CN111751650A - On-line monitoring system and fault identification method for non-intrusive household electrical equipment - Google Patents

Abstract

The invention relates to the technical field of power data analysis, and discloses a non-invasive household electrical equipment online monitoring system and a fault identification method, a non-invasive electrical signal acquisition device and a real-time power consumption information multi-feature extraction system to complete the household electrical equipment The acquisition of the generated waveform signals and the extraction of multivariate power consumption characteristics, the autoregressive moving average model ARMA, the multi-objective optimization model, and the LSTM classification system analyze and process the multivariate power consumption characteristics, and obtain each multivariate time series power consumption characteristic vector. The abnormal probability and normal probability of each currently running electrical equipment or its line, and finally the joint judgment model judges whether it is faulty according to the joint probability: when the joint abnormal probability > joint normal probability, then judge the current operating electrical equipment or its The line you are on is faulty. The present invention solves the technical problem that it is difficult to perform fault identification on household electrical equipment according to signals containing multiple electrical components, reduces the cost of fault identification, and improves the identification accuracy.

Description

On-line monitoring system and fault identification method for non-intrusive household electrical equipment

technical field

The invention relates to the technical field of power data analysis, in particular to a non-intrusive household electrical equipment online monitoring system and a fault identification method.

Background technique

Statistics show that more than 70% of household fire accidents are caused by electrical reasons, and the painful lessons of the accident have become a pain point for electric power practitioners. Most of the failures caused by the deterioration of electrical performance will appear device damage, short circuit, partial discharge and other phenomena, resulting in abnormal load characteristics. If the household electricity load can be monitored in real time and the fault condition can be effectively identified, it can respond in time at the early stage of the fault, cut off the power supply, and give an early warning to avoid the occurrence or expansion of the accident. Traditional load monitoring needs to set up a signal sampling device at the signal output end of each monitored equipment, and then analyze and process the sampled data. When the number of monitored equipment is large, the traditional monitoring method shows high cost and poor monitoring equipment. Disadvantages such as inconvenient installation and maintenance.

At present, some researchers have introduced the concept of non-intrusive type in the online monitoring of large-scale equipment in the power system, that is, collecting information such as voltage and current on the incoming line or bus. However, the research on the non-intrusive monitoring of the electricity consumption of residential users is still insufficient. It mainly stays on the analysis of electricity load, and the fault identification of household electrical equipment is almost blank.

Household electrical appliances include lighting appliances, cooking appliances, air conditioners, washing machines, etc. Each appliance has its own unique operating characteristics. The voltage and current curves of the same appliance are similar when it starts, runs, and shuts down, but it is not always possible at the same time. Only the same electrical appliance operates, so the signals monitored by non-intrusive equipment often contain components of multiple electrical appliances, which is also one of the difficulties in fault identification of electrical equipment.

SUMMARY OF THE INVENTION

In view of the above-mentioned deficiencies of the prior art, the present invention provides a non-invasive on-line monitoring system for household electrical equipment, which solves the technical problem that it is difficult to identify faults for household electrical equipment based on signals containing multiple electrical components.

In order to solve the above-mentioned technical problems, the present invention adopts the following technical scheme: a non-invasive household electrical equipment online monitoring system, including a non-invasive electrical signal acquisition device, a data preprocessing system, an LSTM classification system and a joint judgment Model;

The non-invasive electrical signal acquisition device is used to monitor the real-time waveform signal generated by the household electrical equipment; and upload it to the data preprocessing system through the communication system;

The data preprocessing system is used to extract the current multi-component power consumption characteristic data from the real-time waveform signal, and identify the current running electric equipment according to the current multi-component power consumption characteristic data; the multi-component power consumption characteristic data includes voltage and current. , A collection of various time-domain electricity characteristic data including power and phase angle;

The LSTM classification system includes a time series power consumption feature extraction unit and several LSTM classifiers; the time series power consumption feature extraction unit is used to process the current multi-dimensional power consumption characteristic data into corresponding current multi-dimensional time series power consumption characteristic data; Electrical characteristic data is a collection of various time series electrical characteristic vectors including voltage, current, power and phase angle;

The LSTM classifier is used to calculate the abnormal probability and normal probability of each currently running electrical equipment or its line under each multivariate time series power consumption characteristic vector in the current multivariate time series power consumption characteristic data; Types include single-device LSTM classifiers running for a single electrical device and multi-device LSTM classifiers running simultaneously for different devices; the LSTM classification system can select the corresponding LSTM classifier according to the identification results of the currently running electrical devices;

The joint judgment model is used to judge whether each currently running electrical equipment is faulty. For each currently running electrical equipment, it is judged according to the following criteria: if the joint abnormal probability > joint normal probability, then judge the currently running electrical equipment. The electrical equipment or its line is faulty; the joint abnormal probability is equal to the sum of the abnormal probabilities of the currently running electrical equipment or its line under each time series power consumption characteristic vector; the joint normal probability is equal to each time series power consumption characteristic vector The sum of the normal probabilities of the currently operating electrical equipment or its line.

Further, the data preprocessing system includes a real-time power consumption information multi-feature extraction system, a device power consumption feature database, an autoregressive moving average model ARMA, and a multi-objective optimization model;

The real-time power consumption information multiple feature extraction system is used to extract the current multiple power consumption feature data from the real-time waveform signal, and use the current multiple power consumption feature data as the current data;

The equipment power consumption characteristic database is used to store or update the following data: household power consumption basic characteristic data, historical data-classification result response relationship and real-time environmental data; the household power consumption basic characteristic data includes household personnel composition and electricity consumption habits ; The historical data-classification result response relationship includes historical data-electricity behavior response relationship and historical data-abnormal probability response relationship; historical data includes multivariate power consumption characteristic data at historical moments and corresponding seasonal data, weather data and date data ; Described real-time environmental data includes season data, weather data, date data corresponding to described current data;

The autoregressive moving average model ARMA includes an electricity consumption behavior prediction model established according to the historical data-electricity consumption behavior response relationship and basic household electricity consumption characteristic data; the autoregressive moving average model ARMA is used to predict the electricity consumption behavior through the Model, and predict the electricity consumption behavior according to real-time environmental data and current data, and then predict the corresponding electricity load according to the electricity consumption behavior;

The multi-objective optimization model is used to identify the currently running electrical equipment in combination with the expert knowledge base, the electricity load prediction results and the current data; the expert knowledge base contains multiple power consumption characteristics when a single electrical equipment is running Data and multivariate power consumption characteristic data when multiple power consumption devices are running at the same time.

Further, it also includes a power supply switch installed on the household power line, the power supply switch receives the remote control signal through the communication system, and when it is identified that the current running electrical equipment is faulty, the power supply switch can receive the remote control signal to turn off .

Further, the edge-cloud collaborative architecture is adopted: the autoregressive moving average model ARMA, the multi-objective optimization model, the LSTM classification system and the decision tree joint classification model are all configured in the edge server, and the equipment power consumption feature database is configured in the cloud server.

The present invention also provides a fault identification method for household electrical equipment, which is characterized by comprising the following steps:

Obtain the real-time waveform signal generated during the operation of household electrical equipment, and extract the current multi-component power consumption characteristic data from it; identify the currently operating electrical equipment according to the current multi-component power consumption characteristic data; the multi-component power consumption characteristic data includes voltage , a collection of various time-domain power consumption characteristic data including current, power and phase angle;

The current multivariate power consumption characteristic data is processed into the corresponding current multivariate time series characteristic data; the multivariate time series power consumption characteristic data is a collection of various time series power consumption characteristic vectors including voltage, current, power and phase angle;

According to the identification results of the currently running electrical equipment, select the corresponding LSTM classifier to process the current multivariate time series feature data:

If the identification result of the currently operating electrical equipment is that the currently operating electrical equipment is a certain type of single electrical equipment, then the current multivariate time series electrical consumption characteristic data is input into the corresponding type of single equipment LSTM classifier, and calculated in the current The abnormal probability and normal probability of a single currently running electrical equipment or its line under each time series power consumption characteristic vector in the multivariate time series power consumption characteristic data;

If the identification result of the currently operating electrical equipment is that the currently operating electrical equipment is a combination of several types of electrical equipment, input the current multivariate time series electrical power consumption feature data into the multi-device LSTM classifier of the corresponding combination type , calculate the abnormal probability and normal probability of each currently running electrical equipment or its line under each time series power consumption characteristic vector in the current multivariate time series power consumption characteristic data;

Calculate the corresponding joint abnormal probability and joint normal probability for each currently operating electrical equipment, and judge whether each currently operating electrical equipment fails according to the following criteria: If the joint abnormal probability > joint normal probability, then judge The current operating electrical equipment or its line is faulty; the joint abnormal probability is equal to the sum of the abnormal probabilities of the currently operating electrical equipment or its line under the power consumption characteristic vector of each time series; the joint normal probability is equal to each time series. The sum of the normal probabilities of the current running electrical equipment or its line under the electrical eigenvector.

Further, the real-time waveform signal generated during the operation of the household electrical equipment is collected by the non-intrusive electrical signal acquisition device;

Extract the current multi-dimensional power consumption feature data from the real-time waveform signal through the real-time power consumption information multi-feature extraction system;

Predict the electricity load by using the autoregressive moving average model ARMA: the autoregressive moving average model ARMA includes the electricity consumption behavior prediction model established according to the historical data-electricity consumption behavior response relationship and the basic characteristic data of household electricity consumption; The regression moving average model ARMA is used to predict the power consumption behavior according to the real-time environmental data and current data through the power consumption behavior prediction model, and then predict the corresponding power consumption load according to the power consumption behavior; the real-time environmental data includes: Seasonal data, weather data, date data corresponding to the current data; household electricity consumption basic characteristic data, historical data-electricity consumption behavior response relationship and real-time environmental data are all stored in the equipment electricity consumption characteristic database;

Through the multi-objective optimization model combined with the expert knowledge base, electricity load prediction results and current data, the current running electrical equipment is identified; the expert knowledge base contains multiple power consumption characteristic data and multiple power consumption characteristic data when a single electrical equipment is running. Multivariate power consumption characteristic data when each power consumption device is running at the same time.

Further, the autoregressive moving average model ARMA feeds back the electricity consumption behavior predicted according to real-time environmental data and current data to the equipment electricity consumption characteristic database, thereby updating the historical data-electricity consumption behavior response relationship in the equipment electricity consumption characteristic database. ; The LSTM classification system feeds back the abnormal probability of the currently running electrical equipment or its line under each time series power consumption characteristic vector to the equipment power consumption characteristic database, and the equipment power consumption characteristic database combines the real-time environmental data to historical data-abnormality The probability response relationship is updated.

Further, the LSTM classification system continuously uses the updated historical data-abnormal probability response relationship for training and updating; the autoregressive moving average model ARMA uses the updated historical data-electricity consumption behavior response relationship to update the electricity consumption behavior prediction model.

Compared with the prior art, the present invention has the following beneficial effects:

1. The online monitoring system of the present invention can separately identify the fault state of each currently running electrical equipment according to the waveform signal generated by the household electrical equipment, which greatly reduces the cost compared with the traditional monitoring method. The data preprocessing system is used to extract the current multi-dimensional power consumption characteristic data from the real-time waveform signal, and identify the current running power consumption equipment. Then the LSTM classification system performs data processing to obtain the abnormal probability of each current running electrical equipment under each time series power consumption feature vector, but the abnormal probability calculated based on a single time series power consumption characteristic vector is one-sided, and the accuracy rate Therefore, the present invention uses the joint judgment model to combine the abnormal probability of each time sequence power consumption feature to finally judge whether the current running power consumption equipment is faulty, which can greatly improve the fault identification accuracy.

2. Through the remote control of the power supply switch, the power supply switch can be controlled to close in time when a fault is identified, and the fire hazard can be eliminated in time.

3. The online monitoring system of the present invention adopts an edge-cloud coordination architecture, sinks calculation, analysis and real-time response to edge computing nodes, solves the problem of data traffic overload and timely response on the demand side, and better serves household electricity demand.

4. The present invention updates the historical data-classification results by feeding back the relevant classification results made according to the current data to the equipment power consumption characteristic database, and retrains the identification system using the updated historical data-classification results, thereby expanding the scope of the system. The training samples are used to update the parameters of the identification system, so that the performance of the system is continuously improved in the process of use.

Description of drawings

1 is a data flow diagram of a method for identifying faults of non-intrusive household electrical equipment in the present embodiment;

2 is a schematic diagram of the architecture of a non-intrusive household electrical equipment online monitoring system in this specific embodiment;

FIG. 3 is a flow chart of identifying the currently running electrical equipment in this specific implementation manner.

Detailed ways

The present invention will be further described in detail below with reference to the accompanying drawings and preferred embodiments.

Referring to Figure 1, a method for identifying faults of household electrical equipment includes the following steps:

Obtain the real-time waveform signal generated during the operation of household electrical equipment, and extract the current multi-component power consumption characteristic data from it; identify the currently operating electrical equipment according to the current multi-component power consumption characteristic data; the multi-component power consumption characteristic data includes voltage , a collection of various time-domain power consumption characteristic data including current, power and phase angle;

The current multivariate power consumption characteristic data is processed into the corresponding current multivariate time series characteristic data; the multivariate time series power consumption characteristic data is a collection of various time series power consumption characteristic vectors including voltage, current, power and phase angle;

According to the identification results of the currently running electrical equipment, select the corresponding LSTM classifier to process the current multivariate time series feature data:

If the identification result of the currently operating electrical equipment is that the currently operating electrical equipment is a certain type of single electrical equipment, then the current multivariate time series electrical consumption characteristic data is input into the corresponding type of single equipment LSTM classifier, and calculated in the current The abnormal probability and normal probability of a single currently running electrical equipment or its line under each time series power consumption characteristic vector in the multivariate time series power consumption characteristic data;

If the identification result of the currently operating electrical equipment is that the currently operating electrical equipment is a combination of several types of electrical equipment, input the current multivariate time series electrical power consumption feature data into the multi-device LSTM classifier of the corresponding combination type , calculate the abnormal probability and normal probability of each currently running electrical equipment or its line under each time series power consumption characteristic vector in the current multivariate time series power consumption characteristic data;

Calculate the corresponding joint abnormal probability and joint normal probability for each currently operating electrical equipment, and judge whether each currently operating electrical equipment fails according to the following criteria: If the joint abnormal probability > joint normal probability, then judge The current operating electrical equipment or its line is faulty; the joint abnormal probability is equal to the sum of the abnormal probabilities of the currently operating electrical equipment or its line under the power consumption characteristic vector of each time series; the joint normal probability is equal to each time series. The sum of the normal probabilities of the current running electrical equipment or its line under the electrical eigenvector.

The method for identifying faults of household electrical equipment in this specific embodiment adopts the following monitoring system to realize:

Referring to FIG. 1 and FIG. 2, a non-intrusive household electrical equipment online monitoring system includes a non-intrusive electrical signal acquisition device, a data preprocessing system, an LSTM classification system and a joint decision model;

The non-invasive electrical signal acquisition device is used to monitor the real-time waveform signal generated by the household electrical equipment; and upload it to the data preprocessing system through the communication system;

The data preprocessing system is used to extract the current multi-component power consumption characteristic data from the real-time waveform signal, and identify the current running electric equipment according to the current multi-component power consumption characteristic data; the multi-component power consumption characteristic data includes voltage and current. A collection of various time-domain power consumption characteristics data including power and phase angle; first extract the voltage time-domain power consumption characteristics and current time-domain power consumption characteristics from the real-time waveform signal, and then extract the voltage time-domain power consumption characteristics and the current time-domain power consumption characteristics from the real-time waveform signal. Extracting the power time-domain power consumption characteristics and the phase angle time-domain power consumption characteristics from the current time-domain power consumption characteristics;

The LSTM classification system includes a time series power consumption feature extraction unit and several LSTM classifiers; the time series power consumption feature extraction unit is used to process (slice by time) the current multi-dimensional power consumption characteristic data into corresponding current multi-dimensional time series power consumption characteristics. Data; multivariate time series power consumption characteristic data is a collection of various time series power consumption characteristic vectors including voltage, current, power and phase angle; voltage time series power consumption characteristic vector, current time series power consumption characteristic vector, power time series power consumption characteristic The vector and phase angle time series power consumption characteristic vectors are obtained by time slices from the voltage time domain power consumption characteristics, the current time domain power consumption characteristics, the power time domain power consumption characteristics and the phase angle time domain power consumption characteristics respectively;

The LSTM classifier is used to calculate the abnormal probability and normal probability of each currently running electrical equipment or its line under each multivariate time series power consumption characteristic vector in the current multivariate time series power consumption characteristic data; Types include single-device LSTM classifiers running for a single electrical device and multi-device LSTM classifiers running simultaneously for different devices; the LSTM classification system can select the corresponding LSTM classifier according to the identification results of the currently running electrical devices;

The joint judgment model is used to judge whether each currently running electrical equipment is faulty. For each currently running electrical equipment, it is judged according to the following criteria: if the joint abnormal probability > joint normal probability, then judge the currently running electrical equipment. The electrical equipment or its line is faulty; the joint abnormal probability is equal to the sum of the abnormal probabilities of the currently running electrical equipment or its line under each time series power consumption characteristic vector; the joint normal probability is equal to each time series power consumption characteristic vector The sum of the normal probabilities of the currently operating electrical equipment or its line.

Due to the variety of electrical appliances and the need for non-intrusive monitoring, that is, the mixed waveform signals generated by a variety of electrical appliances are detected at the household electrical wire, and the multi-feature extraction of information is performed in the following steps, that is, the household three-phase signal is extracted for a period of time. Information such as voltage, current, power, phase angle, etc., so the acquisition of these signals must have sufficient accuracy. In this specific embodiment, the non-intrusive electrical signal acquisition device is an instantaneous signal acquisition sensor with a sampling interval of less than milliseconds, such as an isolation transmitter, a transformer, and the like.

In this specific embodiment, the LSTM classification system continuously uses the updated historical data-abnormal probability response relationship for training and update; renew. By feeding back the relevant classification results based on the current data to the equipment power consumption feature database, updating the historical data-classification results, and using the updated historical data-classification results to retrain the identification system, the training samples are expanded, so that The parameters of the identification system are updated, so that the performance of the system is continuously improved during use.

In this specific embodiment, the communication system adopts a wireless communication system, such as wireless communication such as ZigBee, Lora, etc., to realize the preprocessing and security encryption of the perception information.

In this specific embodiment, it also includes a power supply switch installed on the household power line. The power supply switch receives a remote control signal through a communication system. When it is identified that the current running electrical equipment is faulty, the power supply switch can receive the remote control signal. signal to close.

In this specific implementation, the edge-cloud collaboration architecture is adopted, and the deployment on the BIE-AI-BOX container is realized under the Baidu open-source edge computing framework OpenEdge: autoregressive moving average model ARMA, multi-objective optimization model, LSTM classification system and decision-making The tree joint classification model is configured in the edge server, and the equipment power consumption characteristic database is configured in the cloud server.

In this specific embodiment, the data preprocessing system includes a real-time power consumption information multi-feature extraction system, a device power consumption feature database, an autoregressive moving average model ARMA, and a multi-objective optimization model.

The multi-feature extraction system for real-time electricity consumption information is used to extract current multi-component electricity consumption characteristic data from the real-time waveform signal, and use the current multi-component electricity consumption characteristic data as the current data; the multi-component electricity consumption characteristic data includes voltage A collection of various time-domain electrical characteristics data including current, power and phase angle.

The equipment power consumption characteristic database is used to store or update the following data: household power consumption basic characteristic data, historical data-classification result response relationship and real-time environmental data; the household power consumption basic characteristic data includes household personnel composition and electricity consumption habits ; The historical data-classification result response relationship includes historical data-electricity behavior response relationship and historical data-abnormal probability response relationship; historical data includes multivariate power consumption characteristic data at historical moments and corresponding seasonal data, weather data and date data ; The real-time environmental data includes seasonal data, weather data and date data corresponding to the current data.

The autoregressive moving average model ARMA includes an electricity consumption behavior prediction model established according to the historical data-electricity consumption behavior response relationship and basic household electricity consumption characteristic data; the autoregressive moving average model ARMA is used to predict the electricity consumption behavior through the Model, and according to the real-time environmental data and current data to predict the electricity consumption behavior, and then predict the corresponding electricity load according to the electricity consumption behavior.

Referring to Figure 3, the multi-objective optimization model is used to identify the currently running electrical equipment in combination with the expert knowledge base, the electricity load prediction results and the current data; the expert knowledge base includes the operation of a single electrical equipment. The multivariate power consumption characteristic data when multiple power consumption equipments are running simultaneously.

The autoregressive moving average model (ARMA) proposed by American statisticians JenKins and Box is an important method for studying time series. It is composed of an autoregressive model (AR) and a moving average model (MA). The long-term tracking research of user behavior can obtain the dynamic characteristics of the pattern change, obtain the prediction model, and combine with the current data to adaptively optimize the model. Electricity changes. This result supports the storage and processing of power consumption big data on the one hand, and also serves as the basis for decision-making of power dispatching on the other hand. The power consumption behavior analysis and power consumption load prediction performed by the autoregressive moving average model ARMA belong to the prior art, and the specific principle will not be repeated here. Please refer to Chinese patent (CN108073996A) "Urban Energy Panorama Interactive Big Data Platform Management System and Method".

The multi-objective optimization model needs to identify the running electrical equipment according to the electrical load, and the specific principle belongs to the prior art. Reference may be made to the Chinese Patent (CN111092434A) "Method and Device for Electric Load Control in Residential Areas Based on Non-Invasive Electricity Consumption Data".

The LSTM classification system includes multiple LSTM classifiers. The LSTM classifier is mainly composed of LSTM (Long Short-TermMemory) long and short-term memory network and fully connected network. Then the corresponding failure probability can be obtained after the output of the softmax function (used to make the output result meet the requirements of the probability distribution).

After identifying the currently running electrical equipment, find the corresponding LSTM model. For example, the current operating electrical equipment is a refrigerator and a microwave oven, then select the corresponding LSTM model for the simultaneous operation of these two electrical appliances, from the voltage sequence power consumption feature vector, current sequence power consumption feature vector, power The electrical eigenvectors of the time series and the phase angle time series use the electrical eigenvectors for LSTM processing.

For example, the voltage-sequential power consumption feature vector of a refrigerator and a microwave oven running at the same time is input into the multi-device LSTM network, and the following output is obtained: ) and the abnormal probability of the microwave oven and the normal probability of the voltage (the sum of which is 1). Current, power, phase angle, etc. are the same.

Due to the different effectiveness of each feature classifier in evaluating the fault state and lack of prior knowledge, it is necessary to make a joint judgment after obtaining the classifier constructed by each feature, and to give early warning and take protective measures such as cutting off the power supply according to the results.

For example, for a refrigerator running at the same time as a microwave oven, four sets of probabilities can be obtained by the multi-device LSTM classifier:

P _U = (P _U1 , P _U2 );

P _I = (P _I1 , P _I2 );

P _P = (P _P1 , P _P2 );

指在相角时序用电特性向量下的两个概率。Among them, _{PU refers to the two probabilities under the voltage sequence power consumption characteristic vector, P U1 is} the abnormal probability of the refrigerator under the voltage sequence power consumption characteristic vector, P _U2 is the normal probability of the refrigerator under the voltage sequence power consumption characteristic vector, And P _U1 +P _U2 =1. The same is true for the other three groups, _PI refers to the two probabilities under the power consumption characteristic vector of the current sequence, P _P refers to the two probabilities under the power consumption characteristic vector of the power sequence,

Refers to the two probabilities under the phase angle sequence power consumption characteristic vector.

则最终判决结果为电冰箱或其所在线路发生故障，否则结果为正常。Judgment Criteria: If

The final judgment result is that the refrigerator or its circuit is faulty, otherwise the result is normal.

Claims (10)

1. a non-invasive household electrical equipment online monitoring system, is characterized in that: comprise non-invasive electrical signal acquisition device, data preprocessing system, LSTM classification system and joint judgment model; The non-invasive electrical signal acquisition device is used to monitor the real-time waveform signal generated by the household electrical equipment; and upload it to the data preprocessing system through the communication system; The data preprocessing system is used to extract the current multi-component power consumption characteristic data from the real-time waveform signal, and identify the current running electric equipment according to the current multi-component power consumption characteristic data; the multi-component power consumption characteristic data includes voltage and current. , A collection of various time-domain electricity characteristic data including power and phase angle; The LSTM classification system includes a time series power consumption feature extraction unit and several LSTM classifiers; the time series power consumption feature extraction unit is used to process the current multi-dimensional power consumption characteristic data into corresponding current multi-dimensional time series power consumption characteristic data; Electrical characteristic data is a collection of various time series electrical characteristic vectors including voltage, current, power and phase angle; The LSTM classifier is used to calculate the abnormal probability and normal probability of each currently running electrical equipment or its line under each multivariate time series power consumption characteristic vector in the current multivariate time series power consumption characteristic data; Types include single-device LSTM classifiers running for a single electrical device and multi-device LSTM classifiers running simultaneously for different devices; the LSTM classification system can select the corresponding LSTM classifier according to the identification results of the currently running electrical devices; The joint judgment model is used to judge whether each currently running electrical equipment is faulty. For each currently running electrical equipment, it is judged according to the following criteria: if the joint abnormal probability > joint normal probability, then judge the currently running electrical equipment. The electrical equipment or its line is faulty; the joint abnormal probability is equal to the sum of the abnormal probabilities of the currently running electrical equipment or its line under each time series power consumption characteristic vector; the joint normal probability is equal to each time series power consumption characteristic vector The sum of the normal probabilities of the currently operating electrical equipment or its line. 2. A non-intrusive household electrical equipment online monitoring system, characterized in that: the data preprocessing system comprises a real-time electrical information multivariate feature extraction system, a device electrical feature database, an autoregressive moving average model ARMA, a target optimization model; The real-time power consumption information multiple feature extraction system is used to extract the current multiple power consumption feature data from the real-time waveform signal, and use the current multiple power consumption feature data as the current data; The equipment power consumption characteristic database is used to store or update the following data: household power consumption basic characteristic data, historical data-classification result response relationship and real-time environmental data; the household power consumption basic characteristic data includes household personnel composition and electricity consumption habits ; The historical data-classification result response relationship includes historical data-electricity behavior response relationship and historical data-abnormal probability response relationship; historical data includes multivariate power consumption characteristic data at historical moments and corresponding seasonal data, weather data and date data ; Described real-time environmental data includes season data, weather data, date data corresponding to described current data; The autoregressive moving average model ARMA includes an electricity consumption behavior prediction model established according to the historical data-electricity consumption behavior response relationship and basic household electricity consumption characteristic data; the autoregressive moving average model ARMA is used to predict the electricity consumption behavior through the Model, and predict the electricity consumption behavior according to real-time environmental data and current data, and then predict the corresponding electricity load according to the electricity consumption behavior; The multi-objective optimization model is used to identify the currently running electrical equipment in combination with the expert knowledge base, the electricity load prediction results and the current data; the expert knowledge base contains multiple power consumption characteristics when a single electrical equipment is running Data and multivariate power consumption characteristic data when multiple power consumption devices are running at the same time. 3 . The on-line monitoring system for non-invasive household electrical equipment according to claim 1 , wherein the non-invasive electrical signal acquisition device is an instantaneous signal acquisition sensor with a sampling interval of milliseconds or less. 4 . 4 . The non-invasive online monitoring system for household electrical equipment according to claim 1 , wherein the communication system adopts a wireless communication system. 5 . 5. The non-intrusive household electrical equipment online monitoring system according to claim 1, characterized in that: further comprising a power supply switch for being installed on the household power line, the power supply switch receives a remote control signal through a communication system, and when When it is identified that the current running electrical equipment is faulty, the power supply switch can receive a remote control signal to turn it off. 6. The non-intrusive home electrical equipment online monitoring system according to claim 2, characterized in that: using edge-cloud collaborative architecture: autoregressive moving average model ARMA, multi-objective optimization model, LSTM classification system and decision tree joint classification The models are all configured in the edge server, and the equipment power consumption characteristic database is configured in the cloud server. 7. A method for identifying faults of household electrical equipment, comprising the following steps: Obtain the real-time waveform signal generated during the operation of household electrical equipment, and extract the current multi-component power consumption characteristic data from it; identify the currently operating electrical equipment according to the current multi-component power consumption characteristic data; the multi-component power consumption characteristic data includes voltage , a collection of various time-domain power consumption characteristic data including current, power and phase angle; The current multivariate power consumption characteristic data is processed into the corresponding current multivariate time series characteristic data; the multivariate time series power consumption characteristic data is a collection of various time series power consumption characteristic vectors including voltage, current, power and phase angle; According to the identification results of the currently running electrical equipment, select the corresponding LSTM classifier to process the current multivariate time series feature data: If the identification result of the currently operating electrical equipment is that the currently operating electrical equipment is a certain type of single electrical equipment, then the current multivariate time series electrical consumption characteristic data is input into the corresponding type of single equipment LSTM classifier, and calculated in the current The abnormal probability and normal probability of a single currently running electrical equipment or its line under each time series power consumption characteristic vector in the multivariate time series power consumption characteristic data; If the identification result of the currently operating electrical equipment is that the currently operating electrical equipment is a combination of several types of electrical equipment, input the current multivariate time series electrical power consumption feature data into the multi-device LSTM classifier of the corresponding combination type , calculate the abnormal probability and normal probability of each currently running electrical equipment or its line under each time series power consumption characteristic vector in the current multivariate time series power consumption characteristic data; Calculate the corresponding joint abnormal probability and joint normal probability for each currently operating electrical equipment, and judge whether each currently operating electrical equipment fails according to the following criteria: if the joint abnormal probability > joint normal probability, then judge the current The operating electrical equipment or its line fails; among them, the joint abnormal probability is equal to the sum of the abnormal probabilities of the currently running electrical equipment or its line under the power consumption characteristic vector of each time series; the joint normal probability is equal to the abnormal probability of each time series. The sum of the normal probabilities of the current running electrical equipment or its line under the electrical eigenvector. 8 . The fault identification method for household electrical equipment according to claim 7 , wherein the real-time waveform signal generated during the operation of the household electrical equipment is collected by the non-intrusive electrical signal acquisition device; 9 . Extract the current multi-dimensional power consumption feature data from the real-time waveform signal through the real-time power consumption information multi-feature extraction system; Predict the electricity load by using the autoregressive moving average model ARMA: the autoregressive moving average model ARMA includes the electricity consumption behavior prediction model established according to the historical data-electricity consumption behavior response relationship and the basic characteristic data of household electricity consumption; The regression moving average model ARMA is used to predict the power consumption behavior according to the real-time environmental data and current data through the power consumption behavior prediction model, and then predict the corresponding power consumption load according to the power consumption behavior; the real-time environmental data includes: Seasonal data, weather data, date data corresponding to the current data; household electricity consumption basic characteristic data, historical data-electricity consumption behavior response relationship and real-time environmental data are all stored in the equipment electricity consumption characteristic database; Through the multi-objective optimization model combined with the expert knowledge base, electricity load prediction results and current data, the current running electrical equipment is identified; the expert knowledge base contains multiple power consumption characteristic data and multiple power consumption characteristic data when a single electrical equipment is running. Multivariate power consumption characteristic data when each power consumption device is running at the same time. 9 . The non-invasive household electrical equipment fault identification method according to claim 8 , wherein the autoregressive moving average model ARMA feeds back the electricity consumption behavior predicted according to real-time environmental data and current data to the equipment user. 10 . Electricity characteristic database, so as to update the historical data-electricity consumption behavior response relationship in the equipment electricity consumption characteristic database; The equipment power consumption characteristic database, and the equipment power consumption characteristic database is combined with real-time environmental data to update the historical data-abnormal probability response relationship. 10. The non-invasive household electrical equipment fault identification method according to claim 8, characterized in that: the LSTM classification system continuously utilizes the updated historical data-abnormal probability response relationship for training and updating; the autoregressive moving average model ARMA uses The updated historical data-electricity consumption behavior response relationship updates the electricity consumption behavior prediction model.

Images