CN-121234207-B - Meter box multi-parameter prediction method and system integrating big data processing

CN121234207BCN 121234207 BCN121234207 BCN 121234207BCN-121234207-B

Abstract

The application relates to a table box multiparameter prediction method and a system integrating big data processing, wherein the table box multiparameter prediction method comprises the steps of collecting original data of a table box running multidimensionally in real time through an RTU collecting module arranged in the table box, uploading the original data to a cloud server after encryption, cleaning, standardizing and extracting features to obtain structured data, storing the structured data to a distributed storage system, indexing and compressing the data, calling the structured data after compression to construct a multiparameter prediction model, performing supervision training on the multiparameter prediction model by utilizing historical statistical data to obtain a trained prediction model, inputting the original data collected in real time into the trained prediction model, outputting a single-parameter prediction result, and performing multiparameter collaborative analysis based on the single-parameter prediction result and the original data to generate a collaborative prediction result.

Inventors

LUO BIN
XU FEIYU
CHEN SHI
Tu Yanzhao

Assignees

日坤(北京)科技有限公司

Dates

Publication Date: 20260508
Application Date: 20250901

Claims (7)

1. The table box multiparameter prediction method integrating big data processing is characterized by comprising the following steps of: The method comprises the steps of acquiring original data of the operation of a meter box in a multi-dimension manner in real time through an RTU acquisition module arranged in the meter box, wherein the original data comprise temperature, current and voltage of a live wire and a zero wire, phase information, a box door switch state, position coordinates of the meter box and metering data of an ammeter; Transmitting the original data to an Internet of things device through a special communication chip, uploading the encrypted original data to a cloud server, and cleaning, standardizing and extracting features of the original data received by the cloud to obtain structured data; Storing the structured data into a distributed storage system, and indexing and compressing the data, wherein the distributed storage system stores real-time monitoring data, relational database storage equipment information and historical statistical data by adopting a time sequence database; The structured data after compression processing is called to construct a multi-task prediction model, wherein the multi-task prediction model comprises an LSTM neural network for predicting short-term trend of temperature and current and voltage, a XGBoost model for identifying fault types and a degradation model for evaluating ageing degree of a meter box, and historical statistical data is utilized to conduct supervision training on the multi-task prediction model to obtain a trained prediction model; inputting the real-time collected original data into the trained prediction model, outputting a single-parameter prediction result, and carrying out multi-parameter collaborative analysis based on the single-parameter prediction result and the original data to generate a collaborative prediction result; The method for constructing the multi-task prediction model by calling the structured data after compression processing comprises an LSTM neural network for predicting short-term trend of temperature and current and voltage, a XGBoost model for identifying fault types and a degradation model for evaluating ageing degree of a meter box, and performing supervision training on the multi-task prediction model by utilizing historical statistical data to obtain a trained prediction model, wherein the method comprises the following steps of: Extracting the historical statistical data from a distributed storage system, and dividing the historical statistical data into a training data set and a verification data set according to a preset proportion, wherein the training data set needs to comprise a normal operation sample and a fault sample; taking the training data set as input, constructing an initial multi-task prediction model comprising three collaborative sub-models, and performing supervision training by adopting a joint loss function; verifying trend prediction precision of an LSTM sub-model, fine tuning sub-model parameters which do not reach standards, and outputting the trained prediction model which passes verification; the method for constructing the initial multi-task prediction model comprising three collaborative sub-models by taking the training data set as input, and performing supervision training by adopting a joint loss function comprises the following steps: constructing an LSTM neural network submodel, wherein an input layer receives real-time parameters and time characteristics of current, voltage and temperature, 2 hidden layers are arranged, 64 neurons are arranged on each hidden layer, 3 neurons are arranged on an output layer, and three neurons of the output layer correspond to future trend predicted values of the temperature, the current and the voltage for 1 hour respectively; constructing XGBoost sub-models, enabling an input layer to receive characteristics of temperature, current and box door states and fault labels, and realizing fault type identification by learning a mapping relation between the characteristics and fault types through 10 decision trees; Constructing a degradation model, wherein an input layer receives aging related characteristics of temperature fluctuation frequency, humidity and operation age, and carries out aging degree assessment by fitting a table box performance attenuation curve through degradation parameters; Obtaining an initial multi-task prediction model comprising the three cooperative sub-models, wherein each sub-model realizes data cooperation through a shared input feature layer, and adopts a joint loss function to carry out supervision training; the step of inputting the real-time collected original data into the trained prediction model, outputting a single-parameter prediction result, and performing multi-parameter collaborative analysis based on the single-parameter prediction result and the original data to generate a collaborative prediction result, wherein the step of generating the collaborative prediction result comprises the following steps: based on the original data, preprocessing the original data into feature vectors compatible with a trained prediction model through format conversion and feature matching, and outputting preprocessed real-time input data; Based on the real-time input data and the trained prediction model, respectively outputting a temperature current voltage trend prediction value, a fault type probability distribution and an aging degree prediction value through an LSTM neural network, a XGBoost model and a degradation model to obtain the single-parameter prediction result; Based on the single parameter prediction result and the original data, constructing trend, state and aging collaborative analysis rules, and outputting collaborative analysis rules; And based on the collaborative analysis rule and the single parameter prediction result, integrating to generate a collaborative prediction result comprising short-term trend prediction, fault early warning level and ageing maintenance suggestion.
2. The method for predicting multiple parameters of a table box for integrated big data processing according to claim 1, wherein the steps of transmitting the original data to an internet of things device through a special communication chip, uploading the encrypted original data to a cloud server, and cleaning, normalizing and extracting features of the original data received by the cloud to obtain structured data comprise: transmitting the original data to an Internet of things device through a special communication chip built in an RTU, and outputting the original data transmitted to the Internet of things device, wherein the transmission rate of the special communication chip is matched with the sampling frequency of the original data; Encrypting the original data by adopting a preset encryption algorithm, and uploading the encrypted original data to a cloud server through an Internet of things device; Performing outlier rejection and missing value filling processing on the original data at the cloud to obtain cleaned pretreatment data; Converting parameters of different dimensions in the preprocessed data into unified dimensions, outputting standardized data, extracting time characteristics and state characteristics from the standardized data, and outputting structured data.
3. The method for predicting multiple parameters of a table box for integrated big data processing according to claim 2, wherein the storing the structured data in a distributed storage system, and indexing and compressing the data, the distributed storage system stores real-time monitoring data, relational database storage device information and historical statistics by using a time sequence database, and the method comprises: Distributing the structured data to different databases of a distributed storage system according to data types, namely distributing real-time monitoring data to a time sequence database, and distributing equipment information and the historical statistical data to a relational database; Performing downsampling compression processing on high-frequency acquisition data based on the real-time monitoring data, reserving peak value, valley value and time stamp key nodes of a data sequence, performing lossless compression on low-frequency data, and outputting the compressed real-time monitoring data; constructing a data index based on the equipment information and the historical statistical data of the relational database, establishing a primary key index of an equipment ID for the equipment information, establishing a joint index of a time stamp and a table box ID for the historical statistical data, and outputting indexed information; Based on the compressed real-time monitoring data and the index information, the two types of data are stored in an associated mode through a collaborative storage mechanism of the distributed storage system, and the structured data stored in the distributed storage system are output.
4. A method for predicting multiple parameters of a table box integrated with big data processing according to claim 3, wherein the obtaining an initial multi-task prediction model comprising the three collaborative sub-models, wherein each sub-model realizes data collaboration through a shared input feature layer and adopts a joint loss function for supervision training, comprises: regarding the LSTM sub-model, regarding the mean square error of the predicted trend value and the actual value as a loss term, regarding the XGBoost sub-model, regarding the fault recognition accuracy as an optimization target based on a cross entropy loss function, regarding the degradation model, regarding the absolute error of the predicted value and the actual evaluation value of the aging degree as a loss term; And iteratively optimizing each sub-model parameter through the joint minimization total loss function of the Adam optimizer, and outputting the preliminarily trained multi-task prediction model, wherein the feature sharing and performance linkage of each sub-model parameter are realized through collaborative training.
5. A table-box multiparameter prediction device integrating big data processing, characterized in that it comprises: The data acquisition module is used for acquiring original data of the operation of the meter box in a multi-dimension manner in real time through the RTU acquisition module arranged in the meter box, wherein the original data comprise temperature of a fire wire and a zero wire, current and voltage, phase information, a box door switch state, position coordinates of the meter box and metering data of an ammeter; The data processing module is used for transmitting the original data to the Internet of things device through the special communication chip, uploading the encrypted original data to the cloud server, and cleaning, standardizing and extracting features of the original data received by the cloud to obtain structured data; the data storage module is used for storing the structured data to a distributed storage system and carrying out indexing and compression processing on the data, and the distributed storage system stores real-time monitoring data, relational database storage equipment information and historical statistical data by adopting a time sequence database; the model training module is used for calling the structured data after compression processing to construct a multi-task prediction model, wherein the multi-task prediction model comprises an LSTM neural network for predicting short-term trend of temperature and current and voltage, a XGBoost model for identifying fault types and a degradation model for evaluating ageing degree of a meter box, and the multi-task prediction model is supervised and trained by utilizing historical statistical data to obtain a trained prediction model; the result prediction module is used for inputting the real-time collected original data into the trained prediction model, outputting a single-parameter prediction result, and carrying out multi-parameter collaborative analysis based on the single-parameter prediction result and the original data to generate a collaborative prediction result; The model training module is specifically used for extracting the historical statistical data from the distributed storage system, and dividing the historical statistical data into a training data set and a verification data set according to a preset proportion, wherein the training data set needs to comprise a normal operation sample and a fault sample; taking the training data set as input, constructing an initial multi-task prediction model comprising three collaborative sub-models, and performing supervision training by adopting a joint loss function; verifying trend prediction precision of an LSTM sub-model, fine tuning sub-model parameters which do not reach standards, and outputting the trained prediction model which passes verification; The model training module is also used for constructing an LSTM neural network sub-model, the input layer receives real-time parameters and time characteristics of current, voltage and temperature, 2 layers of hidden layers are arranged, 64 neurons are arranged on each hidden layer, 3 neurons are arranged on the output layer, and three neurons of the output layer correspond to future trend predicted values of the temperature, the current and the voltage respectively; constructing XGBoost sub-models, enabling an input layer to receive characteristics of temperature, current and box door states and fault labels, and realizing fault type identification by learning a mapping relation between the characteristics and fault types through 10 decision trees; Constructing a degradation model, wherein an input layer receives aging related characteristics of temperature fluctuation frequency, humidity and operation age, and carries out aging degree assessment by fitting a table box performance attenuation curve through degradation parameters; Obtaining an initial multi-task prediction model comprising the three cooperative sub-models, wherein each sub-model realizes data cooperation through a shared input feature layer, and adopts a joint loss function to carry out supervision training; The result prediction module is specifically used for preprocessing the original data into a feature vector compatible with a trained prediction model through format conversion and feature matching, and outputting preprocessed real-time input data; Based on the real-time input data and the trained prediction model, respectively outputting a temperature current voltage trend prediction value, a fault type probability distribution and an aging degree prediction value through an LSTM neural network, a XGBoost model and a degradation model to obtain the single-parameter prediction result; Based on the single parameter prediction result and the original data, constructing trend, state and aging collaborative analysis rules, and outputting collaborative analysis rules; And based on the collaborative analysis rule and the single parameter prediction result, integrating to generate a collaborative prediction result comprising short-term trend prediction, fault early warning level and ageing maintenance suggestion.
6. A control apparatus, characterized in that the apparatus comprises: Comprising a memory and a processor, said memory having stored thereon a computer program capable of being loaded by said processor and performing the method according to any of claims 1 to 4.
7. A computer readable storage medium, characterized in that a computer program is stored which can be loaded by a processor and which performs the method according to any of claims 1 to 4.

Description

Meter box multi-parameter prediction method and system integrating big data processing Technical Field The application relates to the technical field of intelligent meter boxes, in particular to a meter box multi-parameter prediction method and system integrating big data processing. Background The electric meter box is key equipment of the terminal distribution link of the electric power system, bears the core functions of electric energy metering, user electricity distribution and safety protection, and the running state of the electric meter box is directly related to the safety and stability of a power grid, the reliability of the user electricity and the management efficiency of electricity larceny prevention. With the deep promotion of smart power grid construction and the continuous increase of power loads, the quantity of meter boxes is increased rapidly, the distribution range is enlarged, and the modern power operation and maintenance requirements are difficult to meet by the traditional management mode relying on manual inspection and single parameter monitoring. At present, problems of overheat of a live wire/zero wire, abnormal current and voltage, illegal opening of a box door, unbalanced three-phase load, ageing of the meter box and the like are faced in the operation of the meter box, and line short circuit, power failure accidents, electricity stealing behaviors and equipment damage are easily caused. The manual inspection period is long, short-term faults cannot be captured in real time, single parameter monitoring in the meter box only pays attention to local indexes, and the synergistic influence of multiple parameters such as current, voltage, phase and box door state is ignored, so that fault misjudgment or missed judgment is easily caused. Based on the method, the application provides a table box multi-parameter prediction method and a system for integrated big data processing. Disclosure of Invention In order to solve the problems that the manual inspection period is long, short-term faults cannot be captured in real time, only local indexes are focused on for single parameter monitoring in a meter box, and the multi-parameter synergistic effect of current, voltage, phase, box door state and the like is ignored, so that fault misjudgment or missed judgment is easily caused, the application provides a meter box multi-parameter prediction method and a meter box multi-parameter prediction system integrating big data processing. In a first aspect, the application provides a table box multi-parameter prediction method for integrated big data processing, which adopts the following technical scheme: The method comprises the steps of acquiring original data of the operation of a meter box in a multi-dimension manner in real time through an RTU acquisition module arranged in the meter box, wherein the original data comprise temperature, current and voltage of a live wire and a zero wire, phase information, a box door switch state, position coordinates of the meter box and metering data of an ammeter; Transmitting the original data to an Internet of things device through a special communication chip, uploading the encrypted original data to a cloud server, and cleaning, standardizing and extracting features of the original data received by the cloud to obtain structured data; Storing the structured data into a distributed storage system, and indexing and compressing the data, wherein the distributed storage system stores real-time monitoring data, relational database storage equipment information and historical statistical data by adopting a time sequence database; The structured data after compression processing is called to construct a multi-task prediction model, wherein the multi-task prediction model comprises an LSTM neural network for predicting short-term trend of temperature and current and voltage, a XGBoost model for identifying fault types and a degradation model for evaluating ageing degree of a meter box, and historical statistical data is utilized to conduct supervision training on the multi-task prediction model to obtain a trained prediction model; Inputting the real-time collected original data into the trained prediction model, outputting a single-parameter prediction result, and carrying out multi-parameter collaborative analysis based on the single-parameter prediction result and the original data to generate a collaborative prediction result. Preferably, the transmitting the original data to the internet of things device through the special communication chip, uploading the encrypted original data to a cloud server, and cleaning, normalizing and extracting features of the original data received by the cloud to obtain structured data, including: transmitting the original data to an Internet of things device through a special communication chip built in an RTU, and outputting the original data transmitted to the Internet of things device, wherein the transmission rate of th