CN110516837B

CN110516837B - AI-based intelligent diagnosis method, system and device

Info

Publication number: CN110516837B
Application number: CN201910620275.8A
Authority: CN
Inventors: 马欣; 孙钊
Original assignee: Individual
Current assignee: Individual
Priority date: 2019-07-10
Filing date: 2019-07-10
Publication date: 2020-12-18
Anticipated expiration: 2039-07-10
Also published as: CN110516837A

Abstract

The invention relates to the field of smart power grids, in particular to an AI-based intelligent diagnosis method, system and device. The core is AI + IoT, which is composed of standardized and industrialized IoT edge computing hardware equipment and a customized AI module, and comprises: collecting power utilization data by a collecting unit of the IoT equipment; the cloud platform trains an AI model according to historical data, the AI model is continuously and dynamically perfected according to new data, and the edge computing unit of the IoT equipment inputs collected power utilization data into the trained artificial intelligence AI model to obtain related data. The real-time prediction of treatment is realized, the real-time monitoring level of abnormal and standard exceeding of energy indexes is practically improved, and the working mode of detection and manual verification of the former semi-manual system is changed into a full-automatic intelligent AI analysis mode. Finally, the brand-new software and hardware forms of the cloud platform, the edge computing device, the intelligent field terminal and the intelligent IoT chip set module are realized, the user-side energy system and the power distribution management system are redefined, and the era of software-defined energy is started.

Description

AI-based intelligent diagnosis method, system and device

Technical Field

The invention relates to the field of smart power grids, in particular to an AI-based intelligent diagnosis method, system and device.

Background

At present, the construction of ubiquitous power internet of things is steadily carried out, and tasks in six fields such as basic support, data sharing, internal and external services and the like are successively carried out. The line loss is one of core economic and technical indexes of a power grid enterprise, the operating cost and the economic benefit of the enterprise are reflected, and the strengthening of line loss management is a long-term strategic task and system engineering of the power grid enterprise. The intelligent power grid is a multi-target function with balance as constraint, the factors are extensive, the conditions are random, the line loss conditions are variable, the cause is complex, the actual line loss is usually caused by the combined action of a plurality of loss increasing factors, meanwhile, all factors of the line loss are dynamically changed, physical modeling is difficult, and the problems of low analysis efficiency, high misjudgment rate, incomplete finding of influencing factors and the like still exist in manual experience and the existing means.

The overall construction and application of the internet of things still have defects, and the basic architecture mainly has the following defects:

and in the aspect of a terminal: the acquisition and monitoring coverage of terminals such as a power distribution side and a power utilization side is insufficient, and the instantaneity is not strong; the terminal intellectualization and bidirectional interaction level are low; lack of uniform planning designs and standards;

and (3) network aspect: the communication access network has insufficient coverage depth and insufficient bandwidth; a bandwidth bottleneck exists in a special channel for a power grid regulation and control service; the solution of the end-to-end ubiquitous communication network is lacked, and the network resource allocation capacity is insufficient;

in the aspect of a platform: the cross-professional unified Internet of things management is lacked, and the access and application management capabilities of the super-large-scale terminal are insufficient; open sharing for the internet of things application is insufficient; the platform layer and the terminal layer have insufficient cooperative computing and real-time response capabilities.

At present, the problems of energy index treatment work mainly include the following aspects:

1. energy index treatment is not predictive, namely the current energy index management system is mainly based on monitoring and data storage, a large amount of data does not play a role, and the energy index cannot be predicted in real time, so that the abnormal condition of the energy index cannot be accurately judged;

2. the currently used energy index management system mostly takes data calculation as a main part, has weak analysis capability, cannot accurately position the real-time abnormal state of each station room, equipment, line, platform area and user, has high treatment difficulty of energy indexes, and has the risks of misjudgment, missing judgment and the like because the diagnosis of the abnormal energy index cause depends heavily on the professional level of personnel;

3. the management decision depends on manual work, the current energy index management is in a semi-automatic level, the management depends on manual experience and manual decision seriously, and effective technology and data support are lacked.

4. The data value is not utilized sufficiently, most of analysis work of energy index management at present depends on each large service system for linkage, and the data are not mined sufficiently in the aspect of energy index governance due to the reasons of large data quantity, large time span, complex data association relation and the like.

Disclosure of Invention

The embodiment of the invention provides an AI-based intelligent diagnosis method, system and device, the core of which is AI + IoT, and the method is composed of standardized and industrialized IoT edge computing hardware equipment and a customized AI module, and comprises the following steps: collecting power utilization data by a collecting unit of the IoT equipment; the cloud platform trains an AI model according to historical data, the AI model is dynamically improved according to new data, and the edge computing unit of the IoT equipment inputs collected power utilization data into the trained artificial intelligence AI model to obtain related data. The real-time prediction of treatment is realized, the real-time monitoring level of abnormal and standard exceeding of energy indexes is practically improved, and the working mode of detection and manual verification of the former semi-manual system is changed into a full-automatic intelligent AI analysis mode.

According to a first aspect of embodiments of the present invention, an AI-based intelligent diagnosis method includes:

the IoT acquisition unit acquires power consumption data;

and inputting the collected power utilization data into the trained artificial intelligence AI model to obtain related data.

The artificial intelligence AI model comprises an electricity stealing AI model and detects electricity stealing data, wherein the electricity stealing AI model is obtained by training according to the following procedures: collecting power utilization data with labels; preprocessing the collected electricity utilization data; and training the electricity stealing AI model through the preprocessed electricity utilization data.

Further comprising:

collecting new power consumption data with labels at fixed time intervals, and preprocessing the newly collected power consumption data;

and updating the electricity stealing AI model by utilizing the preprocessed newly collected electricity utilization treatment.

The electricity stealing AI model is a classification regression tree model or a progressive gradient classification tree model.

And for the electricity consumption data which are more than N data/day, a classification regression tree model is adopted, for the electricity consumption data which are less than or equal to N data/day, a progressive gradient classification tree model is selected, and N is an integer which is more than 1.

Training the electricity stealing AI model through the preprocessed electricity utilization data comprises the following steps:

selecting partial power utilization data from the preprocessed power utilization data to train the model;

selecting the rest part of power utilization data from the preprocessed power utilization data to test the model;

and when the test index of the test result is within the threshold range, finishing training, and modifying the model parameters to retrain if the test result is not within the threshold range.

Pre-processing the collected electricity usage data, comprising:

and backfilling the lost power utilization data by adopting a point estimation method based on Gaussian distribution and a polynomial regression method aiming at the collected user data, and correcting the power utilization data with the error larger than a preset threshold value.

Inputting collected power utilization data into a trained area table AI model, and detecting abnormal fluctuation data of the energy index of the area, wherein the area table AI model is obtained by training according to the following procedures: collecting power utilization data with labels; preprocessing the collected electricity utilization data; and training the AI model of the transformer area table through the preprocessed power utilization data.

Inputting the collected power consumption data into a trained technical loss AI model, and predicting real-time line loss data, wherein the technical loss AI model is obtained by training according to the following procedures: collecting power utilization data with labels; preprocessing the collected electricity utilization data; and training a technical loss AI model through the preprocessed power utilization data.

Inputting collected power consumption data into a trained user variable relationship AI model, calculating the influence of the user variable relationship on energy indexes, and screening out inaccurate user variable relationships, wherein the user variable relationship AI model is obtained by training according to the following procedures: collecting power utilization data with labels; preprocessing the collected electricity utilization data; and training the user-variable relationship AI model through the preprocessed power consumption data.

Inputting the collected power utilization data into a trained collection AI model, and screening out inaccurate data to be collected, wherein the collection AI model is obtained by training according to the following procedures: collecting power utilization data with labels; preprocessing the collected electricity utilization data; and training the acquired AI model through the preprocessed power utilization data.

Inputting collected electricity utilization data into a trained metering AI model, calculating the influence of metering on energy indexes, and screening out metering results with inaccurate metering, wherein the metering AI model is obtained by training according to the following procedures: collecting power utilization data with labels; preprocessing the collected electricity utilization data; and training the measurement AI model through the preprocessed power utilization data.

Inputting the collected power utilization data into a trained archive AI model, analyzing the influence of archive errors on energy indexes, and screening out problematic archives, wherein the archive AI model is obtained by training according to the following procedures: collecting power utilization data with labels; preprocessing the collected electricity utilization data; and training the archive AI model through the preprocessed power utilization data.

An AI-based intelligent IOT edge computing energy indicator diagnostic system comprising at least one processor, and at least one memory, wherein the memory stores a computer program that, when executed by the processor, causes the processor to perform the steps of the method.

The utility model provides an intelligent IOT edge calculation energy index diagnostic device based on AI, includes edge calculation module and a plurality of data acquisition module, and the edge calculation module is connected to the data acquisition module, and the data acquisition module is used for gathering the power consumption data, gives the edge calculation module with data transmission.

Still include power module, data acquisition module is including gathering submodule piece, transmission submodule piece, wherein:

the acquisition submodule is used for acquiring data, the transmission module is connected to the acquisition submodule and used for transmitting the data acquired by the acquisition module to the edge calculation module, and the power supply module is respectively connected with power supply ends of the acquisition submodule and the transmission submodule and used for supplying power to the acquisition submodule and the transmission submodule.

The edge calculation module comprises a processing submodule, a data receiving submodule and a storage submodule, wherein the data receiving submodule is used for receiving the data acquired by the acquisition submodule; the storage submodule is connected with the processing submodule and used for storing the data acquired by the acquisition submodule; the power supply module is used for supplying power to the processing submodule, the data receiving submodule and the storage submodule.

The data acquisition module also comprises a single chip microcomputer system and an interface submodule, wherein the single chip microcomputer system is connected with the output end of the interface submodule.

The interface sub-module comprises an RS485 interface and an RS232 interface.

The data acquisition module further comprises a temperature sensor and/or a watchdog system.

The transmission submodule comprises a LORA module and an antenna.

The data receiving submodule of the edge computing module comprises a LORA NC unit and a data receiving antenna.

The processing submodule of the edge computing module comprises a multi-core Central Processing Unit (CPU) and a memory unit.

The data sending module comprises a 4G unit and/or a 5G unit and further comprises a data sending antenna used for transmitting data to the control center.

The technical scheme provided by the embodiment of the invention has the following beneficial effects:

the intelligent diagnosis method, system and device based on AI adopt hierarchical deployment, and the edge node of the user end completes data acquisition and return. The edge nodes in the distribution substation area have the capabilities of cloud Application Programming Interface (API), edge computing, edge storage and the like, can bear intelligent Advanced Information (AI), and realize low-delay service response. When the edge calculation is completed by the edge nodes of the distribution transformer area, the data are transmitted back to the area center layer (national grid/provincial and municipal level data center) to provide edge data with rich types for the data center side, so that the center side can analyze and mine the big data, and the center side can better cooperate with an IoT edge calculation platform.

The AI intelligent system can deeply dig the big data value of related devices and platforms such as an intelligent electric meter, an acquisition terminal and power consumption information, the lean and intelligent level of energy index management is improved, and the energy index is effectively improved. And applying AI artificial intelligence combined with new technologies such as data mining and edge calculation to develop intelligent diagnosis research of energy indexes so as to promote the construction of ubiquitous power Internet of things.

Intelligent AI is a branch of computer science that reacts in a similar way to human intelligence, and research in this field includes robotics, language recognition, image recognition, natural language processing, and expert systems, among others.

The edge calculation processes the collected data nearby on an intelligent gateway at the edge side of the network without uploading a large amount of data to a remote data center, and an application program is initiated at the edge side to generate a faster network service response, thereby meeting the basic requirements of the industry on real-time services, application intelligence, safety, privacy protection and the like.

The intelligent energy index management system has the advantages that various complex factor artificial intelligence modularized treatments of energy index management work are realized, real-time prediction of energy index management is realized, the real-time monitoring level of abnormal overproof energy indexes is practically improved, the working mode of detection and artificial verification of the conventional semi-artificial energy index management system is changed into a full-automatic intelligent AI analysis mode, the real-time discovery rate is increased from 65% to 91% according to the actual feedback of field use, and the problems of more missed judgment and high misjudgment of abnormal fluctuation of the energy indexes are solved.

The intelligent AI system has the characteristics of real-time prediction of a distribution room and a high-voltage area, not only can carry out real-time deep data mining and prediction at a user side, but also can provide accurate data support for modification projects, and a model can record the actual operation state of equipment and reduce the cost caused by transitional modification. Meanwhile, an intelligent AI machine model can be used for collecting and processing huge data which is old and new and cannot be processed by manual and semi-automatic systems, and backfilling and correcting missing parts of the existing data, so that the working requirements of the intelligent AI system are met, and linkage assistance can be provided for other semi-automatic systems of a power grid company. On the basis, a data bridge of the power grid equipment and the user is established, the abnormity of the equipment and the user is accurately judged and predicted, and the treatment rate of the feedback energy index for the system on the implementation site is improved by 30%.

The intelligent AI system is deeply applied to the Internet of things technology, edge calculations such as synchronous electricity selling quantity, station area line loss rate and the like are carried out by using an intelligent distribution transformer terminal and a new-generation intelligent acquisition terminal by using a distribution transformer as a unit, and two types including edge calculation of an equipment end and edge data center of three stations are integrated (a transformer substation is transformed into a transformer substation, an energy storage station and a data center). The method uses artificial intelligence, combines edge calculation, deeply excavates edge data value, and converts the data value into business benefit value.

The intelligent AI system utilizes equipment information and access data of ubiquitous electric power thing networking to carry out intelligent analysis, carries out intelligent identification to the equipment in the thing networking, defines the access authority through authentication, and when logging on, the visit of crossing the security level takes place, the operation such as early warning, interception, record will be judged according to intelligence to the intelligent AI system to the intelligence, realizes intelligent defense, including the dynamic perception to thing networking security situation, the automatic distribution of early warning information, the intelligent analysis of security threat, the linkage of corresponding measure deals with etc..

Drawings

The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the invention and together with the description, serve to explain the principles of the invention.

FIG. 1 is a flow chart of an AI-based intelligent diagnostic method;

FIG. 2 is a graph of the output of a classification regression tree model training;

FIG. 3 is a block diagram of an AI-based intelligent diagnostic system;

FIG. 4 is a schematic diagram of an AI-based intelligent diagnostic device;

fig. 5 is a schematic view of the whole flow of the AI-based intelligent diagnosis device.

Detailed Description

Example one

As shown in fig. 1, the present invention provides an AI-based intelligent diagnosis method, including:

collecting power utilization data;

Specifically, the training of the model is realized by collecting real power utilization data of the platform area through the IOT edge device. The data feature set comprises time, single-phase/three-phase voltage, single-phase/three-phase current, zero line voltage/current, phase included angle, whether the cover is opened or not, user classification data (normal users and electricity stealing users) with labels as actual user classification data serve as actual results of user classification, and the model is trained.

Preprocessing the collected power utilization data, backfilling the lost data by adopting a point estimation method based on Gaussian distribution and a polynomial regression method, and correcting the data with larger errors, so that the quality of training data is improved, and the learning effect of the model is ensured; data set

Wherein

The representative tags respectively represent a normal user and a power stealing user,

the characteristic vectors comprise time, single-phase/three-phase voltage, single-phase/three-phase current, zero line voltage/current, phase included angle and whether to open a cover, wherein the time is the time of data reading moment, whether to open the cover in the cover is 1 and whether to open the cover is 0;

training the AI model, selecting partial data for training, using the rest data as test data, and finishing the training when set conditions are met;

preferably, the AI model may be a classification regression tree model or a progressive gradient classification tree model, and preferably, for data/day greater than N, a classification regression tree model is used, and for data/day less than or equal to N, a progressive gradient classification tree model is used; n is an integer greater than 1, and N may be 10 ten thousand.

As shown in FIG. 2, a classification regression tree model is selected for training, part of data in the classification regression tree model is randomly selected as training data, the model is trained, the model belongs to a decision tree method in the field of classical machine learning, the relation between features and labels can be learned very efficiently under a moderate data set, and the model can be visually displayed in a tree form. The core construction of the model is determined by calculating Gini coefficients of different data sets, wherein the formula of the Gini coefficients is as follows:

where k denotes the data class, p_kWhat is shown is the occupation ratio of the kth class data. After the model is constructed, the model can be visualized into a tree;

testing the model by using the rest data as test data to obtain the accuracy, the recall ratio and the comprehensive evaluation index F1, and finishing the training when the accuracy, the recall ratio and the comprehensive evaluation index F1 are within the threshold range; when the comprehensive evaluation index F1 is not in the threshold range, adjusting the feature selection and the tree structure, and re-training until the precision and the recall rate are reached, wherein the comprehensive evaluation index F1 is in the threshold range;

deploying the trained classification regression tree model in a place with a large IOT edge data volume, such as edge equipment used for data summarization at a station area end;

selecting a progressive gradient classification tree model for training, randomly selecting part of data as training data,

initialization:

n is the number of input data, i is the position of the input data,

the representative labels respectively represent normal users and electricity stealing users, and the loss functions of the actual value and the predicted value of L (#); c is the calculated predicted value;

calculation for M1, 2, … M, i 1,2, … N

To r_m,iFitting the regression tree to obtain the leaf node region R of the mth tree_m,j,j＝1,2…J

For J equal to 1,2 … J, calculate each region R_m,jOutput value of (c):

updating

Obtaining a regression tree:

testing the model by using the data of the rest part to obtain accuracy, recall rate and comprehensive evaluation index F1, finishing training if the accuracy, recall rate and comprehensive evaluation index F1 are in the threshold range, and adjusting the calculated predicted value c if the accuracy, recall rate and comprehensive evaluation index F1 are not in the threshold range;

deploying the trained progressive gradient classification tree model at a place with small data volume and needing finer processing, such as an electric meter end at a user side;

after model training and optimization are completed, the model is deployed to each IOT edge device through an IOT master control platform to run the model. At the moment, the IOT equipment can continuously acquire new data at intervals of a period of time (for example, 15 minutes), wherein the new data comprises indexes such as time, single-phase/three-phase voltage, single-phase/three-phase current, zero line voltage/current, phase included angle, whether the cover is opened and the like, and the acquired data is substituted into the trained model, so that whether the electricity stealing user is obtained through calculation.

Wherein, for the progressive gradient classification tree, the formula is as follows. Where M represents the number of decision trees, h_mRepresents the m-th decision tree, x represents the input feature vector, Θ_mRepresenting model parameters obtained by training. When a power stealing subscriber is detected, the IOT edge device can respond to the subscriber at the first time, for example, notify the nearby utility company to trigger an alarm, and perform power stealing in the shortest time.

And (3) collecting new electricity stealing type electricity utilization data at fixed intervals, and performing a new round of training after the electricity utilization data comprise (characteristic + label data), so that the model adapts to the continuously changing electricity stealing types. The model training process is the same as the previous process and is not described in detail herein. After training is completed, the model is deployed to each IOT device through the IOT master control platform, and therefore updating operation of the model is completed.

Example two

The invention relates to an AI-based intelligent diagnosis method, which comprises the following steps:

a large amount of real power utilization data of the platform area are collected through an IOT edge device and used for training a model. The total amount of training data collected by each IOT device is about 20w, and the training data comprises time, single-phase/three-phase voltage, single-phase/three-phase current, zero line voltage/current, phase included angle, whether the cover is opened and other index data as characteristic sets, and actual user classification data (normal users and electricity stealing users) as actual results of user classification as labels to train the model.

In the 20w data collected by the IOT edge device, the partially missing and erroneous data are included, so we need to pre-process the data to make the data reach the standard of the training data. Through an EDA (empirical Data analysis) method, the Distribution rule of good-quality Data is obvious, the characteristic of Gaussian Distribution is presented, and strong correlation is presented among indexes, so that lost Data are backfilled through the good-quality Data by adopting a point estimation method based on the Gaussian Distribution and a Polynomial Regression (Polynomial Regression) method, and the Data with large errors are corrected, so that the quality of training Data is improved, and the learning effect of a model is ensured.

After data preprocessing is completed, 70% of data are randomly selected as training data, And then a Classification And Regression Tree (Classification Tree) model is adopted for training. The model belongs to a Decision Tree (Decision Tree) method in the field of classical machine learning, can learn the relationship between features and labels very efficiently under a moderate data set, and can visually display the model in a Tree form in a visualized manner. The core construction of the model is determined by calculating Gini coefficients of different data sets, wherein the formula of the Gini coefficients is as follows:

where k denotes the data class, p_kWhat is shown is the occupation ratio of the kth class data. After the model is constructed, the model can be visualized as a tree, as shown in fig. 1.

We tested the model with the remaining 30% of the data as test data to obtain the precision (precision) P of 100%, the recall (recall) R of 91%, and the overall evaluation index (F1-Measure) F1 of (2P R)/(P + R) of 95%. Here, we find that model indexes perform very well, but there is a phenomenon of Overfitting (Overfitting), so we decide to adopt an ensemble learning model in the field of machine learning, namely a progressive Gradient classification Tree (Gradient Boost classification Tree), to continue to improve the model performance. The core idea of model training is to sequentially and iteratively train each learner in sequence and pay attention to samples wrongly divided by the learners in sequence, so that the occurrence of errors is reduced by continuously iterating experience training of the learners, and a good learning effect is achieved.

After model training and optimization are completed, the model is deployed to each IOT edge device through an IOT master control platform to run the model. At the moment, the IOT equipment can continuously acquire new data at intervals of a period of time (for example, 15 minutes), wherein the new data comprises indexes such as time, single-phase/three-phase voltage, single-phase/three-phase current, zero line voltage/current, phase included angle, whether the cover is opened and the like, and then the acquired data is substituted into a formula of a trained progressive gradient classification tree, so that whether the electricity stealing user exists or not can be calculated. The formula is shown below. Where M represents the number of decision trees, h_mRepresents the m-th decision tree, x represents the input feature vector, Θ_mRepresenting model parameters obtained by training. When a power stealing subscriber is detected, the IOT edge device can respond to the subscriber at the first time, for example, notify the nearby utility company to trigger an alarm, and perform power stealing in the shortest time.

Updating the model: over time, new types of electricity stealing must emerge in endless ways. To accommodate this change, we can let the model receive a new round of training after a certain period of time (e.g., 1 day) for a new class of electricity stealing data input (feature + tag data), thereby letting the model adapt to the changing type of electricity stealing. The model training process is the same as the previous process and is not described in detail herein. After training is completed, the model is deployed to each IOT device through the IOT master control platform, and therefore updating operation of the model is completed.

The AI-based intelligent energy index diagnosis system consists of seven AI modules, namely electricity stealing modules, a cell table, technical loss modules, a family change relationship module, an acquisition module, a metering module, a file module and the like, solves the problems of inaccurate data and artificial limitation in the energy index management process through data AI, and establishes a set of fully-automatic AI system for energy index monitoring, analysis, diagnosis, prediction and intelligent decision.

EXAMPLE III

The AI-based intelligent diagnosis method further comprises the following steps:

collecting power utilization data;

Preferably, the artificial intelligence AI model further includes a cell table AI model, a technical loss AI model, a user variable relationship AI model, an acquisition AI model, a metering AI model, and a archive AI model.

Wherein the correlation model is as follows:

cell table AI model

The distribution area meter is used for measuring the difference between the power supply quantity and the electricity selling quantity (called line loss electricity quantity), and the distribution area meter counts the power supply quantity and the electricity selling quantity of the 10kV/400V low-voltage distribution area and provides basic data of the line loss of the distribution area. The AI module of the cell list mainly completes the following tasks:

1. firstly, backfilling and correcting data to improve the accuracy of the data to more than 95%;

2. according to the logic of expert manual judgment, the machine learns over 20 ten thousand of correct and incorrect data, a cell table AI model is established, and abnormal fluctuation data of the cell energy indexes are screened out;

3. putting the electricity stealing AI model into a production environment simulation database, and further correcting the model and increasing parameters to further plump the model;

4. and after discovering the abnormal fluctuation data of the energy index, the AI module of the cell table is linked with other modules to analyze the cause of the abnormal fluctuation of the energy index.

AI model of technical loss

The line loss is mainly divided into two types of technical line loss and management line loss according to the reasons of the line loss, the technical line loss is also called theoretical line loss, the calculation of the technical line loss is used as an important work of line loss management, a theoretical basis is provided for making a loss reduction scheme, a technical measure plan, determining a line loss rate index and line loss examination, and the line loss management guidance function is provided. The technical loss AI module mainly completes the following tasks:

2. according to expert judgment logic, the machine learns more than 20 ten thousand of technical loss data to obtain a technical loss AI model;

3. putting the technical loss AI model into a production environment simulation database, and further correcting the model and increasing parameters to further plump the model;

4. the real-time theoretical line loss rate of each distribution area can be accurately predicted through the distribution area theoretical line loss AI model, and the results of line loss abnormal fluctuation are screened out, so that scientific basis is provided for management of distribution area line loss;

user-variant relationship AI model

The correctness of the transformer area household transformation relation is the basis for effectively carrying out transformer area energy index management, and whether the transformer area household transformation relation corresponds to each other is the problem to be solved firstly in transformer area energy index management. The accurate family change relationship is not only the basis of the accurate statistics of the energy indexes of the transformer area, but also the guarantee of the subsequent accurate analysis and the timely correction, the correct family change relationship is not mastered, and the management of the energy indexes of the transformer area cannot be carried out at all. Therefore, the accuracy of the user variable relationship is very important, and the AI module of the user variable relationship mainly completes the following tasks:

2. according to the expert manual judgment logic, the machine learns more than 20 ten thousand of correct and incorrect data to obtain an AI model;

3. putting the user-variant relation AI model into a production environment simulation database, and further correcting the model and increasing parameters to further plump the model;

4. the user variable relationship AI module can very accurately calculate the influence of the user variable relationship on the energy index and screen out the inaccurate user variable relationship.

Acquisition AI model

The collection is a basic link of energy index management, the normal work of the collection equipment is the basis of accurate data, and the collection AI module mainly completes the following tasks:

2. according to the logic of expert manual judgment, the machine learns more than 20 ten thousand of correct and incorrect data to obtain an acquired AI model;

3. putting the acquired AI model into a production environment simulation database, and further correcting the model and increasing parameters to further plump the model;

4. the acquired AI model can very accurately calculate the influence of acquisition on the energy index and screen out the data with inaccurate acquisition.

Measurement AI model

Measurement management is a key link of energy index management, the accuracy of energy index basic statistical data is directly determined, accurate measurement is not carried out, and subsequent analysis and improvement measures are air attics. The main means for ensuring the metering accuracy is to strengthen the metering management of the metering gauge and to strengthen the monitoring by means of technology, and the metering AI module mainly completes the following tasks:

2. according to the logic of expert manual judgment, the machine learns more than 20 ten thousand of correct and incorrect data to obtain a measurement AI model;

3. putting the measurement AI model into a production environment simulation database, and further correcting the model and increasing parameters to further plump the model;

4. the measurement AI module can accurately calculate the influence of measurement on the energy index, and screens out the measurement result with inaccurate measurement.

Archive AI model

The reason that the marketing system file is wrong and the marketing system file and the file information in the electricity information acquisition system are not synchronous and the like can cause the abnormal statistics of the electricity selling quantity of the acquisition system and cause the abnormal energy index of the distribution room. The archive AI module will mainly address this issue, completing the following tasks:

2. according to the expert manual judgment logic, the machine learns more than 20 ten thousand of correct and incorrect data to obtain an archive AI model;

3. putting the file AI model into a production environment simulation database, and further correcting the model and increasing parameters to further plump the model;

4. the archives AI module can be accurate the analysis play archives mistake to the influence of energy index, can also select the archives that probably have the problem.

As shown in fig. 3, a computing system according to the present invention may include at least one processor and at least one memory. Wherein the memory stores program code which, when executed by the processor, causes the processor to perform the steps of the resource exhibition method according to various exemplary embodiments of the present invention described above in the present specification.

Components of an AI-based intelligent diagnostic system can include, but are not limited to: the at least one processor, the at least one memory, and a bus connecting the various system components (including the memory and the processor).

A bus represents one or more of any of several types of bus structures, including a memory bus or memory controller, a peripheral bus, a processor, or a local bus using any of a variety of bus architectures.

The memory may include readable media in the form of volatile memory, such as Random Access Memory (RAM) and/or cache memory, and may further include Read Only Memory (ROM).

The memory may also include a program/utility having a set (at least one) of program modules including, but not limited to: an operating system, one or more application programs, other program modules, and program data, each of which, or some combination thereof, may comprise an implementation of a network environment.

The system may also communicate with one or more external devices (e.g., keyboard, pointing device, etc.), with one or more devices that enable a user to interact with the computing device, and/or with any device (e.g., router, modem, etc.) that enables the computing device to communicate with one or more other computing devices. Such communication may be through an input/output (I/O) interface. Also, the system may communicate with one or more networks (e.g., a Local Area Network (LAN), a Wide Area Network (WAN), and/or a public network, such as the internet) via a network adapter. As shown, the network adapter communicates with other modules for the system over a bus. It should be understood that although not shown in the figures, other hardware and/or software modules may be used in conjunction with the system, including but not limited to: microcode, device drivers, redundant processors, external disk drive arrays, RAID systems, tape drives, and data backup storage systems, among others. The system comprises a module corresponding to the AI-based intelligent diagnosis method, in particular to

The system comprises an acquisition module and an artificial intelligence AI module, wherein the acquired electricity utilization data is input into the trained artificial intelligence AI module to obtain related data;

the acquisition module acquires real power utilization data of the transformer area through IOT edge equipment; the data with the labels can be used as training data to train an AI module;

the artificial intelligence AI module comprises an electricity stealing AI module for detecting electricity stealing data,

the electricity stealing AI module is obtained by training according to the following procedures: collecting power utilization data with labels; preprocessing the collected electricity utilization data; and training the electricity stealing AI module through the preprocessed electricity utilization data.

The data preprocessing comprises preprocessing the collected power utilization data, backfilling the lost data by a Gaussian distribution point estimation method and a polynomial regression method, and correcting the data with larger errors;

the electric larceny AI module can be classified into a regression tree module or a progressive gradient classification tree module, the electric larceny AI module is trained by training data with labels, and when the electric larceny AI module meets training conditions, the collected electric data can be detected to detect an electric larceny user. The electricity stealing AI module needs to be retrained every fixed time.

Preferably, the acquisition module comprises a single chip microcomputer system and an interface module, and is used for operating a software program and completing various processing of data acquisition and transmission, and the single chip microcomputer system is connected with the output end of the interface module; the interface module comprises an RS485 interface and an RS232 interface, the RS485 interface is used for being in butt joint with the ammeter, and the interface module is provided with an electronic isolation module in consideration of the reliability of bus connection transmission; RS232, because of the standard point-to-point butt joint, does not have the bus conflict problem, does not have the electronic isolation module.

Electricity stealing AI module

In the aspect of management and loss reduction, the working strength of regulation and regulation of illegal electricity utilization and anti-electricity-stealing is emphasized. The following tasks are mainly completed by using the electricity stealing AI module:

2. according to the logic of the expert manual judgment, the machine learns more than 20 ten thousand of correct and incorrect data to obtain an electricity stealing AI model;

4. the electricity stealing AI module can accurately complete conventional judgment (based on expert experience judgment) of electricity stealing, and can obtain screening results of electricity stealing users except the conventional judgment.

Area table AI module

1.2.3 technical loss AI Module

4. the real-time theoretical line loss rate of each distribution area can be accurately predicted through the distribution area theoretical line loss AI model, and the results of line loss abnormal fluctuation are screened out, so that scientific basis is provided for management of distribution area energy indexes;

1.2.4 household variable relation AI module

1.2.5 acquisition AI Module

4. the AI acquisition module can very accurately calculate the influence of acquisition on the energy index and screen out inaccurate data.

1.2.6 metering AI Module

1.2.7 archives AI module

As shown in fig. 4 and 5, an AI-based intelligent diagnosis device includes an edge calculation module, a plurality of data acquisition modules, and each data acquisition module includes an acquisition submodule, a transmission submodule, and a power supply module;

the acquisition submodule acquires data, the transmission submodule is connected with the acquisition submodule, receives the data of the acquisition submodule and transmits the data acquired by the acquisition submodule to the edge calculation module, and the power supply module is connected with power supply terminals of the acquisition submodule and the transmission submodule and supplies power to the acquisition submodule and the transmission submodule;

the edge calculation module comprises a processing submodule, a data receiving submodule, a storage submodule and a power supply module, wherein the data receiving submodule receives data acquired by an acquisition submodule of the data acquisition module; the storage submodule is connected with the processing submodule and stores the data acquired by the acquisition submodule; the power supply module supplies power to the processing submodule, the data receiving submodule and the storage submodule.

The acquisition module is arranged at the user terminal and the distribution transformer area terminal, and comprises a single chip microcomputer system and an interface submodule for operating a software program and finishing various processing of data acquisition and transmission, wherein the single chip microcomputer system is connected with the output end of the interface submodule; the interface sub-module comprises an RS485 interface and an RS232 interface, the RS485 interface is used for being in butt joint with the ammeter, and the interface sub-module is provided with an electronic isolation module in consideration of the reliability of bus connection transmission; RS232, because of the standard point-to-point butt joint, does not have the bus conflict problem, does not have the electronic isolation module.

The data collected by the interface sub-module comprises current, voltage, electric quantity and three-phase included angle.

Preferably, the data acquisition module further comprises a watchdog system, so that the reliability of the equipment is improved, and when the running of the CPU system program is abnormal, the system can be automatically reset, and the system is enabled to be recovered to be normal.

Preferably, the data acquisition module further comprises a temperature sensor, the temperature sensor is used for detecting the temperature of the working environment of the equipment, and the MCU is internally provided with the temperature sensor.

The power supply module mainly solves the power supply problem, and can directly take power from 220V or adopt an adapter power supply mode.

The transmission sub-module comprises an LORA module and an antenna; the LORA module is a low-power wireless transmission module and is used for transmitting the acquired data back to the edge calculation module. LORA is one of low-power wide area network communication technologies, is an ultra-long distance wireless transmission technology based on spread spectrum technology adopted and popularized by Semtech corporation, and is a unique modulation format generated by the Semtech radio frequency part. The LORA module is a wireless data transmission module developed on the basis of a Semtech SX1276/1278 chip, and the chip is small in integration scale and high in efficiency, so that the LORA module has high receiving sensitivity.

The processing module of the edge computing module comprises a multi-core CPU and a memory unit, and optionally, a 1.4GHz 64-bit 4-core ARM Cortex-a53 CPU may be adopted, where the memory has a capacity of at least 1G in consideration of relatively large local side operation requirement.

The multi-core CPU processing module system preferably uses a linux system to support multithreading.

The CPU processing module comprises a data preprocessing module, a classification regression tree module, a progressive gradient classification tree module, a data processing module and an updating module.

The data volume of the multi-core CPU is more than 3 GB/month (according to 10000 point locations in each region, data is collected every 15 minutes, and single data is estimated to be 100 bytes), and the size of the memory card can be flexibly configured by supporting the TF card, so that the requirements of different application scenes are met.

The data receiving submodule of the edge computing module comprises a LORA NC unit and a data receiving antenna. The LORA NC unit is a data aggregation end of the LORA

The storage submodule of the edge calculation module is a storage card and is used for storing processed data.

The power supply module of the edge computing module mainly solves the problem of module power supply, and directly gets power from 220V, or adopts an adapter power supply mode, and the battery power supply is required to be supported in consideration of temporary power failure or power failure.

The edge calculation module also comprises a data sending submodule, the data sending module comprises a 4G unit or a 5G unit and a data sending antenna, and the data sending submodule sends data to the control center.

The above description is only a preferred embodiment of the application and is illustrative of the principles of the technology employed. It will be appreciated by those skilled in the art that the scope of the invention herein disclosed is not limited to the particular combination of features described above, but also encompasses other arrangements formed by any combination of the above features or their equivalents without departing from the spirit of the invention. For example, the above features may be replaced with (but not limited to) features having similar functions disclosed in the present application.

Claims

1. An AI-based intelligent diagnosis method, comprising:

collecting power utilization data;

inputting the collected power utilization data into a trained artificial intelligence AI model to obtain related data;

the artificial intelligence AI model comprises an electricity stealing AI model and detects electricity stealing data, wherein the electricity stealing AI model is obtained by training according to the following procedures: collecting power utilization data with labels; preprocessing the collected electricity utilization data; training an electricity stealing AI model through the preprocessed electricity utilization data; wherein the content of the first and second substances,

the preprocessing the collected electricity consumption data comprises:

backfilling lost data by adopting a point estimation method based on Gaussian distribution and a polynomial regression method, and correcting data with larger errors and a data set

Wherein

the characteristic vectors comprise time, single-phase/three-phase voltage, single-phase/three-phase current, zero line voltage/current, phase included angle and whether the cover is opened or not;

the training of the power stealing AI model comprises the following steps:

for more than N data/day, the AI model is trained by adopting a classification regression tree model, part of randomly selected data is used as training data for training, wherein N is an integer more than 1, the core construction of the classification regression tree model is determined by calculating Gini coefficients of different data sets, and the formula of the Gini coefficients is as follows:

where k represents the data class,

representing the occupation ratio of the kth class data;

testing the model by using the rest data as test data to obtain the accuracy, the recall ratio, the comprehensive evaluation index, the accuracy and the recall ratio, and ending the training when the comprehensive evaluation index is within the threshold range; when the accuracy, the recall rate and the comprehensive evaluation index are not in the threshold range, adjusting the feature selection and the tree structure, and re-training until the accuracy, the recall rate and the comprehensive evaluation index are in the threshold range;

deploying the trained classification regression tree model on edge equipment used for data summarization at a platform area end;

for data/day less than or equal to N, the AI model is trained by adopting a progressive gradient classification tree model, part of data is randomly selected as training data, wherein N is an integer greater than 1

Initialization:

in order to input the number of data,

in order to input the location of the data,

a loss function of the actual value and the predicted value;

is the calculated predicted value;

for the

，

Computing

To pair

Fitting a regression tree to obtain

Leaf node region of a tree

,

To pair

Calculating each region

Output value of (c):

updating

Obtaining a regression tree:

testing the model by using the data of the rest part to obtain the accuracy, the recall rate and the comprehensive evaluation index, finishing the training if the accuracy, the recall rate and the comprehensive evaluation index are in the threshold range, and adjusting the calculated predicted value if the accuracy, the recall rate and the comprehensive evaluation index are not in the threshold range;

and deploying the trained progressive gradient classification tree model at the electric meter end of the user side.

2. The method of claim 1, wherein the collected power consumption data is input into a trained area table AI model, and abnormal fluctuation data of the area energy index is detected, wherein the area table AI model is obtained by training according to the following procedures: collecting power utilization data with labels; preprocessing the collected electricity utilization data; and training the AI model of the transformer area table through the preprocessed power utilization data.

3. The method of claim 1, wherein the collected power consumption data is input into a trained technical loss AI model to predict real-time line loss data, wherein the technical loss AI model is trained according to the following process: collecting power utilization data with labels; preprocessing the collected electricity utilization data; and training a technical loss AI model through the preprocessed power utilization data.

4. The method according to claim 1, wherein the collected electricity data is input into a trained occupant-dependent relationship AI model, the influence of the occupant-dependent relationship on the energy index is calculated, and inaccurate occupant-dependent relationships are screened out, wherein the occupant-dependent relationship AI model is trained according to the following process: collecting power utilization data with labels; preprocessing the collected electricity utilization data; and training the user-variable relationship AI model through the preprocessed power consumption data.

5. The method of claim 1, wherein the collected power consumption data is input into a trained collected AI model, and inaccurate data is screened out, wherein the collected AI model is trained according to the following process: collecting power utilization data with labels; preprocessing the collected electricity utilization data; and training the acquired AI model through the preprocessed power utilization data.

6. The method according to claim 1, characterized in that the collected electricity data is input into a trained metering AI model, the influence of metering on the energy index is calculated, and metering results with inaccurate metering are screened out, wherein the metering AI model is obtained by training according to the following procedures: collecting power utilization data with labels; preprocessing the collected electricity utilization data; and training the measurement AI model through the preprocessed power utilization data.

7. The method of claim 1, wherein the collected electricity consumption data is input into a trained archive AI model, and the influence of archive errors on the energy index is analyzed to screen out problematic archives, wherein the archive AI model is obtained by training according to the following procedures: collecting power utilization data with labels; preprocessing the collected electricity utilization data; and training the archive AI model through the preprocessed power utilization data.

8. An AI-based intelligent diagnostic system, comprising at least one processor and at least one memory, wherein the memory stores a computer program that, when executed by the processor, causes the processor to perform the method of any of claims 1-7.

9. An AI-based intelligent diagnosis device, comprising an edge calculation module and a plurality of data acquisition modules, wherein the data acquisition modules are connected with the edge calculation module, and are used for acquiring power consumption data and transmitting the acquired data to the edge calculation module, and the AI-based intelligent diagnosis device is used for realizing the method of any one of claims 1 to 7.

10. The apparatus of claim 9, further comprising a power module, the data acquisition module comprising an acquisition sub-module, a transmission sub-module, wherein:

the acquisition submodule is used for acquiring data, the transmission submodule is connected to the acquisition submodule and used for transmitting the data acquired by the acquisition module to the edge calculation module, and the power supply module is respectively connected with power supply ends of the acquisition submodule and the transmission submodule and used for supplying power to the acquisition submodule and the transmission submodule.

11. The apparatus of claim 10, wherein the edge calculation module comprises a processing sub-module, a data receiving sub-module, and a storage sub-module, the data receiving sub-module is configured to receive data collected by the collection sub-module; the storage submodule is connected with the processing submodule and used for storing the data acquired by the acquisition submodule; the power supply module is used for supplying power to the processing submodule, the data receiving submodule and the storage submodule.

12. The device of claim 11, wherein the data acquisition module further comprises a single chip microcomputer system and an interface sub-module, and the single chip microcomputer system is connected with an output end of the interface sub-module.

13. The apparatus of claim 12, wherein the interface submodule comprises an RS485 interface and an RS232 interface.

14. The apparatus of claim 11, wherein the data acquisition module further comprises a temperature sensor and/or a watchdog system.

15. The apparatus of claim 11, wherein the transmission sub-module comprises a LORA module and an antenna.

16. The apparatus of claim 11, wherein the data reception sub-module of the edge calculation module includes a LORA NC unit and a data reception antenna.

17. The apparatus of claim 11, wherein the processing sub-modules of the edge computation module comprise a multi-core Central Processing Unit (CPU) and a memory unit.

18. The apparatus of claim 11, wherein the data transmission module comprises a 4G unit and/or a 5G unit, and further comprises a data transmission antenna for transmitting data to the control center.