Power distribution network situation anomaly deduction analysis method and device based on multivariate control theory

Abstract

The invention discloses a power distribution network situation anomaly deduction analysis method and device based on a multivariate control theory, and the method comprises the following steps: obtaining power quality data of a power distribution network, carrying out the preprocessing of the power quality data, and obtaining a power quality multi-dimensional data set; constructing a voltage deviation mapping model, and training by using the multi-dimensional data set to obtain a trained voltage deviation mapping model; acquiring an electric energy quality index value by using the trained voltage deviation mapping model, and performing difference judgment on the acquired index value by adopting a three-level and four-level abnormality judgment mechanism; according to the method, an accurate mapping model from the source load power to the voltage quality is constructed, and cooperative monitoring and anomaly detection of power distribution network electric energy quality multi-dimensional parameters are realized through organic combination of a deep learning technology and a multivariate statistical control theory.

Description

Technical Field

[0001] The present invention belongs to the field of distribution networks, and particularly relates to a method and device for inferring and analyzing abnormal situations of distribution network postures based on a multi - control theory. Background Art

[0002] With the rapid development of distributed generation technology and the in - depth promotion of smart grid construction, the structure of the distribution network has become increasingly complex, and the power generation and consumption behaviors on the user side show diverse characteristics. This complexity has brought unprecedented challenges to the power quality management of the distribution network. Traditional power quality monitoring methods often only focus on parameters such as voltage and harmonics at a single node, and it is difficult to comprehensively grasp the mutual influence between nodes and their spatio - temporal distribution characteristics. Especially in low - voltage distribution areas, the radiation influence mechanism of power generation and consumption behaviors on the surrounding node voltages and power quality on the user side has not been fully studied, and the ability of predictive analysis of power quality problems and judgment of abnormal situations is seriously insufficient, resulting in an increase in the operation risk of the distribution network and affecting the power supply reliability and power quality level.

[0003] In view of the above problems, it is necessary to study a method for inferring power quality situations and analyzing spatio - temporal situation anomalies based on deep learning and multi - variable statistical control theory. Summary of the Invention

[0004] In order to solve the problems existing in the background art, the purpose of the present invention is to provide a method for inferring and analyzing abnormal situations of distribution network postures based on a multi - control theory. The method includes the following steps:

[0005] Obtain the power quality data of the distribution network and pre - process it to obtain a multi - dimensional power quality data set;

[0006] Construct a voltage deviation mapping model and train it using the multi - dimensional data set to obtain a trained voltage deviation mapping model;

[0007] Use the trained voltage deviation mapping model to obtain power quality index values, and use a three - level and four - grade abnormal judgment mechanism to judge anomalies for the obtained index values.

[0008] Further, the method for using a three - level and four - grade abnormal judgment mechanism to judge anomalies for the obtained index values is as follows:

[0009] First, at the first level, judge whether the power quality index exceeds 1.2 times the national standard limit. If it exceeds, trigger a four - level anomaly judgment (the most serious level); when judging at the second level, if it exceeds the national standard limit but does not reach 1.2 times the national standard limit, trigger a three - level anomaly judgment; if the power quality index is better than 70% of the national standard limit, it indicates that the power quality condition is good and no anomaly judgment signal is sent; for index values between 70% and the national standard limit of the national standard, enter the third - level anomaly detection, based on T 2The multivariate statistical control method of the control chart is used for anomaly detection. If an anomaly is detected, a secondary out-of-control signal is triggered, indicating the existence of a potential anomaly. If no anomaly is detected, a primary out-of-control signal is issued, indicating that the power quality index is close to the national standard limit. Through this multi-level out-of-control mechanism, the refined management of the power quality status is realized.

[0010] Furthermore, the voltage deviation mapping model is a five-layer fully connected neural network with residual connections.

[0011] Furthermore, the first layer of the voltage deviation mapping model converts the input dimension into a 64-dimensional hidden representation and applies the ReLU activation function. The second layer keeps the 64-dimensional hidden layer size unchanged and introduces a residual connection, directly adding the output of the first layer to the result after activation of the second layer to form cross-layer information transmission. The third layer continues to process the 64-dimensional feature representation and applies ReLU activation. The fourth layer applies a residual connection again, directly adding the output of the third layer to the result after activation of the fourth layer. The output layer maps the 64-dimensional hidden layer to the output dimension, corresponding to the voltage value prediction of all observation nodes. The specific input and output dimensions are determined by the number of source loads and the number of power quality observation points within the power system.

[0012] Furthermore, based on T 2 The method for anomaly detection using the multivariate statistical control method of the control chart is as follows:

[0013] S1. First, calculate the mean vector and covariance matrix S based on historical power quality data, and calculate the T 2 statistic:

[0014]

[0015] where T i 2 is the T 2 statistic of this data set at time i in history; is the power quality parameter vector composed of data at all observation points at time i, and its dimension is 1×h; is the historical mean vector;

[0016] Then identify the observation points where the T 2 statistic exceeds the T 2 upper control limit U CL , regard these points as potential anomaly points and eliminate them, and then recalculate the T 2 statistic with the remaining data until the relative increment of the number of anomaly points is less than 1%, and record the control upper limit at this time as the two-stage control upper limit U CL2 ;

[0017] S2. Use the historical mean vector and covariance matrix S to obtain the prediction data application stage, and the T at time b2 The calculation formula for the statistic is as follows:

[0018]

[0019] represents a vector composed of h power quality observation indicators at time b;

[0020] S3. If T b 2 the statistic exceeds T b 2 the control upper limit U CL2 , then the power quality data of the observation vector at time b is abnormal data.

[0021] Furthermore, the T 2 the control upper limit U CL is:

[0022]

[0023] t is the number of historical observation vectors, h is the number of power quality observation points, and F 1-α represents the 1-α quantile of the F-distribution with degrees of freedom h and t - h, and α is the significance level.

[0024] Furthermore, the method further includes: visualizing the outlier judgment result and determining a processing response strategy according to the visualization result;

[0025] The processing response strategy includes: for a level-four outlier judgment, immediate intervention and taking emergency measures are required; for a level-three outlier judgment, timely rectification and formulating a special plan are required; for a level-two outlier judgment, close monitoring and making preventive preparations are required; for a level-one outlier judgment, regular monitoring and attention are maintained.

[0026] A device for abnormal deduction and analysis of the distribution network situation based on multivariate control theory includes:

[0027] A multi-dimensional data set acquisition module, configured to acquire power quality data of the distribution network, preprocess it, and obtain a power quality multi-dimensional data set;

[0028] A trained voltage deviation mapping model acquisition module, configured to construct a voltage deviation mapping model and train it using the multi-dimensional data set to obtain a trained voltage deviation mapping model;

[0029] A power quality index value outlier judgment module, configured to use the trained voltage deviation mapping model to obtain power quality index values, and perform outlier judgment on the obtained index values using a three-level and four-grade outlier judgment mechanism.

[0030] A non-transitory computer-readable storage medium stores a computer program thereon, and when the computer program is executed by a processor, the method for inferring and analyzing the abnormal situation of the distribution network based on the multivariate control theory as described above is implemented.

[0031] An electronic device includes a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the program, the method for inferring and analyzing the abnormal situation of the distribution network based on the multivariate control theory as described above is implemented.

[0032] The beneficial technical effects of the present invention are as follows:

[0033] 1. The method for inferring the power quality situation and analyzing the spatio-temporal situation anomaly based on deep learning and multivariate statistical control theory provided by the present invention constructs an accurate mapping model from source-load power to voltage quality, overcoming the limitations of traditional power quality analysis methods that rely on complex parameters and physical models; through the organic combination of deep learning technology and multivariate statistical control theory, this method realizes the collaborative monitoring and anomaly detection of multi-dimensional parameters of the distribution network power quality, designs a refined grading judgment mechanism with three levels and four grades, can quickly and accurately infer the power quality situation of the entire network based on limited monitoring data, realizes the transformation of the management mode from "post-processing" to "foreseeing and preventing", provides strong support for the problem governance and judgment in areas with frequent power quality problems, and significantly improves the reliability of the distribution network operation and the power supply service quality.

[0034] 2. The multivariate statistical control theory of the present invention breaks through the limitations of the traditional national standard analysis mode and single-variable control charts. By establishing the covariance matrix between multiple variables, it captures the internal correlations and mutual influences of power quality problems between different nodes, and effectively identifies hidden anomalies and compound situation anomalies that are difficult to discover by single-variable control methods; combined with the mapping model from source-load power to voltage deviation constructed by the deep learning neural network, it can realize the comprehensive perception and dynamic inference of the power quality state of the distribution network, predict potential power quality problems in advance, and perform grading and anomaly judgment according to the degree of spatio-temporal situation anomaly; this power quality monitoring method based on multi-dimensional data analysis provides more comprehensive and accurate technical support for the operation and maintenance decision-making of the distribution network, and has important theoretical and practical value for improving the reliability of the distribution network operation and the power supply service quality. Description of the Drawings

[0035] Figure 1 is the logic block diagram of the method for inferring and analyzing the abnormal situation of the distribution network based on the multivariate control theory in the embodiment of the present invention;

[0036] Figure 2 is in the embodiment of the present invention based on T 2 The multivariate statistical anomaly judgment logic block diagram of the statistic;

[0037] Figure 3 It is the neural network structure diagram in the embodiment of the present invention;

[0038] Figure 4 It is the comparison between the actual example voltage deviation fluctuation curve and the prediction curve in Embodiment 2 of the present invention;

[0039] Figure 5 It is the grading and anomaly judgment result of the voltage deviation of the actual example in Embodiment 2 of the present invention. Specific implementation manner

[0040] The following further clearly and completely describes a power capacity market clearing method and device considering flexibility provided by the present invention with reference to the accompanying drawings:

[0041] Embodiment 1

[0042] This embodiment provides a method for inferring and analyzing the abnormal situation of the distribution network based on the multi - control theory. The method includes the following steps:

[0043] Obtain the power quality data of the distribution network and pre - process it to obtain a multi - dimensional power quality data set; specifically, collect the original data such as voltage, current, and power of each node in the distribution network, and perform pre - processing operations such as denoising, normalization, and time alignment to form a standardized multi - dimensional power quality data set, laying a foundation for subsequent analysis;

[0044] Construct a voltage deviation mapping model and train it using the multi - dimensional data set to obtain a trained voltage deviation mapping model; specifically, based on the pre - processed multi - dimensional power quality data set, select key monitoring points in the distribution network, use the power generation power and power consumption power of low - voltage users as input features, and the node voltage deviation as the output feature to construct a voltage deviation mapping model based on a deep - learning neural network. This model contains a multi - layer perceptron structure and captures the complex mapping relationship between source - load power and voltage quality through a non - linear activation function; when training the model: use the obtained multi - dimensional power quality data set to train the constructed neural network model, adopt the batch gradient descent method to optimize the model parameters, evaluate the model performance through cross - validation and perform hyperparameter tuning, and finally obtain a voltage deviation mapping model with high - precision inference ability;

[0045] The trained voltage deviation mapping model is used to obtain the power quality index values, and a three - level and four - grade anomaly determination mechanism is adopted for anomaly determination on the obtained index values. First, at the first level, it is judged whether the power quality index exceeds 1.2 times the national standard limit. If it exceeds, a grade - four anomaly determination (the most serious level) is triggered; if it exceeds the national standard limit but is less than 1.2 times the national standard limit, a grade - three anomaly determination is triggered; if the power quality index is better than 70% of the national standard limit, it indicates that the power quality condition is good and no anomaly determination signal is sent; for the index values between 70% and the national standard limit of the national standard limit, it enters the third - level anomaly detection, based on the T 2 The multivariate statistical control method of the control chart is used for anomaly detection. If an anomaly is detected, a grade - two anomaly determination is triggered, indicating the existence of a potential anomaly; if no anomaly is detected, a grade - one anomaly determination is sent, indicating that the power quality index is close to the national standard limit; this method conducts a multivariate joint analysis on the power quality time - series data of different observation points in the regional power grid, identifies the correlated abnormal fluctuations commonly shown by different measuring points, makes up for the lack of flexibility in the traditional rigid - threshold grading method, and can form a coupled criterion with the national standard to form a multi - level flexible abnormal disturbance identification method;

[0046] It can also be considered that the method further includes: visualizing the anomaly determination results and determining the processing response strategy according to the visualization results; the processing response strategy includes: grade - four anomaly determination requires immediate intervention and taking emergency measures; grade - three anomaly determination requires timely rectification and formulating a special plan; grade - two anomaly determination requires close monitoring and making preventive preparations; grade - one anomaly determination maintains routine monitoring and attention; specifically, integrating the voltage deviation mapping model and the anomaly determination mechanism based on the deep - learning neural network, developing a power quality situation deduction software platform. According to different anomaly levels corresponding to different processing response strategies, this platform can, based on the real - time input source - load power data, use the trained model to deduce the whole - network voltage quality distribution, and display the power quality anomaly detection results and anomaly levels, supporting power grid operation personnel for predictive management and decision - making optimization.

[0047] Specifically, the voltage deviation mapping model is a five - layer fully - connected neural network with residual connections. The first layer converts the n - dimensional input into a 64 - dimensional hidden representation and applies the ReLU activation function; the second layer keeps the 64 - dimensional hidden layer size unchanged, and at the same time introduces a residual connection, directly adding the output of the first layer to the result after activation of the second layer to form cross - layer information transfer; the third layer continues to process the 64 - dimensional feature representation and applies ReLU activation; the fourth layer applies a residual connection again, directly adding the output of the third layer to the result after activation of the fourth layer; the output layer maps the 64 - dimensional hidden representation to the h - dimensional output space, corresponding to the predicted voltage deviation values of h observation points. Among them, n represents the number of source - load nodes, and h represents the number of power quality observation points.

[0048] The described T2 The multivariate statistical control method for control charts includes:

[0049] In the first stage, first calculate the mean vector and covariance matrix S based on historical power quality data. Among them, the mean vector is the average value of each observation point in the historical power quality data:

[0050]

[0051] Among them, is the historical mean of the j-th observation point, t is the total number of samples of historical data, and x ij is the observed value of the j-th observation point at the i-th moment. The covariance matrix is a matrix with dimensions of h×h, and the calculation formula for each element is:

[0052]

[0053] Among them, S jk is the covariance between the data of the j-th observation point and the data of the k-th observation point, is the historical mean of the k-th observation point, and x ik is the observed value of the k-th observation point at the i-th moment.

[0054] Calculate the T 2 statistic, and its calculation formula is:

[0055]

[0056] In the formula, T i 2 is the T 2 statistic of this data set at the i-th moment in history; is the power quality parameter vector composed of the data of all observation points at the i-th moment, and its dimension is 1×h; is the historical mean vector, and its dimension is 1×h.

[0057] Then identify the observation points where the T 2 statistic exceeds the T 2 control upper limit U CL The calculation formula for the control upper limit is:

[0058]

[0059] t is the number of historical observation vectors, h is the number representing the power quality observation points, and F 1-α represents the 1-α quantile of the F distribution with degrees of freedom h and t - h, and α is the significance level.

[0060] First, the first control limit calculation is performed based on the global historical data, and the data outside the upper control limit are regarded as potential abnormal points and eliminated, and then the T is recalculated using the remaining data. 2 Statistics, until the relative increase of abnormal points is less than 1%, record the control upper limit at this time as the second stage control upper limit U CL2 .

[0061] Phase II, calculate T based on phase I parameters 2 Statistics, using historical mean vector and covariance matrix S, obtain the prediction data application phase, T at time b 2 The statistical calculation formula is:

[0062]

[0063] represents the vector composed of h power quality observation indicators at time b;

[0064] If T b 2 Statistics exceed T b 2 Upper control limit U CL2 , then the power quality data of the observation vector at time b is abnormal data. The proposed method fully considers the coupling relationship of multiple variables and carries out 2 The joint analysis of control charts enables the identification of spatiotemporal abnormal disturbance phenomena in power quality data at multiple points in the region.

[0065] Example 2

[0066] As an example, this embodiment provides a specific analysis example based on the method provided in Example 1:

[0067] The distribution network in a rural area of Henan Province was selected, and the annual historical monitoring data of 6 distributed power sources, 6 power loads and 23 key power quality observation points in the distribution network in a rural area of Henan Province were obtained. The data of the typical 30 days before the day were selected as the training data set of the neural network and the T 2 control Figure Ⅰ Stage T 2 Upper control limit U CL The calculation is based on the multivariate coupling relationship between voltage deviations at 23 observation points on a typical day, and the spatiotemporal situation anomaly analysis is carried out. By analyzing the coupling correlation of the power quality time series data of the 23 observation points in the historical and forecast data sets, the abnormal disturbance phenomenon of voltage deviation is identified, which makes up for the lack of flexibility in the rigid threshold warning method based on national standards. The specific steps are:

[0068] Clean and supplement the source load and power quality data obtained from area monitoring, and construct training data sets for the typical 30 days before spring and summer respectively;

[0069] Use a fully connected neural network to fit the non-linear relationship between source load information and power quality information. The specific input structure is as follows: The system uses a 12-dimensional feature vector as the input, and these features are divided into two categories: DG1-6 and L1-6. The former represents the active power output of distributed power sources, and the latter represents the active power of electrical loads. In the data preprocessing stage, these input features are standardized so that their mean is 0 and the standard deviation is 1 to improve the training efficiency and performance of the model. The standardized features are organized into batch data through a custom data structure for batch training of the neural network. The network structure is a five-layer fully connected neural network with residual connections. The first layer converts the 12-dimensional input into a 64-dimensional hidden representation and applies the ReLU activation function; the second layer keeps the size of the 64-dimensional hidden layer unchanged, and at the same time introduces a residual connection, directly adding the output of the first layer to the result after activation of the second layer to form cross-layer information transfer; the third layer continues to process the 64-dimensional feature representation and applies ReLU activation; the fourth layer is similar to the second layer, and a residual connection is applied again, directly adding the output of the third layer to the result after activation of the fourth layer; the output layer maps the 64-dimensional hidden representation to a 23-dimensional output space, corresponding to the voltage value prediction of 23 nodes. After the training loss function of the model converges, the trained pth file is exported for power quality prediction of typical days.

[0070] Based on the prediction data and the national standards for power quality of voltage deviation, divide the abnormal levels into level three and level four. If the voltage deviation is greater than 1.2 times the national standard threshold, it is judged as a level four abnormality; if the voltage deviation is greater than the national standard threshold, it is judged as a level three abnormality.

[0071] Based on historical data and prediction data, construct a T 2 multivariate statistical control chart for level one and level two abnormality classification. For the parameter estimation link, use the power quality data of the historical 30 days as the reference data set in stage I, and use the maximum likelihood estimation method to calculate the parameters of the multivariate normal distribution: the mean vector and the covariance matrix. In practical applications, to improve the robustness of parameter estimation, an iterative optimization mechanism is introduced. This mechanism first calculates the initial mean vector and covariance matrix using all historical data, and then identifies the T 2 values exceeding the upper control limit U CL of the observation points, regards these points as potential abnormal points and eliminates them, and then re-estimates the T 2 statistic using the remaining data until the relative increment of the number of abnormal points is less than 1%.

[0072] In stage I, first calculate the mean vector based on historical power quality data and the covariance matrix S. Among them, the mean vector is the average value of each observation point in the historical power quality data:

[0073]

[0074] Among them, is the historical mean of the j-th observation point, t is the total number of samples of the historical data, and x ij is the observed value of the j-th observation point at the i-th moment. The covariance matrix is a 23×23 dimensional matrix, and the calculation formula for each element is:

[0075]

[0076] Among them, S jk is the covariance between the data of the j-th observation point and the data of the k-th observation point, is the historical mean of the k-th observation point, and x ik is the observed value of the k-th observation point at the i-th moment.

[0077] Calculate the T 2 statistic, and its calculation formula is:

[0078]

[0079] In the formula, T i 2 is the T 2 statistic of this data set at the i-th moment in history; is the power quality parameter vector composed of the data of all observation points at the i-th moment, and its dimension is 1×23; is the historical mean vector, and its dimension is 1×23.

[0080] Then identify the observation points where the T 2 statistic exceeds the T 2 control upper limit U CL . The calculation formula for the control upper limit is:

[0081]

[0082] t is the number of historical observation vectors, h is the number representing the power quality observation points, and F 1-α represents the 1-α quantile of the F distribution with degrees of freedom h and t-h, and α is the significance level.

[0083] Regard the data outside its control upper limit as potential outliers and eliminate them, and then recalculate the T 2 statistic with the remaining data until the relative increment of the number of outliers is less than 1%, and record the control upper limit at this time as the two-stage control upper limit U CL2 .

[0084] Phase II: Calculate T based on the parameters in Phase I 2 Statistic, using the historical mean vector and covariance matrix S, obtain the T statistic at time b in the prediction data application stage 2 The calculation formula for the T statistic is:

[0085]

[0086] denotes the vector composed of voltage deviation indicators within 23 observation points at time b;

[0087] If the T b 2 statistic exceeds the upper control limit U b 2 of the T CL2 control stage, then the power quality data of this observation vector at time b is abnormal data. If the voltage deviation value of a certain node at a certain time is greater than 0.7 times the national standard threshold and is in an abnormal state, a secondary anomaly is given; if the voltage deviation value of the node at a certain time is greater than 4.9% but not in an abnormal state, a primary anomaly is given. The national standard is 7%. 4.9% is 0.7 times the national standard. (According to the requirements for power quality in the "Regulations on Power Supply Voltage, Grid Harmonics and Technical Line Loss Management of State Grid Corporation of China", if the monitoring value is less than 70% of the national standard, no early warning is carried out. This value is cited in this article, and the situation where the monitoring value is less than 70% is regarded as a situation with relatively good power quality, that is, the 70% national standard value is regarded as the excellent value of power quality)

[0088] In this implementation plan, the data of the 30 days before two typical days, April 9th in spring and July 24th in summer, are selected as the training data set of the neural network and the basis for calculating the upper control limit U 2 of the T Figure Ⅰ control stage, and the voltage deviation of the two typical days is deduced, predicted and analyzed for spatio-temporal situation anomalies. The constructed neural network structure is as CL shown. The voltage deviation prediction results of some nodes are as Figure 3 shown. The spatio-temporal situation anomaly analysis results of the voltage deviation on April 9th in spring and July 24th in summer are as Figure 4 shown. The spatio-temporal situation anomaly analysis results of the voltage deviation on April 9th in spring and July 24th in summer are as Figure 5As shown. From the uniform step distribution effect in the spatio-temporal situation anomaly analysis diagram, it can be seen that the trend anomaly recognition of the multivariate statistical control chart plays a buffering role in the phenomenon of exceeding the national standard limit, effectively identifying the data anomaly disturbance phenomenon between the first-level anomaly and the third- and fourth-level anomalies: During the power quality problem period of node V12 from 8 o'clock to 23 o'clock in summer, the first-level anomaly first occurred at this node, and then the anomaly level gradually increased, finally reaching the fourth-level anomaly. After that, the anomaly level gradually decreased again and exited the abnormal state. This gentle transition phenomenon of the anomaly level means that the national standard criterion and the multivariate statistical control chart criterion form a complementarity: The control chart accurately identified the T existing in the power quality time series data before the national standard limit was exceeded 2 The abnormal trend of the statistic, and made a second-level anomaly judgment in time.

[0089] Embodiment 3

[0090] This embodiment provides a device for inferring and analyzing the situation anomaly of a distribution network based on the multivariate control theory, including:

[0091] A multi-dimensional data set acquisition module, which is used to acquire the power quality data of the distribution network, preprocess it, and obtain a multi-dimensional power quality data set;

[0092] A trained voltage deviation mapping model acquisition module, which is used to construct a voltage deviation mapping model and train it using the multi-dimensional data set to obtain a trained voltage deviation mapping model;

[0093] A power quality index value anomaly judgment module, which is used to obtain the power quality index value using the trained voltage deviation mapping model, and perform anomaly judgment on the obtained index value using a three-level and four-grade anomaly judgment mechanism.

[0094] A non-transitory computer-readable storage medium is also provided, on which a computer program is stored. When the computer program is executed by a processor, it implements the method for inferring and analyzing the situation anomaly of a distribution network based on the multivariate control theory as described above.

[0095] Furthermore, the present invention adopts the following technical solutions:

[0096] An electronic device, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the program, it implements the method for inferring and analyzing the situation anomaly of a distribution network based on the multivariate control theory as described above.

[0097] Through the description of the above embodiments, those skilled in the art can clearly understand that the facilities of the present invention can be implemented by means of software plus a necessary general hardware platform. The embodiments of the present invention can be implemented using existing processors, or by dedicated processors used for this purpose or other purposes in a suitable system, or by a hardwired system. The embodiments of the present invention also include non-transitory computer-readable storage media, which include machine-readable media for carrying or having machine-executable instructions or data structures stored thereon; such machine-readable media can be any available media accessible by a general or special-purpose computer or other machine with a processor. For example, such machine-readable media can include RAM, ROM, EPROM, EEPROM, CD-ROM or other optical disk memories, magnetic disk memories or other magnetic storage devices, or any other medium that can be used to carry or store the required program code in the form of machine-executable instructions or data structures and can be accessed by a general or special-purpose computer or other machine with a processor. When information is transmitted or provided to a machine through a network or other communication connection (hardwired, wireless, or a combination of hardwired and wireless), this connection is also regarded as a machine-readable medium.

[0098] So far, the technical solution of the present invention has been described in conjunction with the preferred embodiments shown in the accompanying drawings. However, it is easy for those skilled in the art to understand that the protection scope of the present invention is obviously not limited to these specific embodiments. Without departing from the principle of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will fall within the protection scope of the present invention.

Claims

1. A method for inferring and analyzing the abnormal situation of a distribution network based on the multi - control theory, characterized in that, The method includes the following steps: Obtain the power quality data of the distribution network, preprocess it, and obtain a multi-dimensional dataset of power quality; Construct a voltage deviation mapping model, and use the multi-dimensional dataset for training to obtain a trained voltage deviation mapping model; Use the trained voltage deviation mapping model to obtain power quality index values, and use a three-level and four-grade anomaly determination mechanism to determine anomalies for the obtained index values.

2. The method for inferring and analyzing the abnormal situation of the distribution network based on the multi - element control theory according to claim 1, characterized in that, The method for using a three-level and four-grade anomaly determination mechanism to determine anomalies for the obtained index values is as follows: First, determine whether the power quality index exceeds 1.2 times the national standard limit. If it exceeds, trigger a fourth-level anomaly judgment; if it exceeds the national standard limit but does not reach 1.2 times the national standard limit, trigger a third-level anomaly judgment; if the power quality index is better than 70% of the national standard limit, it indicates that the power quality condition is good and no anomaly judgment signal is sent; for the index values between 70% of the national standard limit and the national standard limit, based on the T 2 control chart multivariate statistical control method for anomaly detection. If an anomaly is detected, trigger a second-level anomaly judgment, indicating the existence of a potential anomaly; if no anomaly is detected, issue a first-level anomaly judgment, indicating that the power quality index is close to the national standard limit.

3. The method for inferring and analyzing the abnormal situation of a distribution network based on the multi - element control theory according to claim 1, wherein, The voltage deviation mapping model is a five-layer fully connected neural network with residual connections.

4. The method for abnormal situation deduction and analysis of a distribution network based on the multi - element control theory according to claim 1, characterized in that The first layer of the voltage deviation mapping model converts the input dimension into a 64-dimensional hidden representation and applies the ReLU activation function; the second layer keeps the 64-dimensional hidden layer size unchanged, introduces a residual connection at the same time, and directly adds the output of the first layer to the result after activation of the second layer to form cross-layer information transmission; the third layer continues to process the 64-dimensional feature representation and applies ReLU activation; the fourth layer applies a residual connection again and directly adds the output of the third layer to the result after activation of the fourth layer; the output layer maps the 64-dimensional hidden layer to the output dimension, corresponding to the voltage value prediction of all observation nodes.

5. The method for abnormal situation deduction and analysis of a distribution network based on the multi - element control theory according to claim 1, characterized in that, Based on T 2 The method for anomaly detection based on the multivariate statistical control method of control charts is as follows: S1. First, calculate the mean vector and covariance matrix S based on historical power quality data, and calculate the T 2 statistic: where T i 2 is the T of this data set at time i in history 2 statistic; is the power quality parameter vector composed of data at all observation points at time i, and its dimension is 1×h; is the historical mean vector; Then identify T 2 The statistic exceeds T 2 The upper control limit U CL For the observation points, regard these points as potential outliers and eliminate them. Then recalculate the T statistic using the remaining data 2 until the relative increment of the number of outliers is less than 1%. Record the upper control limit at this time as the two-stage upper control limit U CL2 ; S2. Use the historical mean vector and the covariance matrix S to obtain the T statistic at time b during the prediction data application phase. 2 The calculation formula for the statistic is as follows: A vector composed of h power quality observation indexes at time b; S3. If T b 2 The statistic exceeds T b 2 The upper control limit U CL2 , then the power quality data of the observation vector at time b is abnormal data.

6. The method for inferring and analyzing the abnormal situation of a distribution network based on the multi - element control theory according to claim 5, wherein The said T 2 Upper control limit U CL is:[[]] t is the number of historical observation vectors, h is the number of power quality observation points, and F 1-α represents the 1-α quantile of the F-distribution with degrees of freedom h and t-h, where α is the significance level.

7. The method for inferring and analyzing the abnormal situation of the distribution network based on the multi - element control theory according to claim 5, wherein, The method further includes: visualizing the anomaly determination results, and determining a processing response strategy according to the visualization results; The processing response strategy includes: immediate intervention is required for a grade-four anomaly determination, and emergency measures should be taken; timely rectification is required for a grade-three anomaly determination, and a special plan should be formulated; close monitoring is required for a grade-two anomaly determination, and preventive preparations should be made; for a grade-one anomaly determination, regular monitoring and attention should be maintained.

8. A distribution network situation abnormal deduction and analysis device based on multi - control theory, characterized in that, It includes: A multi-dimensional dataset acquisition module, which is used to obtain the power quality data of the distribution network, preprocess it, and obtain a multi-dimensional dataset of power quality; A trained voltage deviation mapping model acquisition module, which is used to construct a voltage deviation mapping model and use the multi-dimensional dataset for training to obtain a trained voltage deviation mapping model; A power quality index value anomaly determination module, which is used to use the trained voltage deviation mapping model to obtain power quality index values, and use a three-level and four-grade anomaly determination mechanism to determine anomalies for the obtained index values.

9. A non-transitory computer-readable storage medium, on which a computer program is stored, characterized in that, When the computer program is executed by a processor, it implements the distribution network situation anomaly deduction analysis method based on the multi-element control theory as described in any one of claims 1 to 7.

10. An electronic device, comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the distribution network situation anomaly deduction analysis method based on the multi-element control theory as described in any one of claims 1 to 7.

Images