Data acquisition, aggregation, compression and transmission method and system based on low-voltage branch box

Abstract

The invention provides a data acquisition, aggregation, compression and transmission method and system based on a low-voltage branch box. The method comprises the following steps: acquiring related branch box operation state data information; performing aggregation processing on the data through standardization processing of the multi-source data on the basis of different difference characteristics of characteristic attributes of the acquired data sources; data are compressed and transmitted according to an arrangement sequence, high-speed transmission of the data is realized through priority ordering of the data, data standardization processing can be realized through data aggregation, the problem of time sequence fluctuation offset generated in a data acquisition process is eliminated to a certain extent, and the data acquisition efficiency is improved. According to the method, the dimension and magnitude unification of various data types is effectively realized, the data is compressed and transmitted according to the data arrangement sequence, key data and key data are transmitted preferentially, the abnormal state of the branch box can be analyzed as soon as possible, and the real-time performance of monitoring the state of the branch box is improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of data acquisition and transmission, and particularly relates to a data acquisition, aggregation, compression and transmission method and system based on a low-voltage branch box. Background Art

[0002] At present, there are still certain limitations in the status monitoring of branch boxes. In terms of data acquisition, electrical parameters such as voltage and current, as well as the temperature and humidity inside the box, are often collected in real time. However, these collected data are one-sided and cannot comprehensively represent the operating status of the low-voltage branch box. During the transmission process of the collected data, congestion problems often occur due to the large amount and disorder of the data, which greatly limits the improvement of the data transmission rate and affects the data transmission efficiency. Moreover, the collected data have inconsistent formats, which also makes data transmission complex and difficult.

[0003] The existing technologies mainly have the following deficiencies: 1) The types of collected data are single, and the specific situation of the branch box status cannot be grasped; 2) The types of collected data information are diverse and the formats are inconsistent, which brings difficulties to data transmission and subsequent processing; 3) At present, parallel communication may often be used for simultaneous transmission of multiple data, and the data transmission speed is faster, but the line layout is relatively complex, the cost is high, and it is not suitable for long-distance transmission. While unified communication data transmission can be carried out over long distances, but the transmission speed is slow. According to the on-site requirements, cost calculation and the problem of long power supply distance, data acquisition through a unified channel is more in line with the actual situation. However, when using a unified channel for data transmission, congestion problems often occur due to the large amount and disorder of the data, affecting the data transmission efficiency and resulting in data loss or loss of timeliness.

[0004] Therefore, the present invention proposes a data acquisition, aggregation, compression and transmission method and system based on a low-voltage branch box. Summary of the Invention

[0005] To solve the deficiencies in the existing technologies, the present invention provides a data acquisition, aggregation, compression and transmission method and system based on a low-voltage branch box, which can achieve standardized processing of data through data aggregation, eliminate the problem of timing fluctuation offset generated in the data acquisition process to a certain extent, effectively achieve the unity of the dimension and magnitude of various data types, and compress and transmit the data according to the data arrangement order, realizing the priority transmission of key data and important data, being able to analyze the abnormal status of the branch box early, and improving the real-time performance of the branch box status monitoring.

[0006] The present invention adopts the following technical solutions.

[0007] The present invention provides a data acquisition, aggregation, compression and transmission method based on a low-voltage branch box, including the following steps: S1: Collect the operation status data information of relevant distribution boxes. The data information includes voltage, current and other relevant electrical quantities, temperature and humidity environment information, water immersion image data, and smoke and gas data information caused by insulation heating. S2: After decomposing and processing the distribution box data information in step S1 according to the time series characteristics, use the clustering algorithm to calculate the overall consistency of the data. Select the corresponding standardization based on the different data formats of the distribution box data and use the obtained overall consistency for processing to complete the data aggregation transformation. S3: Use the fuzzy analytic hierarchy process for the distribution box data that has completed the data aggregation transformation in step S2 to determine the weights of each index, calculate the scores of each index to obtain the comprehensive score matrix, and determine the data arrangement order according to the scores. S4: Compress and transmit the sorted data in step S3 using the hybrid data compression method based on the convolutional neural network and the K-means clustering algorithm, and then transmit the data to the background.

[0008] Preferably, the calculation of the overall consistency of the data in step S2 specifically includes: Decompose and process the collected distribution box data according to the time series characteristics, and use it as the input for the standardization transformation in matrix form Implement the overall consistency of the data using the clustering algorithm. Let be the set of clusterers; The mean solution result of each cluster is , where represents the set of all data in the i th clusterer; For a single sample existing in the space, is the cluster label of the sample value. If , the i th sample clustering value , then the clustering result obtained based on the data set is ; The formula for the overall consistency is: ; In the formula: represents the clustering result; represents the overall consistency; n represents the number of data sets.

[0009] Preferably, after the collected distribution box data is processed by BC-Zscore data standardization: ; ; ; ; ; ; In the formula: represents the average value of the solution variable ; represents the variance of the solution variable ; n represents the number of data sets.

[0010] Preferably, the determination of the index weights in step S3 specifically includes: Perform pairwise comparison judgments on the importance of p evaluation indicators regarding a certain evaluation objective to construct an index discrimination matrix, and use the fuzzy analytic hierarchy process to determine the weights of each index; Construct the index discrimination matrix , and the matrix is rows columns matrix; The matrix A satisfies the following properties: ; Normalize the matrix A by columns to obtain the matrix , and the formula is: ; And add the rows of the matrix Q to get the vector , where: ; Hierarchical single sorting normalizes to obtain the eigenvector corresponding to the largest eigenvalue: ; And calculate the index information entropy , and the calculation formula is: ; In the formula: represents the number of data sets; represents the number of evaluation indicators; Thus, the weight values of each index can be obtained: 。

[0011] Preferably, the specific steps of calculating the comprehensive score matrix by calculating the scores of each index in step S3 include: Calculating each variable For the importance of the branch box, the calculation formula is: ; In the formula: represents the relaxation factor, ; represents the correlation of each variable with the branch box; Calculating each variable for the correlation with the branch box , and the correlation calculation formula is as follows; ; ; ; In the formula: represents the time series influence coefficient of each variable ; The calculation formula for the failure level is as follows: ; In the formula: N represents the number of failures of the branch box that result in serious failure consequences; represents the conditional probability of failure consequences after a failure; V represents the hazard degree of the failure; D represents the degree of influence of the branch box failure on power outage; o represents the weighted parameter of abnormal operation of the branch box; The calculation formula for the usage frequency is as follows: ; In the formula: t represents the number of hours of daily usage data; represents the number of daily usage times; The calculation formula for the sorting score is as follows: ; In the formula: P represents the comprehensive score matrix; Represents the importance index scoring vector; Represents the fault level index scoring vector; Represents the usage frequency index scoring vector.

[0012] Preferably, the learning process of the deep convolutional neural network includes convolutional layer learning and downsampling layer learning; In the convolutional layer learning, by optimizing the bias, the output result of the convolutional layer unit jointly constructed by several convolutions is output; In the downsampling layer learning, by determining the neuron distance through the cluster center of the cluster, after completing the neuron clustering process, an upsampling operation is performed on the downsampling layer, thereby realizing the learning process of the downsampling layer.

[0013] Preferably, there is a cluster center representing the weight for each neuron in each layer of the neural network, and the corresponding weight bias formula is: ; ; In the formula: Represents the l th j bias of the bias Represents the bias term; Represents the partial derivative value of the loss function with respect to the weight; Represents the partial derivative of the weight with respect to each cluster center of the l th layer; Represents the convolution kernel; Represents the l variance of the data in the Represents a constant used to enhance the stability of the variance; Represents the cluster center of the neuron clustering in each layer of the neural network; The output result of the convolutional layer unit jointly constructed by several convolutions is: ; ; In the formula: f Represents the activation function of the current layer; Represents the input mapping of the current layer to perform a convolution operation; Indicates the offset of the l feature of the j layer and the Indicates the length of the convolutional kernel of the l layer; Indicates the width of the convolutional kernel of the l layer; And respectively represent the stride.

[0014] Preferably, by calculating the l neurons of the l +1 layer to calculate the neurons of the l layer, using to represent the activation function of the layer neurons, and calculating the products of the activation function with the weight function and the gradient value respectively, that is, to obtain the neurons of the convolutional neural network ; In the formula: l represents the layer number; , represent the neurons before clustering; represents the probability that two neurons fail on the same test set simultaneously; represents the overlap coefficient.

[0015] Preferably, after the neuron clustering process is completed, an upsampling operation is performed on the downsampling layer. The specific process is as follows: ; ; In the formula: represents the l th neuron of the j layer after clustering; represents the sampling factor; represents l the activation function of the layer neurons; , represent the neurons before clustering; represents the sampling function; The formula for the learning process of the downsampling layer is as follows: ; In the formula: r represents the downsampling operation of the upsampling layer.

[0016] The present invention also provides a data acquisition, aggregation, compression, and transmission system based on a low-voltage branch box, which operates according to the aforementioned data acquisition, aggregation, compression, and transmission method based on a low-voltage branch box, and includes: A data information acquisition module, which is used to collect relevant branch box operation status data information by using HPLC; A data format processing module, which is used to select corresponding standardization processing according to different data formats deeply processed by using a clustering algorithm, and obtain a data source with unified transformation of dimension and magnitude; A data scoring and sorting module, which is used to determine the sorting score of the collected data according to the data sorting method that combines multiple parameters and index weights, and arrange the data according to the score; A data compression and transmission module, which is used to compress and transmit the sorted data by using a hybrid data compression method of a convolutional neural network and a K-means clustering algorithm, and transmit the data to the background.

[0017] The beneficial effects of the present invention are as follows. Compared with the prior art: (1) The data aggregation method proposed by the present invention effectively realizes the unity of dimension and magnitude of various data formats, and the overall consistency of the data in the method reduces the data redundancy phenomenon through the standardized processed data, reduces the data volume, makes the data more compact during compression, reduces the occupation of storage space, reduces the errors and inconsistencies in the data, thereby reducing the accumulation of errors during the compression process, ensuring that the decompressed data maintains the original quality, avoiding information loss caused by compression, and effectively eliminating the local detail differences of the data brought about by different acquisition environments, and ensuring the consistency of the data to a certain extent.

[0018] (2) The present invention proposes a data sorting method based on multi-parameter comprehensive evaluation. In the calculation process of using the fuzzy analytic hierarchy process to determine the index weights, the index information entropy is added, so that the calculation of the weights of each index has high credibility and accuracy, and the priority sorting is more accurately determined through multiple indexes, improving the accuracy of data sorting and providing a guarantee for the reasonable and accurate sorting result.

[0019] (3) The present invention proposes a hybrid data compression method combining a convolutional neural network and the K-means clustering algorithm. During the learning of the convolutional layer, by optimizing the bias, the accuracy loss is reduced, the storage space caused by excessive bias is decreased, and the data compression space is reduced. At the same time, neurons are clustered through the K-means clustering algorithm. During the learning of the downsampling layer, the distance between neurons is determined by the cluster center of the clustering clusters to achieve neuron clustering processing, reducing the number of neurons, the computational amount of the convolutional layer, and improving the data compression and transmission efficiency.

[0020] (4) During the data aggregation process, the overall consistency of the data is calculated, eliminating data differences, ensuring strong data consistency, replacing the work of data cleaning and preprocessing, and greatly reducing the workload of data aggregation, i.e., standardization processing.

[0021] (5) The present invention compresses and transmits data according to the arrangement order, solving the problem of data transmission congestion. Through the priority sorting of data, key data and important data are preferentially transmitted, effectively solving the current situation where all data needs to be uploaded for processing and monitoring the status of the branch box, facilitating the staff to timely master the status of the branch box, and reducing economic and safety problems caused by untimely status monitoring. Description of the Drawings

[0022] Figure 1 is a schematic flow chart of a data acquisition, aggregation, compression, and transmission method based on a low-voltage branch box in the present invention; Figure 2 is a flow chart of data aggregation processing in the present invention; Figure 3 is a flow chart of data transmission in the present invention. Specific Embodiments

[0023] To make the objectives, technical solutions, and advantages of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. The embodiments described in this application are only a part of the embodiments of the present invention, rather than all embodiments. Based on the spirit of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts fall within the protection scope of the present invention.

[0024] As Figure 1 shown, Embodiment 1 of the present invention provides a data acquisition, aggregation, compression, and transmission method based on a low-voltage branch box, including the following steps: S1: Collect the operation status data information of relevant branch boxes. The data information includes voltage, current, and other relevant electrical quantities, temperature and humidity environment information, water immersion image data, and smoke and gas data information caused by insulation heating.

[0025] For further illustration, in the implementation of the present invention, the sampling period of electrical quantities such as voltage and current is 250 ms, the sampling period of temperature and humidity environment information is 15 min, the sampling period of water immersion image data is 20 min, and the sampling period of smoke and gas data information caused by insulation heating is 3 min.

[0026] S2: Decompose and process the data information of the branch box in step S1 according to the time series characteristics, and then use the clustering algorithm to calculate the overall consistency of the data. Select the corresponding standardization based on the different data formats of the branch box data and use the obtained overall consistency for processing to complete the data aggregation transformation.

[0027] In the implementation of the present invention, since the formats, dimensions, data types, and orders of magnitude of the characteristic attributes of each data source are different during the operation of the branch box, in order to achieve the accuracy and precision of the data, it is necessary to eliminate the limitations caused by various inconsistent factors. Through the standardization processing of multi-source data, data aggregation is achieved, and thus the problem of time series fluctuation and offset generated during the data collection process is eliminated to a certain extent, effectively realizing the unification of the dimensions and orders of magnitude of various data types.

[0028] Preferably but not restrictively, refer to Figure 2 As shown, step S2 specifically includes the following steps: S2.1: Decompose and process the data of the branch box collected in step S1 according to the time series characteristics, and construct the collected data of the branch box X as the input for the standardization transformation; The X can exist in the form of multi-dimensional data, matrix form, and vector form. Let when appears in matrix form: ; In the formula: represents the i th variable of the j th observation value.

[0029] S2.2: Use the clustering algorithm to perform clustering operations on the multi-dimensional matrix X in step S2.1 to obtain the corresponding clustering results, and calculate the overall consistency of the data based on the clustering results.

[0030] First, let be the set of clusterers, and the mean solution result of each cluster is , where represents the set of all data in the i th clusterer; For a single sample existing in the space, is the clustering label of the sample value. If , for the i th sample clustering value , then based on the data set , the obtained clustering result is ; The formula for the overall consistency is: ; In the formula: represents the clustering result; represents the overall consistency; n represents the number of data sets.

[0031] S2.3: Set BC-Zscore standardization transformation schemes for different data formats according to the collected branch box data of different data formats; Preferably but not limited to, the step S2.3 specifically includes the following steps: Based on the collected branch box data X There are various data formats, including vector form, matrix form, and multi-dimensional data form. To ensure that the final data transformation processing is achieved for various data formats, BC-Zscore standardization transformation schemes for different formats are set respectively; S2.3.1: If the collected branch box data X exists in the form of a vector, the returned transformed result vector is: ; In the formula: represents the mean of the collected branch box data X when it is a vector; represents the variance of the collected branch box data X when it is a vector.

[0032] S2.3.2: If X exists in the form of a matrix, the data of the corresponding columns is standardized one by one using the mean and standard deviation of the column vectors of X , and the returned transformed result matrix is BC _ Z ; S2.3.3: If exists in the form of a multi-dimensional array, the mean and standard deviation are calculated along multiple dimensions of , and then is standardized, and the returned transformed high-dimensional array is .

[0033] S2.4: Process according to the selected corresponding standardized utilization to obtain overall consistency, and obtain a data source with unified transformation of dimension and magnitude.

[0034] The collected sectionalizer data After being processed by BC-Zscore data standardization: ; ; ; ; ; ; In the formula: represents the overall consistency; represents the solution variable the average value of; represents the solution variable the variance of; After being processed by BC-Zscore data transformation in each column of = 1, = 0.

[0035] S2.5: The collected sectionalizer data contains multiple sources with different formats. According to the selected multi-source data processing scheme, perform data iterative processing, and loop through steps S2-3 to step S2-4 to converge the data transformation processing results to obtain a data source with a unified format.

[0036] In the embodiments of the present invention, if the input data comes from multiple sources or has different formats, it is necessary to select a suitable processing scheme for each type of data. For example, different sensor data, different time series data, etc. may require different processing strategies; loop through steps S2.3 to step S2.4 to process each type of data in turn to ensure that they all undergo consistent standardized transformation; converge the standardized data together to obtain a data source with a unified format.

[0037] S2.6: Until all the data sources input for the task have undergone unified transformation of dimension and magnitude, perform task output saving for data calculation and data storage, and the data aggregation transformation ends.

[0038] S3: For the feeder pillar data after data aggregation transformation in step S2, use the fuzzy analytic hierarchy process to determine the weights of each index, calculate the scores of each index to obtain the comprehensive score matrix, and determine the data arrangement order according to the scores.

[0039] Preferably but not limitedly, step S3 specifically includes the following steps: S3.1: Make pairwise comparison judgments on the importance of p evaluation indexes regarding a certain evaluation objective to construct an index discrimination matrix, and use the fuzzy analytic hierarchy process to determine the weights of each index.

[0040] Construct the index discrimination matrix , and the matrix is rows columns matrix; The matrix A satisfies the following properties: ; Normalize the matrix A by columns to obtain the matrix , and the formula is: ; And add the rows of the matrix Q to get the vector , where: ; Hierarchical single sorting normalizes , that is, the eigenvector corresponding to the largest eigenvalue is obtained: ; And calculate the index information entropy , and the calculation formula is: ; In the formula: represents the number of data sets; represents the number of evaluation indexes; Thus, the weight values of each index can be obtained: .

[0041] S3.2: Calculate the importance, fault level and usage frequency of the feeder pillar operation data according to the feeder pillar data obtained in step S1.

[0042] Preferably but not limitedly, step S3.2 specifically includes the following steps: a. The importance refers to the severity level of the impact of data information on the safety and working load of the feeder pillar, and calculate each variable For the importance of the branch box, the calculation formula is as follows: ; In the formula: is the relaxation factor, ; represents the correlation of each variable with the branch box; The said each variable the correlation with the branch box , and the correlation calculation formula is as follows; ; ; ; In the formula: is the time series influence coefficient of each variable .

[0043] b. The fault level refers to the degree of influence of data information on the safe operation of the branch box when the branch box is in a fault state. The fault level calculation formula is as follows: ; In the formula: N represents the number of faults of the branch box that result in serious fault consequences; represents the conditional probability of fault consequences occurring after a fault; V represents the hazard degree of the fault; D represents the degree of influence of the branch box fault on power outage; o represents the weighted parameter of abnormal operation of the branch box; c. The usage frequency is the usage frequency requirement of information, that is, the usage frequency of data when the branch box is running. The usage frequency calculation formula is as follows: ; In the formula: t represents the number of hours of using data per day; represents the number of usage times per day.

[0044] S3.3: Calculate the sorting score of the collected data based on the weight values of each index obtained in step S3.1 and the importance, fault level, and usage frequency of each index obtained in S3.2. The calculation formula is as follows: ; In the formula: P represents the comprehensive scoring matrix; represents the importance index scoring vector; represents the fault level index scoring vector; represents the usage frequency index scoring vector.

[0045] Further explanation, according to the score, determine the data arrangement order. The score of 4.5 - 5.0 is critical data, 3.0 - 4.5 is important data, 2.0 - 3.0 is main data, and below 2.0 is non-essential data.

[0046] S4: Use the hybrid data compression method based on the convolutional neural network and the K-means clustering algorithm to compress and transmit the sorted data in step S3, and then send the data to the background.

[0047] Preferably but not restrictively, as Figure 3 shown, the specific steps of step S4 include the following steps: S4.1: Calculate the distance information of the features between the sorted data sets in step S3 through the K-means clustering algorithm, obtain the data information clustering labels , generate the cluster centers of the neuron clusters in each layer of the neural network , and use the convolutional neural network to divide the data set into a training set TR, a validation set V, and a test set TE, and classify the data in the validation set V.

[0048] S4.2: The data imported in step S3 and the clustering labels obtained in step S4.1 are respectively used as the inputs of the deep convolutional neural network, and the convolutional neural network and the clustering algorithm are used to learn the distance information between the data features.

[0049] Preferably but not restrictively, the specific steps of step S4.2 include the following steps: The learning process of the deep convolutional neural network consists of two links, namely convolutional layer learning and downsampling layer learning. The convolutional layer performs convolutional processing such as local connection and weight sharing on the data obtained from the previous layer based on the principle that the data information from the same source belongs to the same category, so as to reduce the number of connections and parameters.

[0050] The basic idea of weight sharing is that multiple network connections share a single weight value. As can be seen from step S4.1, each neuron in each layer of the neural network has a cluster center representing the cluster weight. Since the weights within the same cluster share the same weight value, only the index value of the cluster of the weight needs to be stored.

[0051] The convolutional layer is based on the principle that data information from the same source belongs to the same category and the network structure The formula for the weight bias is as follows: ; In the formula: represents the bias of the l th j feature of the bias layer; represents the partial derivative of the loss function with respect to the weight; represents the partial derivative of the weight with respect to each cluster center of the l th represents the convolutional kernel; represents the l th layer of the data variance; represents a constant used to enhance the stability of the variance.

[0052] and are respectively the length and width of the convolutional kernel of the l th layer, , where and are respectively their strides.

[0053] In the present invention, the output result of the convolutional layer unit constructed by several convolutions is described by the following formula; The formula for the output result of the convolutional layer unit is: ; ; In the formula: f represents the activation function of the current layer; represents the implementation of the convolutional operation on the input mapping of the current layer; represents the l th j th feature of the represents thel The length of the layer convolution kernel; Indicates the l width of the layer convolution kernel; and represent the stride respectively.

[0054] l The neuron in the +1 layer is the key to calculating the l neuron in the layer. It is represented by l the activation function of the neuron in the layer. Calculate the products of it with the weight function and the gradient value respectively, which is to obtain the neuron of the convolutional neural network

[0055] This invention borrows the clustering algorithm to perform upsampling operation on the downsampling layer.

[0056] ; In the formula: l represents the number of layers; ) represents the probability that two neurons fail simultaneously for the same test set; is the overlap coefficient.

[0057] After the neuron clustering process is completed, perform upsampling operation on the downsampling layer. The specific situation is as follows: ; ; In the formula: represents the l th neuron in the j layer after clustering is completed; represents the sampling factor; represents l the activation function of the neuron in the , represent the neurons before clustering; represents the sampling function.

[0058] The learning of the convolutional layer is prone to overfitting. The sampling layer calculates the upper limit value of the regional feature under the maximum pooling condition, so as to simplify the calculation process, improve the stability of the network model, and avoid the occurrence of overfitting problems. The learning process of the downsampling layer can be described by the following formula: ; In the formula: r represents the downsampling operation of the upsampling layer.

[0059] S4.3: Under the condition that the deep convolutional neural network converges, the last fully connected layer therein can be used as new data features, denoted as ; S4.4: According to obtain new clustering labels , and recalculate the cluster centers of each neuron, and use them as the output of the deep convolutional neural network in the next iteration process; S4.5: Loop the above process until the number of iterations reaches the set upper limit, select a denoising autoencoder, encode the finally obtained features into a specified coding length, and implement sparse autoencoding.

[0060] S4.6: Through the above process, a high-quality data coding effect can be obtained, data compression can be achieved, and the compressed data is transmitted to the background.

[0061] Embodiment 2 of the present invention also provides a data acquisition aggregation compression transmission system based on a low-voltage branch box, which runs the data acquisition aggregation compression transmission method based on a low-voltage branch box in Embodiment 1, including: A data information acquisition module, which is used to collect relevant branch box operation status data information by using HPLC; A data format processing module, which is used to select corresponding standardization processing according to different data formats deeply processed by using a clustering algorithm, and obtain a data source with unified transformation of dimension and magnitude; A data scoring and sorting module, which is used to determine the sorting score of the collected data based on a data sorting method that combines multiple parameters and index weights, and arrange the data according to the score; A data compression and transmission module, which is used to compress and transmit the sorted data by using a hybrid data compression method of a convolutional neural network and a K-means clustering algorithm, and transmit the data to the background.

[0062] The beneficial effects of the present invention are as follows. Compared with the prior art: (1) The data aggregation method proposed by the present invention effectively realizes the unification of the dimensions and magnitudes of various data formats. Moreover, the overall data consistency in the method reduces data redundancy through standardized data, decreases the data volume, makes the data more compact during compression, reduces the occupation of storage space, reduces errors and inconsistencies in the data, thereby reducing the accumulation of errors during the compression process, ensuring that the decompressed data maintains its original quality, avoiding information loss caused by compression, and effectively eliminating the local detail differences in the data due to different acquisition environments, guaranteeing the data consistency to a certain extent.

[0063] (2) The present invention proposes a data sorting method based on multi-parameter comprehensive evaluation. During the calculation process of determining the index weights by using the fuzzy analytic hierarchy process, the index information entropy is added, making the calculation of the weights of each index have higher credibility and accuracy. Moreover, the priority sorting is determined more accurately through multiple indicators, improving the accuracy of data sorting and providing guarantee for the reasonable accuracy of the sorting results.

[0064] (3) The present invention proposes a hybrid data compression method of convolutional neural network and K-means clustering algorithm. In the learning of the convolutional layer, through the optimization of the bias, the precision loss is reduced, the storage space caused by excessive bias is decreased, and the data compression space is reduced. At the same time, the neurons are clustered by the K-means clustering algorithm. In the learning of the downsampling layer, the neuron distance is determined through the cluster centers of the clustering clusters to realize the neuron clustering process, reducing the number of neurons, decreasing the computational amount of the convolutional layer, and improving the data compression and transmission efficiency.

[0065] (4) During the data aggregation process, the overall data consistency is calculated, eliminating the data differences, ensuring that the data has strong consistency, replacing the work of data cleaning and preprocessing, and greatly reducing the workload of data aggregation, i.e., standardized processing.

[0066] (5) The present invention compresses and transmits the data according to the arrangement order, solves the problem of data transmission congestion, realizes the priority transmission of key data and important data through the priority sorting of the data, effectively solves the current situation that all data needs to be uploaded before processing and monitoring the status of the branch box, facilitates the staff to timely master the status of the branch box, and reduces the economic and safety problems caused by untimely status monitoring.

[0067] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those of ordinary skill in the art should understand that: it is still possible to modify the specific implementation manners of the present invention or make equivalent replacements, and any modification or equivalent replacement that does not depart from the spirit and scope of the present invention shall be covered by the protection scope of the claims of the present invention.

Claims

1. A data acquisition, aggregation, compression and transmission method based on a low-voltage branch box, characterized in that It includes the following steps: S1: Collect the operation status data information of the relevant branch box. The data information includes voltage, current and other relevant electrical quantities, temperature and humidity environment information, water immersion image data, and smoke and gas data information caused by insulation heating. S2: Decompose and process the branch box data information in step S1 according to the time series characteristics, and then use the clustering algorithm to calculate the overall consistency of the data. Based on the different data formats of the branch box data, select the corresponding standardization to obtain the overall consistency for processing, and complete the data aggregation transformation. S3: Use the fuzzy analytic hierarchy process for the branch box data that has completed the data aggregation transformation in step S2 to determine the weights of each index, calculate the scores of each index to obtain the comprehensive score matrix, and determine the data arrangement order according to the scores. S4: Compress and transmit the sorted data in step S3 using a hybrid data compression method based on the convolutional neural network and the K-means clustering algorithm, and then transmit the data to the background.

2. A data acquisition, aggregation, compression and transmission method based on a low-voltage branch box according to claim 1, characterized in that: The specific calculation of the overall consistency of the data in step S2 includes: Decompose the collected branch box data according to the time series characteristics and present it in matrix form as the input for standardization transformation, and use the clustering algorithm to achieve the overall consistency of the data. Let be the set of clusterers; The mean solution for each cluster is , where represents the set of all data in the i th clusterer; For a single sample existing in space , is the clustering label of the sample value. If , the i th sample clustering value , then the clustering result obtained based on the data set is ; The calculation formula for the overall consistency is: ； In the formula: Indicates the clustering result; Indicates overall consistency; n Indicates the number of data sets.

3. A data acquisition, aggregation, compression and transmission method based on a low-voltage branch box according to claim 2, characterized in that: The collected sectionalizer data After being processed by BC-Zscore data standardization: ； ； ； ； ； ； In the formula: Indicates overall consistency; Denotes the average value of the solution variables ; Represents the solution variable Variance of 4. A data acquisition, aggregation, compression and transmission method based on a low-voltage branch box according to claim 1, characterized in that: The specific determination of the index weights in step S3 includes: Compare p the importance of each evaluation index for a certain evaluation goal pairwise to construct an index discrimination matrix, and use the fuzzy analytic hierarchy process to determine the weight of each index; Construct an index discrimination matrix , the matrix is rows matrix of columns; Matrix A Satisfies the following properties: ； Normalize the matrix A column by column to obtain the matrix , and the formula is: ； And add the rows of the matrix Q to obtain the vector , where: ； Hierarchical single sorting takes Normalization, that is, obtaining the eigenvector corresponding to the maximum eigenvalue : ； And calculate the index information entropy , and the calculation formula is as follows: ； In the formula: Indicates the number of data sets; Denotes the number of evaluation metrics; Thus, the weight values of each index can be obtained: 。 5. A data acquisition, aggregation, compression and transmission method based on a low-voltage branch box according to claim 1, characterized in that: The specific calculation of the scores of each index in step S3 to obtain the comprehensive score matrix includes: Calculate each variable For the importance of the branch box, the calculation formula is as follows: ； In the formula: denotes the relaxation factor, ; Indicates each variable Regarding the relevance of the branch box; Calculate each variable Regarding the relevance of the branch box , the relevance calculation formula is as follows; ； ； ； In the formula: Represents each variable Time series influence coefficient; The calculation formula for the fault level is as follows: ； In the formula: N The number of faults indicating that a branch box has a fault, resulting in serious fault consequences; Indicates the conditional probability of the consequences of a failure occurring after a failure; V Indicates the criticality of the failure; D Indicates the impact degree of the branch box fault on power outage; o Indicates the weighted parameter for abnormal operation of the branch box; The calculation formula for the usage frequency is as follows: ； In the formula: t Indicates the number of hours of data usage per day; Indicates the number of daily uses; The calculation formula for the sorting score is as follows: ； In the formula: P Denote the comprehensive scoring matrix; Indicates the importance index scoring vector; Indicates the scoring vector of the fault level index; Indicates the usage frequency index scoring vector.

6. A data acquisition, aggregation, compression and transmission method based on a low-voltage branch box according to claim 1, characterized in that: The learning process of the deep convolutional neural network includes convolutional layer learning and downsampling layer learning; In the convolutional layer learning, by optimizing the bias, the output result of the convolutional layer unit jointly constructed by several convolutions is output; In the downsampling layer learning, after determining the neuron distance through the cluster center of the clustering cluster to complete the neuron clustering process, an upsampling operation is performed on the downsampling layer, thereby realizing the learning process of the downsampling layer.

7. A data acquisition, aggregation, compression and transmission method based on a low-voltage branch box according to claim 6, characterized in that: Each neuron in each layer of the neural network has a cluster center representing the weight, and the corresponding weight bias formula is: ； ； In the formula: Indicates the l bias of the j nth feature of the mth layer; bias represents the bias term; Represents the partial derivative value of the loss function with respect to the weights; Indicates the partial derivative of the weight with respect to each cluster center in the l layer; Denotes a convolutional kernel; represents the l variance of the data of the Represents a constant used to enhance the stability of variance; Represent the cluster centers of neuron clusters in each layer of the neural network; The output result of the convolutional layer unit jointly constructed by several convolutions is: ； ； In the formula: f Represents the activation function of the current layer; Indicates that the input mapping of the current layer performs a convolution operation; Indicates the l bias of the j nth feature of the mth layer; Indicates the l length of the convolutional kernel of the Indicates the l width of the convolutional kernel of the layer; and respectively represent the step size.

8. A data acquisition, aggregation, compression and transmission method based on a low-voltage branch box according to claim 6 or 7, characterized in that: By calculation l The neurons of the +1 layer are calculated l The neurons of the layer, using To represent l The activation function of the neurons in the layer, and calculate the products of the activation function with the weight function and the gradient value respectively, which is to obtain the neurons of the convolutional neural network Perform upsampling operation on the downsampling layer through the clustering algorithm; Determine the neuron distance through the cluster center of the clustering cluster, and its formula is: ； In the formula: l Indicates the number of layers; , represent neurons before clustering; Indicates the probability that two neurons fail simultaneously for the same test set; Indicates the overlap coefficient.

9. A data acquisition, aggregation, compression and transmission method based on a low-voltage branch box according to claim 8, characterized in that: After the neuron clustering process is completed, an upsampling operation is performed on the downsampling layer. The specific process is as follows: ； ； In the formula: Indicating the l th neuron in the j th layer that has completed clustering; represents the sampling factor; representation l activation function of the layer neuron; , represent neurons before clustering; represents a sampling function; The sampling layer calculates under the condition of maximum pooling The upper limit value of the regional feature. The formula for the learning process of the downsampling layer is as follows: ； In the formula: r Represents the downsampling operation of the upsampling layer.

10. A data acquisition, aggregation, compression, and transmission system based on a low-voltage branch box, which operates according to a data acquisition, aggregation, compression, and transmission method based on a low-voltage branch box as described in any one of claims 1-9, characterized in that It includes: A data information acquisition module, which is used to collect the operation status data information of relevant branch boxes by using HPLC; A data format processing module, which is used to select corresponding standardization processing according to different data formats after in-depth processing by using a clustering algorithm, and obtain a data source with unified transformation of dimension and magnitude; A data scoring and sorting module, which is used to determine the sorting score of the collected data based on a data sorting method that combines multiple parameters and index weights, and arrange the data according to the score; A data compression and transmission module, which is used to compress and transmit the sorted data by using a hybrid data compression method of a convolutional neural network and a K-means clustering algorithm, and transmit the data to the background.

Images