this step S3 introduces a predefined prior cycle set. In each IMF component, the characteristic cycle set detected by With the prior period set Compare and use the matching confidence as a quantitative indicator to construct the characteristic period matrix of the IMF component Furthermore, the normalized characteristic periodic matrix is ​​obtained by the maximum normalization operation , the normalized matrix describes the proportional contribution of each prior period component in the IMF component from a relative perspective. The value of each column reflects the IMF component in the corresponding prior period The larger the value, the more significant the cyclical component in the IMF component.

[0058] In one embodiment of the present application, in order to extract the periodic features related to the operation law of the industrial park from the three-phase current data, the following steps are performed: a. Define the prior period set ;

[0059] b. Calculate the confidence of each characteristic cycle and the prior cycle set Sequence; c. Select the prior period with the highest confidence for each characteristic period ; d. Construct periodic feature array ; e. Normalized characteristic periodic matrix .

[0060] The details are as follows:

[0061] a. Define the prior period set :

[0062] Setting the basic prior period For 24 hours (day), through different multiples Generate a priori period set , get the set Then, according to the production rhythm of the industrial park, different A collection of values ​​that constitutes a set Corresponding to the possible periodic time scale. Among them, is the basic a priori period, set to 24 hours. It is a multiple, set according to actual conditions, and is used to represent multiple cycles. It is through multiples Generate a priori cycles. Represents the prior period set The number of values ​​in .

[0063] Furthermore, by setting a basic prior cycle and combining different multiples, we generate a set of prior cycles. This set of prior cycles provides a reference benchmark for subsequent cycle matching. These cycles reflect the actual operating patterns of industrial parks, such as daily cycles, weekly cycles, and specific production shift cycles.

[0064] b. Calculate the confidence of each characteristic cycle and the prior cycle set sequence:

[0065] For the characteristic period of the i-th IMF component , it is necessary to evaluate its relationship with the prior period set Each cycle The degree of match. This is an indicator that measures the degree of match between the characteristic period and the prior period, determined by calculating the difference between the two. A higher confidence level indicates a closer match between the characteristic period and the prior period, indicating a higher probability that the IMF component signal contains the prior periodic component. The confidence level can reflect whether a typical periodic pattern is evident in the original component signal.

[0066] In one embodiment of the present application, a confidence model can be constructed based on an exponential decay function to calculate each characteristic period in the i-th IMF component: and the prior period set Each cycle The matching confidence , each characteristic period The length is The confidence sequence of : .in, It is a tuning parameter used to control the matching tolerance. is the absolute difference between the characteristic period and the prior period. It is the confidence calculation formula, and the value range is (0,1]. The larger the value, the higher the matching degree.

[0067] So far, each characteristic period in the i-th IMF component With each in the prior period set Make a difference and calculate the corresponding confidence to get the confidence sequence. Each characteristic cycle has a length of The confidence sequence provides a quantitative basis for subsequent matching.

[0068] c. Select the prior period with the highest confidence for each characteristic period :

[0069] For each characteristic period Select the prior period with the highest confidence as its matching period and obtain its matching confidence The corresponding period This step ensures that each characteristic period is assigned to the most similar prior period. , find the prior period corresponding to the maximum confidence and get its matching degree and the corresponding : ,

[0070] The corresponding prior period is: .in, Indicates that in all Select Confidence The largest . Then, by selecting the prior period with the maximum confidence, for each characteristic period The most similar prior period is found , and get the corresponding confidence .

[0071] d. Construct characteristic periodic matrix :

[0072] Among them, for all M IMF components, a statistical construction is constructed The characteristic periodic matrix of , used to indicate the The prior period set of IMF components Matching strength . Each entry of this matrix Indicates: IMF component Does it contain cyclical components? , and the included confidence ratio. Specifically, the following steps are included:

[0073] 1. Initialize the characteristic periodic matrix

[0074] is M IMF components Build Matrix , initially all elements are 0.

[0075] 2. Statistical matching results

[0076] For the first The characteristic period set of IMF components Each characteristic period Obtained and Conduct statistics.

[0077] 3. Assign confidence

[0078] In the characteristic periodic matrix In, find The corresponding position (i.e. Column) sets the value of the corresponding position to If there is already a value at that position (i.e., another feature period has been matched to the same prior period before), the larger confidence value is taken to retain the highest matching confidence.

[0079] 4. Handling unmatched prior cycles

[0080] If a priori period If there is no corresponding characteristic period match, its confidence is defined as 0.

[0081] For M IMF components, a periodic characteristic matrix is ​​constructed.

[0082]

[0083] in, is the first component of the i-th IMF Priori cycles The corresponding confidence value, if the prior period has no matching characteristic period, then If two characteristic periods match the same prior period (i.e., the matching The larger confidence level is used. .

[0084] At this point, by sorting out the matching results of the characteristic matrix of each IMF component, the characteristic period matrix of M IMF components is obtained .

[0085] e. Normalized characteristic periodic matrix .

[0086] In order to enhance the comparability of the periodicity intensity between the characteristic periodic matrices of different IMF components, the characteristic periodic matrix corresponding to each IMF component (i.e. ) is normalized to obtain the normalized characteristic periodic matrix Each row of the matrix describes the proportional contribution of each prior period component in the i-th IMF component from a relative perspective. The value of each row reflects the IMF component in the corresponding prior period. The relative confidence on , that is, the degree weight of the periodicity of this component dominated by the prior period. The formula is expressed as:

[0087] is the maximum value of the data in the i-th row.

[0088] S4: The energy weights of the intrinsic mode function components and the normalized characteristic periodic matrix are fused to construct the global periodic confidence vector.

[0089] It's important to note that the importance of different IMF components in the original signal is measured by their energy contribution. IMF components with higher energy contribute more to the overall signal and carry more representative periodic information. By weighting the energy contributions of the IMF components and integrating their periodic confidence scores at different prior periods, a global periodic confidence matrix is ​​constructed that reflects the periodic structure of the entire signal. This process quantifies the differences in the weights of different IMF components in expressing periodic information, ensuring that the final result better reflects the periodicity-dominant nature of the signal.

[0090] In one embodiment of the present application, in order to quantify the relative importance of each IMF component in the periodic expression and combine its periodic characteristic matrix to construct a periodic intensity representation that can reflect the overall signal periodic structure, the specific process is as follows: a. Calculate the weight of each IMF component based on the energy proportion;

[0091] b. Fusion of periodic features based on energy weights to generate a global periodicity confidence matrix.

[0092] The details are as follows:

[0093] a. Calculate the weight of each IMF component based on the energy proportion:

[0094] Each IMF component reflects the oscillation component of a certain frequency band in the original signal. The higher the energy of the IMF component, the greater its contribution to the overall signal, and the more representative the periodic information it carries.

[0095] The formula is:

[0096]

[0097] Among them, the weight is the energy proportion of a single IMF component in the overall signal, It is The total energy of the IMF components, It is The weight of each IMF component is calculated, where N is the total length of the time series and M is the number of IMF components. By calculating the energy of each IMF component and calculating the weight based on the energy proportion, the relative importance of each component in the entire signal is quantified.

[0098] b. Based on the energy weight, the period features are fused to generate a global period confidence vector:

[0099] In order to combine the periodic characteristics of each IMF component with its relative importance in the overall signal (i.e., energy weight), a more accurate global periodic confidence vector is obtained. .

[0100] Use Energy Weights Normalized characteristic periodic matrix Perform weighted summation on each column (i.e. each prior period) in , and get a The global periodic confidence vector of By constructing the formula:

[0101]

[0102] Among them, S is equal to the weighted sum of the confidence of each prior cycle according to the energy proportion. The contribution to the overall signal is obtained by taking the weighted average of the contributions (confidence) of all IMF components to the period; the weighting is the IMF weight calculated based on the energy in the previous step. , the impact of important components is greater.

[0103] At this point, by summarizing the normalized characteristic period matrices of all IMF components by column weights, we can obtain the period confidence vector reflecting the entire signal at each prior period. .

[0104] S5: Based on the periodic feature construction of the sliding window, the global periodic confidence vector is used as the input of the long short-term memory network to predict the changing trend of the target signal.

[0105] S6: Based on the changing trend of the target signal, the three-phase load is regulated.

[0106] In one embodiment of the application, for each time point , using its first N time points (forming a sliding window) to calculate the corresponding global cycle confidence vector, and obtain the global cycle confidence vector of the time series. These matrices are used as input to the LSTM model to predict future signal change trends.

[0107] Through this process, the model learns long-term dependencies in time series data and outputs predictions of current values ​​at future points in time. These predictions can be used to intelligently control three-phase loads, optimize energy distribution, and reduce three-phase imbalance, thereby enabling intelligent control of three-phase loads in industrial parks.

[0108] The above implementation scheme enables layer-by-layer decomposition of complex time series data using the empirical mode decomposition method. Different IMF components are weighted according to their energy proportions, ensuring that the periodic feature array in each component matches its contribution to the overall signal periodicity, enabling the LSTM model to more accurately capture periodic changes in the data. Furthermore, it enables adaptive matching of prior periods. By calculating the confidence level between the detected feature period and the prior period, it dynamically selects the optimal matching k value, improving the accuracy and robustness of periodic feature extraction. It can automatically match the prior period under different operating conditions, better meeting actual needs.

[0109] The above are all preferred embodiments of the present application, and are not intended to limit the scope of protection of the present application. Therefore, any equivalent changes made based on the structure, shape, and principle of the present application should be included in the scope of protection of the present application.

Claims

1. A three-phase load intelligent control method based on meter data, characterized in that: The following steps are involved: Acquire three-phase current time series data, and perform empirical mode decomposition on each phase current time series in the three-phase current time series data to obtain multiple intrinsic mode function components; The frequency domain information of each intrinsic mode function component is obtained by fast Fourier transform, the power spectrum density is calculated and the peak value is extracted to obtain the characteristic period set; By defining a priori period set and using a confidence index to evaluate the degree of matching between each period in the characteristic period set and the priori period set, a normalized characteristic period matrix is ​​constructed, including: setting a basic priori period and generating a priori period set based on different multiples, calculating the matching confidence of each characteristic period in the i-th intrinsic mode function component and each prior period in the priori period set, selecting the priori period with the highest confidence as the corresponding target matching period for each characteristic period, and statistically constructing a corresponding normalized characteristic period matrix based on each intrinsic mode function component; fusing the energy weights of the intrinsic mode function components and the normalized characteristic periodic matrix to construct a global periodic confidence vector; Based on the periodic feature construction of the sliding window, the global periodic confidence vector is used as the input of the long short-term memory network to predict the change trend of the target signal; regulating the three-phase load based on the change trend of the target signal; The statistical construction of the corresponding normalized characteristic period matrix includes: constructing an M×K characteristic period matrix for M intrinsic mode function components, initializing all elements to 0; performing statistics on the prior period and matching confidence obtained for each characteristic period in the characteristic period set of the i-th intrinsic mode function component; obtaining the column in the characteristic period matrix that has the largest matching confidence corresponding to the prior period. , and the The value of is set to the value corresponding to the maximum matching confidence, and a corresponding periodic characteristic matrix is ​​constructed; the periodic characteristic matrix is ​​normalized to obtain a normalized characteristic periodic matrix; The constructing of the global period confidence vector includes: calculating the energy weight of each intrinsic mode function component based on the energy proportion; performing weighted summation on each prior period in the normalized characteristic period matrix based on the energy weight to obtain a corresponding global period confidence vector; The periodic feature construction based on the sliding window uses the global periodic confidence vector as the input of the long short-term memory network to predict the change trend of the target signal, including: for each time point t n The corresponding global cycle confidence vector is calculated using the first N time points to obtain the corresponding time series global cycle confidence vector; the time series global cycle confidence vector is used as the input of the long short-term memory network to predict the change trend of the target signal.

2. The three-phase load intelligent control method based on electric meter data according to claim 1 is characterized in that: The performing of empirical mode decomposition on each phase current time series in the three-phase current time series data to obtain a plurality of intrinsic mode function components includes: Collect the a-phase current data I in the three-phase current time series data a ; The a-phase current data I a Empirical mode decomposition is performed to obtain a corresponding plurality of intrinsic mode function components, wherein the plurality of intrinsic mode function components respectively represent energy fluctuations at different frequency levels.

3. The three-phase load intelligent control method based on electric meter data according to claim 1 is characterized in that: Obtaining frequency domain information of each intrinsic mode function component by fast Fourier transform, calculating power spectrum density and extracting peak values ​​to obtain a characteristic period set includes: Perform fast Fourier transform on each intrinsic mode function component to obtain the corresponding spectrum expression; Performing a square operation on the frequency spectrum expression to obtain a corresponding power spectrum density to evaluate the energy distribution of each intrinsic mode function component at each frequency; Identifying a peak point in the energy distribution curve, where the peak point is the frequency corresponding to a local maximum of the power spectrum; Setting a power spectrum density threshold and screening out peak frequency points where the energy distribution is greater than the power spectrum density threshold; Based on each peak frequency point, calculate and obtain the target period value that meets the preset conditions; Based on the target period value, constructing a characteristic period set corresponding to the intrinsic mode function component; The periodic characteristic matrix is ​​normalized to obtain a normalized characteristic periodic matrix.

4. The three-phase load intelligent control method based on electric meter data according to claim 1 is characterized in that: If the a priori period has no corresponding characteristic period matching, the confidence level corresponding to the a priori period is defined as 0.

5. The three-phase load intelligent control method based on electric meter data according to claim 1 is characterized in that: A confidence model is constructed based on an exponential decay function, and a matching confidence between each characteristic period in the intrinsic mode function component and each prior period in the prior period set is calculated.

Images