A distribution box and its power monitoring system for the distribution box

Abstract

This application relates to the field of power monitoring technology, and specifically relates to a distribution box and a power monitoring system for the distribution box. The system includes: a data acquisition module for collecting parameters to form feature rows and feature groups; a reasonable judgment module for determining the final reasonableness of the feature groups according to the similarity of voltage fluctuations in the feature groups after considering the influence of environmental parameters on voltage fluctuations; a feature matrix construction module for determining the overall reasonableness based on the power difference between the feature groups and the corresponding feature rows and the final reasonableness, obtaining the unique future power based on the overall reasonableness and future power weighting, and determining the feature matrix; determining the regular contribution degree according to the overall reasonableness and future power, inputting the feature matrix to obtain the decision tree and the regular term, and subsequently iterating to obtain the future power at the current moment to complete power monitoring. This application improves the energy utilization efficiency.

Description

Technical Field

[0001] This application relates to the technical field of power monitoring, and particularly relates to a distribution box and a power monitoring system for the distribution box. Background Art

[0002] Predicting the future power of a distribution box for power dispatching and load management is an important part in the operation of a power system. Through accurate power prediction, the power distribution network can reasonably allocate power resources at different time periods, avoid overload, waste, and system failures, thereby improving energy utilization efficiency, ensuring power supply safety, and reducing operating costs.

[0003] Existing methods generally use XGBoost to train a model based on a large amount of historical power data to predict the data within the next minute of the distribution box. Specifically, first, all features related to future power need to be selected, and through feature engineering, a large amount of data is transformed into a feature matrix as the input of the model. Then, XGBoost fits the residuals of the current model by gradually constructing decision trees. The training objective of each tree is to optimize the error of the previous tree. At the same time, during the optimization process, in addition to minimizing the loss of the training data, a regularization term is added to prevent the model from overfitting. The model is continuously adjusted through the gradient boosting method, and finally, the prediction results of each tree are weighted and summed to improve the overall performance of the model.

[0004] In existing algorithms, collecting a large amount of historical data may result in many different future powers for the same feature group. At this time, it is difficult to obtain the true future power situation corresponding to this group of features, resulting in poor prediction ability. Moreover, the regularization term of the decision tree generated in each iteration process is closely related to the number of leaf nodes and the weights of leaf nodes. However, if the weights of some unavailable leaf nodes such as short circuits and open circuits learned by this decision tree are also large, XGBoost may regard these noise features as useful prediction information. At this time, the regularization term of this decision tree cannot well prevent the overfitting phenomenon. Summary of the Invention

[0005] To solve the technical problem of poor prediction ability, this application provides a distribution box and a power monitoring system for the distribution box. The specific technical solutions adopted are as follows:

[0006] In a first aspect, this application proposes a power monitoring system for a distribution box, which includes the following modules:

[0007] A data acquisition module, configured to collect parameters and future power through different sensors, use the parameters collected within a preset time as a feature group, and use the feature group and the future power as a feature row to construct an initial matrix. The parameters include power, ambient temperature, ambient humidity, ambient wind speed, and voltage fluctuation;

[0008] A rationality judgment module, configured to obtain the rationality of a feature group according to the similarity degree of all power and voltage fluctuations in the feature group; calculate the influence degree of environmental parameters on voltage fluctuations according to the correlation between all environmental parameters and voltage fluctuations and the correlation between environmental parameters and voltage fluctuations within a feature group; obtain the final rationality of the feature group according to the rationality of the feature group and the influence degree of environmental parameters on voltage fluctuations.

[0009] A feature matrix construction module, configured to obtain the overall rationality of a feature row according to the difference between the power of each feature group and the future power of the corresponding feature row and the final rationality of the feature group; weight the future power of the feature row with the overall rationality of the feature row to obtain the unique future power corresponding to the feature group; determine the feature rows of each feature group based on the unique future power to construct a feature matrix.

[0010] A power monitoring module, configured to obtain the regular contribution degree of a feature group according to the mean value of the overall rationality of each feature group and the mean value of the future power; take the feature matrix as an input to obtain a decision tree, and obtain the regular term of the decision tree by weighting with the regular contribution degree; obtain a prediction value based on the decision tree and iteratively update the decision tree with the future power to obtain a prediction model, and take the feature group corresponding to the current moment as the input of the prediction model to obtain the future power at the current moment, thereby completing power monitoring.

[0011] In the above solution, compared with the problem that there may be many different future powers in the same feature group and it is difficult to obtain the future power truly corresponding to this group of features, the present application starts from the feature rationality, gradually obtains the availability of all future powers corresponding to this group of features, and weights with the availability to obtain the future power truly corresponding to this group of features, solving the situation in the existing method that one feature group corresponds to multiple future powers, resulting in poor prediction effect due to difficult acquisition of future power; compared with the general XGBoost, in which the regular term of the tree generated in each iteration process is affected by some unavailable leaf node weights with large weights in case of sudden situations such as short circuits and open circuits, the present application weights the regular contribution degree with the leaf node weights, enabling the feature groups with regular contribution degrees to be better utilized, better preventing overfitting, and obtaining more accurate prediction results.

[0012] In one embodiment, the method for constructing the initial matrix is as follows:

[0013] Collect a preset time as an interval, and take the power collected in the next minute of the interval as the future power corresponding to the interval; take all parameter data collected in each interval as a feature group, and each feature group and the future power as a feature row, and take the feature row as a row of the matrix, and collect multiple feature groups and their corresponding future powers to construct an initial matrix.

[0014] In one embodiment, the method for obtaining the rationality of a feature group according to the similarity degree of all power and voltage fluctuations of a feature group is as follows:

[0015] Extract all the power of each feature group as a power feature sequence, and extract all the voltage fluctuations of each feature group as a voltage fluctuation feature sequence;

[0016] Calculate the cosine similarity between the power feature sequence and the voltage fluctuation feature sequence of each feature group, take the negative values of all the cosine similarities and normalize them by linear normalization, and use each normalized value as the rationality of each feature group.

[0017] In one embodiment, the method for calculating the influence degree of environmental parameters on voltage fluctuations according to the correlation between all environmental parameters and voltage fluctuations and the correlation between environmental parameters and voltage fluctuations within a feature group is as follows:

[0018] Sort all the collected environmental temperatures, environmental humidities, environmental wind speeds and voltage fluctuations in time series to obtain an overall temperature sequence, an overall humidity sequence, an overall wind speed sequence and an overall voltage fluctuation sequence;

[0019] Calculate the correlations between the overall voltage fluctuation sequence and the overall temperature sequence, the overall humidity sequence, and the overall wind speed sequence respectively as the influence of temperature on voltage fluctuations, the influence of humidity on voltage fluctuations, and the influence of wind speed on voltage fluctuations;

[0020] Sort all the environmental temperatures of a feature group in time series to obtain a temperature feature sequence, sort all the environmental humidities of a feature group in time series to obtain a humidity feature sequence, and sort all the environmental wind speeds of a feature group in time series to obtain a wind speed feature sequence;

[0021] Taking the influence of environmental characteristics on voltage fluctuations as the weight, weight the correlation between environmental characteristics and voltage fluctuations in each feature group to obtain the influence degree of environmental parameters on voltage fluctuations.

[0022] In one embodiment, the expression for the influence degree of environmental parameters on voltage fluctuations is:

[0023] , represents the environmental temperature, represents the environmental humidity, represents the environmental wind speed, represents the overall sequence corresponding to the Kth parameter in the ith feature group, represents the overall voltage fluctuation sequence in the ith feature group, represents the voltage fluctuation feature sequence in the ith feature group, represents the feature sequence corresponding to the Kth parameter in the ith feature group, Indicates the correlation between two sequences Indicates the influence degree of environmental parameters on voltage fluctuation in the i-th feature group

[0024] In one embodiment, the method for obtaining the final rationality of a feature group according to the rationality of the feature group and the influence degree of environmental parameters on voltage fluctuation is as follows:

[0025] , Indicates the influence degree of environmental parameters on voltage fluctuation in the i-th feature group Indicates the rationality of the i-th feature group Indicates the final rationality of the i-th feature group

[0026] In one embodiment, the method for obtaining the overall rationality of a feature row by combining the difference between the power of each feature group and the future power of the corresponding feature row with the final rationality of the feature group is as follows:

[0027] , Indicates power Indicates the power feature sequence in the feature group corresponding to the Indicates the average power of the power feature sequence in the feature group corresponding to the Indicates the final rationality of the feature group corresponding to the Indicates the normalization function Indicates the future power of the Indicates the overall rationality of the

[0028] In one embodiment, the method for obtaining the unique future power corresponding to a feature group by weighting the future power of a feature row with the overall rationality of the feature row and constructing a feature matrix based on the unique future power for each feature group's feature rows is as follows:

[0029] Count all the feature rows corresponding to a feature group, take the ratio of the overall rationality of each feature row to the sum of the overall rationalities of all the feature rows corresponding to the feature group as the weight of each feature row, and take the product of the weights of all the feature rows and the future power as the unique future power of the feature group;

[0030] Take each feature group and its unique future power as a row of the matrix, count all the feature groups and their corresponding unique future powers, and sort them in chronological order from top to bottom to obtain the feature matrix

[0031] In one embodiment, the method for obtaining the regularization term of the decision tree by weighting the regularization contribution is as follows:

[0032] , and are preset hyperparameters, represents the number of leaf nodes of the k-th iteratively obtained decision tree, represents the number of feature groups in the d-th leaf node, represents the regularization contribution of the u-th feature group in the d-th leaf node, represents the unique future power of the u-th feature group in the d-th leaf node, represents the regularization term of the k-th iteratively obtained decision tree.

[0033] In a second aspect, an embodiment of the present application further provides a distribution box, including a memory, a processor, and a computer program stored in the memory and running on the processor. When the processor executes the computer program, the steps of the power distribution box power monitoring system described in any one of the above are implemented.

[0034] The beneficial effects of the present application are as follows:

[0035] Compared with the problem that there may be many different future powers for the same feature group and it is difficult to obtain the real corresponding future power of this group of features, starting from the feature rationality, the present application gradually obtains the availability of all future powers corresponding to this group of features, and uses the availability to weight and obtain the real corresponding future power of this group of features, solving the situation in the existing method where one feature group corresponds to multiple future powers, resulting in poor prediction effects due to difficult acquisition of future powers; compared with the situation where the regularization term of the tree generated in each iteration process of general XGBoost is affected by unavailable leaf node weights with large weights in some unexpected situations such as short circuits and open circuits, the present application weights the regularization contribution through the leaf node weights, enabling the feature groups with regularization contributions to be better utilized, better preventing overfitting, and obtaining more accurate prediction results. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] In order to more clearly illustrate the technical solutions and advantages in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the following described drawings are only some embodiments of the present application. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0037] Figure 1 It is a flowchart of a power distribution box power monitoring system provided by an embodiment of the present application. DETAILED DESCRIPTION

[0038] To further elaborate on the technical means and effects adopted by this application to achieve the intended invention purpose, the following will, in conjunction with the accompanying drawings and preferred embodiments, elaborate in detail on a distribution box and its power monitoring system according to this application, including its specific implementation manners, structures, features, and effects. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. In addition, the specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0039] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the technical field to which this application belongs.

[0040] An embodiment of a distribution box and its power monitoring system:

[0041] The following will specifically describe the specific solution of a power monitoring system for a distribution box provided by this application in conjunction with the accompanying drawings.

[0042] Please refer to Figure 1 , which shows a flowchart of a power monitoring system for a distribution box provided by an embodiment of this application. The system includes the following modules:

[0043] A data acquisition module. In this application, the XGBoost model is used to predict the power of the distribution box in the next minute. The power, ambient temperature, ambient humidity, and ambient wind speed are respectively collected through an intelligent electricity meter, a temperature sensor, a humidity sensor, and a wind speed sensor. The maximum voltage and minimum voltage within one minute are collected through the intelligent electricity meter, and the difference between the maximum voltage and the minimum voltage is used as the voltage fluctuation. The power, ambient temperature, ambient humidity, ambient wind speed, and voltage fluctuation are used as the collected parameters.

[0044] Taking a preset minute as an interval, the interval corresponding time can be repeated. The power collected in the next minute of each interval is used as the future power corresponding to that interval. All the parameter data collected in each interval are used as a feature group. Each feature group and the future power are used as a feature row, and the feature row is used as a row of the matrix. Multiple feature groups and their corresponding future powers are collected to construct an initial matrix.

[0045] In this embodiment, the parameter data within 30 days are collected. Every 10 minutes is used as an interval. The parameter data from the start of collection to the 10th minute are used as a feature group, and this feature group and the power at the 11th minute are used as the first row of the initial matrix. The parameter data from the 2nd minute to the 11th minute are used as a feature group, and this feature group and the power at the 12th minute are used as the second row of the initial matrix, and so on.

[0046] Thus, the initial matrix is obtained.

[0047] The rationality judgment module. Each group of features in the initial matrix obtained by the data acquisition module corresponds to a future power. However, when collecting data, there may be a situation where the same group of features exists in different rows of the matrix, and the future powers corresponding to the same group of features are different. That is, the same group of features may correspond to multiple different future powers. Therefore, this application needs to select the most suitable future power from them.

[0048] Since this application needs to analyze the regularization contribution degree of each group of features through a decision tree later, and due to short circuits and open circuits, the data of the collected feature groups may have problems, resulting in analysis deviations. Therefore, it is first necessary to calculate whether each group of features is reasonable.

[0049] When the power in the distribution network increases, the voltage fluctuation decreases; when the power decreases, the voltage fluctuation increases. Therefore, for each group of features in the initial matrix, the more negative the correlation between power and voltage fluctuation, the higher the rationality of the group of features.

[0050] Extract all the powers of each group of features as the power feature sequence, and extract all the voltage fluctuations of each group of features as the voltage fluctuation feature sequence. The powers and voltage fluctuations in the power feature sequence and the voltage fluctuation feature sequence are sorted according to time series respectively.

[0051] Calculate the cosine similarity between the power feature sequence and the voltage fluctuation feature sequence, thereby obtaining the cosine similarity between the power feature sequence and the voltage fluctuation feature sequence of each group of features. After taking the negative values of all the cosine similarities, normalize them through linear normalization, and use each normalized value as the rationality of each group of features. The larger the normalized value, the more negative the correlation between power and voltage fluctuation, and the higher the rationality of the group of features.

[0052] Since voltage fluctuations are not necessarily all caused by power changes, but may also be due to environmental factors: when the temperature rises, the resistance of the wire increases, resulting in increased line losses and more significant voltage fluctuations during long-term high loads; when the humidity in the air increases, the conductivity may rise, resulting in voltage fluctuations in the power grid under some extreme conditions; wind speed may generate mechanical pressure on high-altitude transmission lines, transformers and other equipment, and may even cause loosening or damage of the lines, thus triggering power fluctuations.

[0053] Sort all the environmental temperatures, environmental humidities, environmental wind speeds and voltage fluctuations collected in this application according to time series to obtain the overall temperature sequence, overall humidity sequence, overall wind speed sequence and overall voltage fluctuation sequence.

[0054] Calculate the correlations between the overall voltage fluctuation sequence and the overall temperature sequence, the overall humidity sequence, and the overall wind speed sequence respectively, which are used as the influence of temperature on voltage fluctuation, the influence of humidity on voltage fluctuation, and the influence of wind speed on voltage fluctuation. In this embodiment, the method for calculating the correlations between the overall voltage fluctuation sequence and the overall temperature sequence, the overall humidity sequence, and the overall wind speed sequence respectively is the Pearson correlation coefficient.

[0055] Sort all the ambient temperatures of a feature group in time series to obtain the temperature feature sequence, sort all the ambient humidities of a feature group in time series to obtain the humidity feature sequence, and sort all the ambient wind speeds of a feature group in time series to obtain the wind speed feature sequence.

[0056] Taking the influence of environmental characteristics on voltage fluctuation as the weight, combine the correlations between the environmental characteristics and voltage fluctuation in each feature group to obtain the influence degree of environmental parameters on voltage fluctuation. The expression is:

[0057] , represents the ambient temperature, represents the ambient humidity, represents the ambient wind speed, represents the overall sequence corresponding to the Kth parameter in the ith feature group, represents the overall voltage fluctuation sequence in the ith feature group, represents the voltage fluctuation feature sequence in the ith feature group, represents the feature sequence corresponding to the Kth parameter in the ith feature group, represents the correlation between two sequences, represents the influence degree of environmental parameters on voltage fluctuation in the ith feature group.

[0058] Through the above steps, the influence degree of other environmental characteristics on voltage fluctuation is obtained. Therefore, it is inaccurate to determine the rationality of each feature group of the initial matrix through the mutual relationship between voltage fluctuation and power. Only the part that is not affected by the environmental influence degree is the part of the rationality truly determined through the mutual relationship between voltage fluctuation and power. Based on this, the final rationality of each feature group is obtained. The expression is:

[0059] , represents the influence degree of environmental parameters on voltage fluctuation in the ith feature group, represents the rationality of the ith feature group, represents the final rationality of the ith feature group.

[0060] Among them, the greater the final rationality of each feature group, the less likely the data corresponding to the feature group is caused by short circuit and open circuit.

[0061] The feature matrix construction module. Since the distribution box usually combines an automated control system for load regulation and the power changes slowly through regulation, the future power corresponding to each feature group cannot differ significantly from the power change of that feature group. Therefore, based on the final rationality, the overall rationality of each feature row is obtained by combining all the future powers corresponding to each feature group with the final rationality of that feature group. The expression is:

[0062] , represents power, represents the power feature sequence in the feature group corresponding to the th feature row, represents the average power of the power feature sequence in the feature group corresponding to the th feature row, represents the final rationality of the feature group corresponding to the th feature row, represents the normalization function, represents the future power of the th feature row, represents the overall rationality of the th feature row.

[0063] When constructing the feature matrix, the result of the future power corresponding to each feature group needs to be unique. Therefore, all the future powers corresponding to the same feature group are counted, and the future power of each feature row is weighted with the overall rationality of each feature row to obtain the unique future power of each feature group. The expression is:

[0064] , represents the overall rationality of the th feature row, represents the number of feature rows corresponding to the i-th feature group, represents the future power of the th feature row, represents the unique future power corresponding to the i-th feature group.

[0065] After that, the unique future powers of all different feature groups are obtained. Each different feature group and its corresponding unique future power are used as a row of the matrix, and the rows are sorted from top to bottom in chronological order. The obtained matrix is used as the feature matrix, and the number of columns of the feature matrix is the same as that of the initial matrix.

[0066] Power monitoring module. XGBoost is an algorithm based on Gradient Boosting, and its goal is to minimize the loss function by gradually building a series of decision trees. In each iteration, XGBoost constructs a new tree to fit the residuals (the residuals are the differences between the predicted values of the previous tree and the true values). To prevent overfitting, it is usually necessary to calculate the regularization term for each tree obtained in each iteration.

[0067] For each decision tree, the regularization term consists of the number of leaf nodes and the weights of the leaf nodes. The weight of a leaf node is the predicted value of this leaf node, and the predicted value of a leaf node is determined by the average value of the future power of all feature rows that can fall into this node. The contribution degrees of these feature rows to the regularization are the same. If there are some unavailable leaf nodes due to sudden situations such as short circuits or open circuits and their weights are also large, then the impact on regularization is very significant.

[0068] The larger the predicted value of a leaf node, the greater the regularization. In order to prevent the influence of short circuits and open circuits, its overall rationality should also be greater. Therefore, based on all the overall rationalities and future powers corresponding to each feature group, the regularization contribution degree of each feature group is determined, and the expression is:

[0069] , represents the overall rationality of the th feature row, represents the number of feature rows corresponding to the i-th feature group, represents the average value of all future powers corresponding to the i-th feature group, represents the normalization function, represents the regularization contribution degree of the i-th feature group.

[0070] Taking the feature matrix as the input, constructing a decision tree through XGBoost, and obtaining the regularization contribution degree of each feature group in the decision tree through the above steps. After constructing the decision tree, the number of leaf nodes can be known.

[0071] When determining the weight of each leaf node of each decision tree in this application, the regularization term of the decision tree is obtained by weighting the regularization contribution degrees of all feature groups falling on it:

[0072] , and are preset hyperparameters, represents the number of leaf nodes of the k-th decision tree obtained in the iteration, represents the number of feature groups in the d-th leaf node, represents the regularization contribution degree of the u-th feature group in the d-th leaf node, Represents the unique future power of the u-th feature group in the d-th leaf node. Represents the regularization term of the k-th decision tree obtained by iteration. In this application, and take values of 6 and 1.

[0073] Accordingly, the regularization term of the decision tree obtained in each iteration process can be corrected, and substituting it into the XGBoost iterative fitting process can obtain a better model for predicting the future power of the distribution box to prevent overfitting.

[0074] Input the feature group corresponding to the current moment into the model, and the power for the next minute at the current moment, that is, the future power at the current moment, can be obtained. Thus, the power monitoring of the distribution box is completed.

[0075] Based on the same inventive concept as the above method, an embodiment of the present invention also provides a distribution box, including a memory, a processor, and a computer program stored in the memory and running on the processor. When the processor executes the computer program, it implements any one of the modules in the above-mentioned power monitoring system of a distribution box.

[0076] It should be noted that: the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements on some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present application, and should all be included in the protection scope of the present application.

[0077] Each embodiment in this specification is described in a progressive manner. The same or similar parts between the embodiments can be referred to each other. Each embodiment focuses on the differences from other embodiments.

Claims

1. A power monitoring system for a distribution box, characterized in that: The system includes the following modules: The data acquisition module is used to collect parameters and future power through different sensors, take the parameters collected at the preset time as the feature group, and use the feature group and the future power as feature rows to construct an initial matrix. The parameters include power, ambient temperature, ambient humidity, ambient wind speed and voltage fluctuation; The rationality judgment module is used to obtain the rationality of a feature group according to the similarity of all power and voltage fluctuations in a feature group; calculate the influence of environmental parameters on voltage fluctuations according to the correlation between all environmental parameters and voltage fluctuations combined with the correlation between environmental parameters and voltage fluctuations in a feature group; The final rationality of the feature group is obtained according to the rationality of the feature group and the influence of environmental parameters on voltage fluctuation; The feature matrix construction module is used to obtain the overall rationality of the feature row according to the difference between the power of each feature group and the future power of the corresponding feature row and the final rationality of the feature group; weight the future power of the feature row with the overall rationality of the feature row as the weight to obtain the unique future power corresponding to the feature group; determine the feature row of each feature group based on the unique future power to construct the feature matrix; The power monitoring module is used to obtain the regular contribution of the feature group based on the mean of the overall rationality of each feature group and the mean of the future power; the feature matrix is ​​used as input to obtain the decision tree, and the regular term of the decision tree is obtained by weighting the regular contribution; the prediction value and future power are obtained based on the decision tree, and the decision tree is iterated to obtain the prediction model, and the feature group corresponding to the current moment is used as the input of the prediction model to obtain the future power at the current moment, thereby completing power monitoring.

2. A distribution box power monitoring system as claimed in claim 1, characterized in that: The method for constructing the initial matrix is: The preset collection time is taken as an interval, and the power collected in the next minute of the interval is taken as the future power corresponding to the interval; all parameter data collected in each interval are taken as feature groups, each feature group and future power are taken as feature rows, and the feature rows are taken as a row of the matrix. Multiple feature groups and their corresponding future powers are collected to construct the initial matrix.

3. A distribution box power monitoring system as claimed in claim 1, characterized in that: The method for obtaining the rationality of a feature group according to the similarity of all power and voltage fluctuations of a feature group is: Extract all powers of each feature group as a power feature sequence, and extract all voltage fluctuations of each feature group as a voltage fluctuation feature sequence; The cosine similarity of the power feature sequence and the voltage fluctuation feature sequence of each feature group is calculated, all cosine similarities are normalized by linear normalization after being negative, and each normalized value is used as the rationality of each feature group.

4. A distribution box power monitoring system as claimed in claim 3, characterized in that: The method for calculating the influence of environmental parameters on voltage fluctuations based on the correlation between all environmental parameters and voltage fluctuations combined with the correlation between environmental parameters and voltage fluctuations in a feature group is: All collected ambient temperatures, ambient humidity, ambient wind speeds and voltage fluctuations are sorted in time sequence to obtain an overall temperature sequence, an overall humidity sequence, an overall wind speed sequence and an overall voltage fluctuation sequence; Calculate the correlation between the overall voltage fluctuation sequence and the overall temperature sequence, the overall humidity sequence, and the overall wind speed sequence as the influence of temperature on voltage fluctuation, the influence of humidity on voltage fluctuation, and the influence of wind speed on voltage fluctuation; All ambient temperatures of a feature group are sorted in time series to obtain a temperature feature sequence, all ambient humidity of a feature group are sorted in time series to obtain a humidity feature sequence, and all ambient wind speeds of a feature group are sorted in time series to obtain a wind speed feature sequence; Taking the influence of environmental characteristics on voltage fluctuation as the weight, the correlation between environmental characteristics and voltage fluctuation in each feature group is weighted to obtain the influence of environmental parameters on voltage fluctuation.

5. A power distribution box power monitoring system as claimed in claim 4, characterized in that: The expression of the influence of the environmental parameters on the voltage fluctuation is: , Indicates the ambient temperature, Indicates the ambient humidity. Indicates the ambient wind speed, represents the overall sequence corresponding to the Kth parameter in the i-th feature group, represents the overall voltage fluctuation sequence in the i-th feature group, represents the voltage fluctuation feature sequence in the i-th feature group, represents the feature sequence corresponding to the Kth parameter in the i-th feature group, represents the correlation between two series. Represents the influence of environmental parameters in the i-th feature group on voltage fluctuation.

6. A distribution box power monitoring system as claimed in claim 1, characterized in that: The method for obtaining the final rationality of the feature group according to the rationality of the feature group and the influence of environmental parameters on voltage fluctuation is: , represents the influence of environmental parameters in the i-th feature group on voltage fluctuation, Indicates the rationality of the i-th feature group, Indicates the final rationality of the i-th feature group.

7. A power distribution box power monitoring system as claimed in claim 1, characterized in that: The method for obtaining the overall rationality of the feature row by combining the difference between the power of each feature group and the future power of the corresponding feature row with the final rationality of the feature group is: , Indicates power, Indicates The power feature sequence in the feature group corresponding to the feature rows, Indicates The average power of the power feature sequence in the feature group corresponding to the feature rows, Indicates The final rationality of the feature group corresponding to the feature rows, represents the normalization function, Indicates The future power of the characteristic line, Indicates The overall rationality of the feature line.

8. A power distribution box power monitoring system as claimed in claim 1, characterized in that: The method of weighting the future power of the feature row by taking the overall rationality of the feature row as the weight to obtain the unique future power corresponding to the feature group, and determining the feature row of each feature group based on the unique future power to construct the feature matrix is: Count all feature rows corresponding to a feature group, take the ratio of the overall rationality of each feature row to the sum of the overall rationality of all feature rows corresponding to the feature group as the weight of each feature row, and take the product of the weight of all feature rows and the future power as the unique future power of the feature group; Each feature group and unique future power are taken as a row of the matrix, all feature groups and their corresponding unique future powers are counted, and the feature matrix is ​​obtained by sorting them from top to bottom in chronological order.

9. A distribution box power monitoring system as claimed in claim 1, characterized in that: The method for obtaining the regularization item of the decision tree by weighting the regularization contribution is: , and To preset hyperparameters, represents the number of leaf nodes of the decision tree obtained by the kth iteration, represents the number of feature groups in the dth leaf node, represents the regular contribution of the u-th feature group in the d-th leaf node, represents the unique future power of the u-th feature group in the d-th leaf node, Represents the regularization term of the decision tree obtained in the kth iteration.

10. A distribution box, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that: When the processor executes the computer program, the steps of a distribution box power monitoring system as described in any one of claims 1-9 are implemented.

Images