CN102765643B - Elevator fault diagnosis and early-warning method based on data drive - Google Patents

Abstract

The invention relates to the field of elevators. In order to realize the early detection and diagnosis of elevator faults, the technical solution adopted by the present invention is to implement a data-driven elevator fault diagnosis and early warning method by means of a remote service center, a fault diagnosis and prediction terminal, and an elevator controller, including the following steps: Firstly, the real-time elevator fault data is mined to obtain the characteristic information in the elevator fault data stream, and the mining results are stored in the elevator fault case database of the fault diagnosis and prediction terminal; Update the elevator fault knowledge base; then conduct case retrieval according to the characteristics of the new elevator fault problem, and use the fault diagnosis method based on case reasoning to diagnose the fault of the elevator system: by retrieving the knowledge or cases of the elevator fault knowledge base, obtain the new elevator fault problem The information with the most similar features solves the problem of diagnosis; the invention is mainly applied to the design and manufacture of image sensors.

Description

Data-driven elevator fault diagnosis and early warning method

technical field

The invention relates to the field of elevators, in particular to a data-driven elevator fault diagnosis and early warning method.

Background technique

Due to the rapid increase in the number of elevators with safety hazards in China in recent years, the maintenance of elevators only through the experience of maintenance personnel or maintenance manuals has problems such as low efficiency, poor accuracy, and often after-the-fact diagnosis, which cannot meet the needs of elevator safety. The elevator needs an intelligent fault diagnosis and early warning system to ensure the safe operation of the system.

There are two main ways to solve the elevator safety problem in China: one is fault diagnosis after the fault occurs, and the other is regular maintenance by maintenance personnel. At present, the widely used fault diagnosis techniques mainly include expert system, fuzzy reasoning, neural network and so on. However, these technologies rely heavily on expert knowledge, and the difficulty in obtaining expert knowledge has become a bottleneck in the implementation of fault diagnosis. In addition, most fault diagnosis methods cannot provide fault prediction function, and passive diagnosis cannot prevent the occurrence of faults, and can only rely on the regular maintenance of elevators. Regular maintenance with unclear purpose is not only costly and inefficient, but also it is difficult to find potential safety hazards of elevators by relying on manual inspection.

Contents of the invention

The purpose of the present invention is to overcome the deficiencies of the prior art and realize the early detection and diagnosis of elevator faults. The implementation of fault diagnosis and prediction terminal and elevator controller includes the following steps:

Firstly, the real-time elevator fault data is mined to obtain the characteristic information in the elevator fault data stream, and the mining results are stored in the elevator fault case database of the fault diagnosis and prediction terminal as the source of the elevator fault knowledge base; then the elevator fault case database is used Update the elevator fault knowledge base on the fault diagnosis and prediction terminal, realize the timely update of the elevator fault knowledge base through similarity matching calculation, and then carry out case retrieval according to the characteristics of new elevator fault problems, and adopt the fault diagnosis method based on case reasoning Fault diagnosis for the elevator system: by retrieving the elevator fault knowledge base knowledge or cases, obtain the information with the most similar characteristics to the new elevator fault problem, and solve the diagnosis problem;

In addition, using the elevator fault identification classifier on the remote service center, the obtained elevator fault data flow is clustered and analyzed, the corresponding elevator fault data flow is associated with the elevator fault type, and the elevator fault data flow is used to correlate with the corresponding fault type Train the classifier, and then test the classifier through another set of elevator fault data streams and corresponding fault types to verify the correctness of the trained classifier; the remote service center continuously updates the classifier and downloads the latest classifier In the local fault diagnosis and prediction terminal, the local fault diagnosis and prediction terminal collects the elevator data stream in real time and inputs it into the classifier, and the real-time data stream output by the classifier is compared with the existing elevator fault data stream for similarity. The larger the value, the greater the possibility of the same type of fault, so the elevator fault prediction is carried out accordingly.

The fault diagnosis of the elevator system using the fault diagnosis method based on case reasoning is carried out on the fault diagnosis and prediction terminal, and is further refined into the following steps:

(1) Elevator fault knowledge base: It is a collection of elevator fault diagnosis knowledge and experience, mainly provided by experts, including elevator basic information, elevator fault classification information, and various key feature attributes and weights required by different types of faults, and Construct elevator fault case database and symptom database accordingly;

(2) Establish an elevator failure case library: maintenance personnel fill in various information about elevator failures according to historical data including elevator failure logs and maintenance logs, and use this as a basis to store cases and generate new cases;

(3) Establish a symptom database: store the fault type data flow information collected when the elevator breaks down, that is, the various parameters of the elevator running when the fault occurs;

(4) Establish a rule base: store the interrelated information between various types of elevator faults, apply the association rule algorithm to the fault case base, carry out data mining, and extract deep-level, hidden information from numerous elevator fault case information. The knowledge contained in it is used for elevator fault diagnosis and guides maintenance personnel to take corresponding maintenance measures;

(5) Reasoning system: composed of case retrieval, case matching, and case adjustment, specifically: search for one or more cases that are most similar to the current fault through case retrieval on the elevator fault case database, and the retrieval algorithm used includes template inspection , inductive retrieval, and nearest neighbor search; then generate solutions based on the retrieved cases and adjust the generated solutions through case correction. The adjustment methods include conversion method, replacement method, and specific goal-driven method;

(6) Case study: According to the feedback information of the maintenance personnel, the case reuse of the elevator fault case library is carried out, that is, if the solution can solve the encountered fault, the maintenance suggestion in the elevator fault case library is saved, otherwise the program is modified Finally, it is stored in the fault case library, so that new knowledge is continuously acquired and old knowledge is improved, and a new maintenance plan is formed, which is added to the case library, so that the case library is continuously expanded and perfected.

The specific implementation steps of case retrieval:

(1) Collect the elevator fault data stream, extract the characteristic information, and preliminarily retrieve the case types that meet the characteristic information according to the classification structure index.

(2) According to the type of the fault case, the fault information feature value is matched with the elevator fault knowledge base.

(3) Calculate according to the improved European algorithm, calculate the matching degree between the target case and all cases in the initial matching case set, sort according to the matching degree, output the first few cases that best match the target case, and complete The case matching process; finally, the case matching details are displayed and prepared for case revision.

The generation process of the classifier includes a data preprocessing module, a feature extraction module, and a classifier generation module. The data preprocessing module adopts steps including standardization and variance reduction, and is responsible for eliminating noise data such as abnormal data and redundant data in the data; feature extraction The module adopts principal component analysis and partial least square method, which is responsible for simplifying data flow and improving training efficiency; the classifier generation module also includes neural network and support vector machine sub-modules.

Technical characteristics and effects of the present invention:

Data mining is the identification of valid, novel, potentially useful, and ultimately understandable types from data. The key and primary problem of fault diagnosis is fault identification. The process of fault diagnosis is also the process of fault type acquisition and fault identification. Considering the unique advantages of data mining technology in knowledge acquisition, it is feasible to introduce this technology in the field of fault diagnosis. It can use the historical data to dig out the potential laws and provide decision-making basis for fault diagnosis, which has practical reference value.

The fault diagnosis and early warning system based on data mining has the following advantages:

(1) Break through the bottleneck of difficulty in obtaining elevator diagnostic knowledge and the lack of knowledge. The diagnostic experience can be obtained automatically without manual summarization and input, which greatly improves the diagnostic efficiency and accuracy, and reduces the diagnostic cost.

(2) For large-scale faults involving multiple variables, the diagnosis method for a single component cannot be solved, and the overall analysis of the elevator operation data by using data mining technology can be effectively diagnosed.

(3) It can not only find the cause and location of the fault for maintenance personnel, but also provide corresponding fault solving measures.

(4) The elevator operation data can be monitored in real time, and the similarity between the real-time data flow and the fault type can be obtained through the classifier, so as to realize early detection and early warning of elevator faults.

(5) The system has the ability of self-learning, and continuously learns new fault data to form new diagnostic knowledge. With the continuous increase of fault data, the fault diagnosis ability of the system will continue to increase.

(6) Provide the basis for preventive maintenance. The preventive maintenance based on fault prediction reduces blindness and enables the elevator to be effectively repaired at the best fault repair time, which not only reduces the number of repairs and costs, but also greatly improves the efficiency of maintenance.

Description of drawings

Figure 1 The overall structure of the data-driven elevator fault diagnosis and early warning system.

Fig. 2 Framework diagram of elevator fault diagnosis based on case-based reasoning.

Figure 3 Schematic representation of case information.

Figure 4. The flow chart of elevator failure case retrieval strategy.

Figure 5 is a flow chart of classifier-based elevator fault prediction.

Fig. 6 is a schematic diagram of the classifier generation of the elevator remote service center.

Figure 7 is a schematic diagram of classifier design based on BP neural network.

Detailed ways

The purpose of the present invention is to propose a data-driven elevator fault diagnosis and early warning system to realize efficient fault diagnosis and accurate fault prediction.

Existing elevator fault diagnosis technology has problems such as difficulty in obtaining expert knowledge, low diagnosis efficiency, and high cost. In view of these problems, the present invention applies the fault diagnosis method based on case reasoning to diagnose the fault of the elevator system. Retrieve the best matching case, and maintain it according to the fault cause, fault location and fault solution method in the case information. At the same time, the fault case library can automatically maintain the case library, including adding cases, merging cases, deleting redundant cases, etc., so that Have a strong learning ability.

At present, most fault diagnosis systems lack the function of fault prediction. The present invention analyzes the historical data of elevators with the help of data mining technology, and summarizes the data flow corresponding to specific faults. Considering expert knowledge and data flow comprehensively, the data flow when the elevator is running Real-time comparison with the known fault data stream, and quantitative calculation of the similarity between the two, when the similarity reaches a certain level, a fault warning can be given to the elevator system, thereby completing the fault prediction function. The system obtains the general parameters of elevator operation and automatically analyzes the operation data, which breaks through the bottleneck of the expert system, and has the advantages of high diagnostic efficiency, low cost and the ability to realize fault prediction. The invention does not need additional sensors, is applicable to elevators of various brands, has good application prospects and economic value, and the system and method also have high reference value for fault diagnosis in other fields.

The present invention uses the elevator fault data mining technology to design an elevator fault diagnosis system, which continuously analyzes the data collected from the elevator system, and automatically and efficiently forms the fault diagnosis knowledge of the elevator system relying on the knowledge acquisition ability of the data mining technology , solve the problem of difficulty in obtaining expert knowledge, and overcome the current technical bottleneck of elevator fault diagnosis. Then, on the basis of this system architecture, a case-based reasoning elevator fault diagnosis method is proposed, and the knowledge formed by the above method is used for diagnosis.

In addition, on the basis of this system architecture, a classifier-based fault prediction function is added, which can monitor the elevator data flow in real time, and use the classifier to analyze and identify these data flows, and calculate the similarity between the current data flow and the fault data flow And trends, and then realize the early detection and diagnosis of elevator faults.

The invention designs a data-driven elevator fault diagnosis and early warning system by means of data mining technology, which has enhanced fault diagnosis and prediction functions.

The present invention first uses data mining algorithm to mine real-time elevator fault data to obtain characteristic information in the elevator fault data flow, and saves the mining result in the elevator fault case database as the source of the elevator fault knowledge base. Then use the elevator fault case database to update the elevator fault knowledge base, and realize the timely update of the elevator fault knowledge base through the similarity matching calculation. Then carry out case retrieval according to the characteristics of the new problem, and obtain the information with the most similar characteristics to the new elevator fault problem by retrieving the knowledge or cases of the elevator fault knowledge base, which is used to solve the diagnosis problem.

In addition, it is designed to be used in elevator fault identification classifiers, cluster analysis is performed on a set of data streams obtained, and the corresponding data streams are associated with fault types, and the classifier is trained with this data stream and corresponding fault types, and then through another The group data flow and the corresponding fault type are tested against the classifier to verify the correctness of the trained classifier. The remote service center continuously updates the classifier and downloads the latest classifier to the local diagnostic terminal. The local diagnostic terminal collects the elevator data stream in real time and inputs the data stream into the classifier, and the classifier outputs the real-time data stream and the existing fault data stream The similarity degree, the greater the similarity, the greater the possibility of the same type of failure, and the elevator failure prediction can be performed in turn.

The present invention will be described in further detail below in conjunction with the accompanying drawings.

The core of the present invention is to analyze the elevator operation data flow, dig out the characteristic information of the data flow during the fault, find the data flow corresponding to the elevator fault type, and convert it into expert experience, store it in the elevator fault diagnosis case library, Then use the case-based reasoning fault diagnosis method to diagnose the fault of the elevator system. In addition, a fault data flow classifier is designed, which can classify the current data flow of the elevator in real time, and calculate the similarity between the current data flow and the elevator fault data flow through a distance-based similarity algorithm. The greater the similarity, the greater the possibility of failure. Large, according to the trend of this similarity or by setting the threshold value method for fault prediction.

Referring to Figure 1, the data-driven elevator fault diagnosis and early warning system includes three parts: a remote service center, a fault diagnosis and prediction terminal, and an elevator controller.

When a fault occurs, the elevator controller records the fault code in the system and various parameters in the current elevator system, such as: traction machine speed, car acceleration, inverter voltage, inverter current, leveling signal, topping signal, bottoming signal, door operator signal, etc. And pass the fault code and current parameters to the local diagnosis platform. During normal operation, it only needs to send the current system parameters to the local diagnostic terminal in real time for fault prediction.

The elevator fault diagnosis and prediction software and SQLServer2005 database software are installed in the local diagnosis terminal. After receiving the elevator fault code and current parameters, the characteristic value of the fault type will be extracted, and then the best matching case will be found in the case library according to this. This fault information and the matching fault cause and solution are sent to the remote maintenance personnel through the Internet or mobile terminals such as mobile phones; Identify it as a new fault type, and send the current fault information to the remote service center. On the other hand, the elevator fault diagnosis and prediction software integrates the classifier module programmed by COM components to complete the elevator fault prediction function.

The remote service center is responsible for collecting the data streams of all elevator system failures, and using the data streams in these fault states to train the classifier and case library stored in the service center, and constantly updating the classifier and case library to make fault type identification and diagnosis more accurate ; The remote service center regularly downloads the latest classifier and case library to the local diagnosis terminal, and responds to the elevator failure warning information.

Referring to Figure 2, the elevator fault diagnosis framework based on case reasoning mainly includes four databases, a reasoning system and a case study module. Each part is described in detail as follows:

(1) Knowledge base: a collection of elevator fault diagnosis knowledge and experience, which is mainly provided by experts, including basic information of elevators, classification information of elevator faults, and various key feature attributes and their weights required by different types of faults, and based on this Construct elevator fault case library and symptom database.

(2) Fault case library: Maintenance personnel fill in various information about elevator faults according to historical data such as elevator fault logs and maintenance logs, and use this as a basis to store cases and generate new cases.

(3) Symptom database: the fault type data flow information collected when the elevator fails, that is, the various parameters of the elevator running when the fault occurs.

(4) Rule base: Interrelated information between various types of elevator faults. It is a maintenance measure that applies association rule algorithm to the fault case database, conducts data mining, and extracts deep and implicit knowledge from numerous elevator fault case information, which is used for elevator fault diagnosis and guides maintenance personnel to respond.

(5) Reasoning system: the core of the diagnosis system, composed of case retrieval, case matching, and case adjustment. Find one or more cases that are most similar to the current fault through case retrieval on the elevator fault case database. The retrieval algorithms that may be used include template inspection, inductive retrieval, and nearest neighbor search. Then generate a solution based on the retrieved case and adjust the generated solution through case revision. There may be conversion method, replacement method, and specific goal-driven method. Most of the case adjustments are completed through human-computer interaction. of. The reasoning system determines the diagnostic efficiency, realizes finding the case most similar to the current problem from the existing case set, and provides corresponding fault solutions.

(6) Case study: According to the feedback information of the maintenance personnel, the case reuse of the elevator fault case library is carried out, that is, if the solution can solve the encountered fault, the maintenance suggestion in the elevator fault case library is saved, otherwise the program is modified Then save it to the fault case library. In this way, new knowledge is continuously acquired and old knowledge is improved, new maintenance schemes are formed, and added to the case base, so that the case base is continuously expanded and improved.

See Figure 3. Each case in the case library is composed of basic case information, fault cause and location, and fault solution. During the diagnosis process, the elevator controller provides basic information such as fault data flow, elevator signal, etc., and the diagnosis system is based on these information. Analyze and return information such as fault cause, location, and fault solution.

See Figure 4. Case retrieval is the key to the entire elevator fault diagnosis process based on case reasoning. The following are the specific implementation steps:

(1) Collect the fault data stream of the faulty elevator, extract the characteristic information, and preliminarily retrieve the case types that meet the characteristic information according to the classification structure index.

(2) Match the fault information feature value with the case set according to the types of fault cases.

(3) Calculate according to the improved European algorithm, calculate the matching degree between the target case and all cases in the initial matching case set, sort according to the matching degree, output the first few cases that best match the target case, and complete Case matching process. Finally, the case match details are displayed and prepared for case revision.

According to each case in the elevator fault case library, the attribute function matrix is constructed as follows:

where A _ij represents the jth attribute of the ith case. Record the average value of the jth attribute as B _j , then:

${\mathrm{<span\; class="notranslate">\; B</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; =</span>\frac{\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; m</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{\mathrm{<span\; class="notranslate">\; a</span>}}_{\mathrm{<span\; class="notranslate">\; ij</span>}}}{\mathrm{<span\; class="notranslate">\; m</span>}}$

Note the intermediate variable M _ij

${\mathrm{<span\; class="notranslate">\; m</span>}}_{\mathrm{<span\; class="notranslate">\; ij</span>}}<span\; class="notranslate">\; =</span>\frac{{\mathrm{<span\; class="notranslate">\; a</span>}}_{\mathrm{<span\; class="notranslate">\; ij</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; B</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}}{{\mathrm{<span\; class="notranslate">\; B</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}}$

Reorder:

${{\mathrm{<span\; class="notranslate">\; a</span>}}_{\mathrm{<span\; class="notranslate">\; ij</span>}}}^{<span\; class="notranslate">\; \&prime;</span>}<span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; e</span>}}^{{<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; m</span>}}_{\mathrm{<span\; class="notranslate">\; ij</span>}}}}{\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; e</span>}}^{<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; m</span>}}_{\mathrm{<span\; class="notranslate">\; ij</span>}}}}$

The improved Euclidean distance between the target case and the source case is:

${\mathrm{<span\; class="notranslate">\; d</span>}}_{\mathrm{<span\; class="notranslate">\; ti</span>}}<span\; class="notranslate">\; =</span>{<span\; class="notranslate">\; \{</span>\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; no</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{\mathrm{<span\; class="notranslate">\; w</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}{<span\; class="notranslate">\; (</span>{{\mathrm{<span\; class="notranslate">\; a</span>}}_{\mathrm{<span\; class="notranslate">\; ij</span>}}}^{<span\; class="notranslate">\; \&prime;</span>}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; a</span>}}_{\mathrm{<span\; class="notranslate">\; ij</span>}}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; \}</span>}^{\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; 2</span>}}}$

In the formula, w _i is the weight value given by expert experience. The larger the value of d _ti is, the smaller the distance between the target case and the source case is, and the higher the similarity is, the source case with the smallest distance is calculated during the retrieval process for diagnosis.

The improved European algorithm introduces the intermediate variable M _ij on the basis of the traditional European algorithm, which adds a process of attribute value normalization, which can effectively prevent the value of some attributes in the same case from being too large, causing the retrieval results to deviate from the actual situation. Appear.

See Figure 5. Through the classifier’s identification of the elevator fault data flow and the calculation of the similarity, the development trend of the fault similarity or the comparison result with the threshold value can be obtained. In order to ensure the accuracy of the other, the classifier has a remote service The center performs regular updates, and finally completes the fault prediction function of the elevator.

Referring to Fig. 6, the generation of the classifier includes two stages (training stage and testing stage). The fault data stream for classifier generation is split into two parts, where two-thirds of the data is used for the training phase and one-third of the data is used for the testing phase. In the training phase, a classifier is initially generated, and in the testing phase, the generated classifier is verified to ensure its accuracy.

The generation process of the classifier includes a data preprocessing module, a feature extraction module, and a classifier generation module. The data preprocessing module is responsible for removing noise data such as abnormal data and redundant data in the data, and the feature extraction module is responsible for simplifying the data flow and improving training. efficiency. The classifier generation module is essentially a module with nonlinear function simulation function such as support vector machine and neural network. The data preprocessing module needs to use statistical and mathematical tools including standardization, variance reduction, etc., and the feature processing module may use mathematical methods such as principal component analysis and partial least squares.

Referring to Figure 7, the BP neural network is used in the classifier generation module, the number of nodes in the input layer is defined as 2, the number of nodes in the output layer is 1, the number of nodes in the hidden layer is 6, and the Logsig type transfer function is used, expressed as follows:

$\mathrm{<span\; class="notranslate">\; Logsig</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; exp</span>}<span\; class="notranslate">\; (</span><span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; )</span>}$

The weight modification formula is:

W _sq (t+1)＝W _sq (t)+η(t)δ _q y _s +αΔW _sq (t)

Among them, η is the gain item, δ _q is the error item, y _s is the output of node s, and α is the set weight. W _sq (t) is the weight of the tth iteration. The classifier is implemented in Visual C++ environment. The input sample set and the corresponding training target set are directly stored in the SQL database to ensure the versatility of the data. Design the classifier module as a COM component, and call the component when needed.

Claims (3)

1. A data-driven elevator fault diagnosis and early warning method is characterized in that, by means of remote service center, fault diagnosis and prediction terminal and elevator controller realizes, comprises the steps: Firstly, the real-time elevator fault data is mined to obtain the characteristic information in the elevator fault data stream, and the mining results are stored in the elevator fault case database of the fault diagnosis and prediction terminal as the source of the elevator fault knowledge base; then the elevator fault case database is used Update the elevator fault knowledge base on the fault diagnosis and prediction terminal, realize the timely update of the elevator fault knowledge base through similarity matching calculation, and then carry out case retrieval according to the characteristics of new elevator fault problems, and adopt the fault diagnosis method based on case reasoning Carry out fault diagnosis on the elevator system: by retrieving the knowledge or cases in the elevator fault knowledge base or temporary elevator fault case database, obtain the information with the most similar characteristics to the new elevator fault problem, and solve the diagnosis problem; In addition, using the elevator fault identification classifier on the remote service center, the obtained elevator fault data flow is clustered and analyzed, the corresponding elevator fault data flow is associated with the elevator fault type, and the elevator fault data flow is used to correlate with the corresponding fault type Train the classifier, and then test the classifier through another set of elevator fault data streams and corresponding fault types to verify the correctness of the trained classifier; the remote service center continuously updates the classifier and downloads the latest classifier In the local fault diagnosis and prediction terminal, the local fault diagnosis and prediction terminal collects the elevator data stream in real time and inputs it into the classifier, and the real-time data stream output by the classifier is compared with the existing elevator fault data stream for similarity. The greater the probability of the same type of failure, the elevator failure prediction is based on this; The specific implementation steps of case retrieval: (1) Collect the elevator fault data stream, extract the characteristic information and index according to the classification structure, and initially retrieve the case types that meet the characteristic information; (2) match the fault information characteristic value with the elevator fault knowledge base according to the type of the fault case; (3) Calculate according to the improved European algorithm, calculate the matching degree of the target case and all the cases in the initial matching case set, sort according to the size of the matching degree, output the first few cases that best match the target case, and complete the case Matching process; finally, case matching details are displayed and prepared for case revision; where: According to each case in the elevator fault case library, the attribute function matrix is constructed as follows: Where A _ij represents the jth attribute of the i-th case, and the average value of the jth attribute is B _j , then: ${\mathrm{<span\; class="notranslate">\; B</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; =</span>\frac{\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; m</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{\mathrm{<span\; class="notranslate">\; a</span>}}_{\mathrm{<span\; class="notranslate">\; ij</span>}}}{\mathrm{<span\; class="notranslate">\; m</span>}}$ Note the intermediate variable M _ij ${\mathrm{<span\; class="notranslate">\; m</span>}}_{\mathrm{<span\; class="notranslate">\; ij</span>}}<span\; class="notranslate">\; =</span>\frac{{\mathrm{<span\; class="notranslate">\; a</span>}}_{\mathrm{<span\; class="notranslate">\; ij</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; B</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}}{{\mathrm{<span\; class="notranslate">\; B</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}}$ Reorder: ${{\mathrm{<span\; class="notranslate">\; a</span>}}_{\mathrm{<span\; class="notranslate">\; ij</span>}}}^{<span\; class="notranslate">\; \&prime;</span>}<span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; e</span>}}^{{<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; m</span>}}_{\mathrm{<span\; class="notranslate">\; ij</span>}}}}{\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; e</span>}}^{{<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; m</span>}}_{\mathrm{<span\; class="notranslate">\; ij</span>}}}}$ The improved Euclidean distance between the target case and the source case is: ${\mathrm{<span\; class="notranslate">\; d</span>}}_{\mathrm{<span\; class="notranslate">\; ti</span>}}<span\; class="notranslate">\; =</span>{<span\; class="notranslate">\; \{</span>\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; no</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{\mathrm{<span\; class="notranslate">\; w</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}{<span\; class="notranslate">\; (</span>{{\mathrm{<span\; class="notranslate">\; a</span>}}_{\mathrm{<span\; class="notranslate">\; ij</span>}}}^{<span\; class="notranslate">\; \&prime;</span>}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; a</span>}}_{\mathrm{<span\; class="notranslate">\; ij</span>}}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; \}</span>}^{\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; 2</span>}}}$ The intermediate variable M _ij is introduced on the basis of the traditional European algorithm, that is, a process of normalizing attribute values is added. 2. The elevator fault diagnosis and early warning method based on data drive as claimed in claim 1 is characterized in that, adopting the fault diagnosis method based on case reasoning to carry out fault diagnosis to elevator system is to carry out on the fault diagnosis and prediction terminal, and further It is refined into the following steps: (1) Elevator fault knowledge base: It is a collection of elevator fault diagnosis knowledge and experience, mainly provided by experts, including elevator basic information, elevator fault classification information, and various key feature attributes and weights required by different types of faults, and Construct elevator fault case database and symptom database accordingly; (2) Establish an elevator failure case library: maintenance personnel fill in various information about elevator failures according to historical data including elevator failure logs and maintenance logs, and use this as a basis to store cases and generate new cases; (3) Establish a symptom database: store the fault type data flow information collected when the elevator breaks down, that is, the various parameters of the elevator running when the fault occurs; (4) Establish a rule base: store the interrelated information between various types of elevator faults, apply the association rule algorithm to the fault case base, carry out data mining, and extract deep-level, hidden information from numerous elevator fault case information. The knowledge contained in it is used for elevator fault diagnosis and guides maintenance personnel to take corresponding maintenance measures; (5) Reasoning system: composed of case retrieval, case matching, and case adjustment, specifically: search for one or more cases that are most similar to the current fault through case retrieval on the elevator fault case database, and the retrieval algorithm used includes template inspection , inductive retrieval, and nearest neighbor search; then generate solutions based on the retrieved cases and adjust the generated solutions through case correction. The adjustment methods include conversion method, replacement method, and specific goal-driven method; (6) Case study: According to the feedback information of the maintenance personnel, the case reuse of the elevator fault case library is carried out, that is, if the solution can solve the encountered fault, the maintenance suggestion in the elevator fault case library is saved, otherwise the program is modified Finally, it is saved to the fault case library, so that new knowledge can be continuously obtained and old knowledge improved, and a new maintenance plan can be formed and added to the case library, so that the case library can be continuously expanded and improved. 3. the elevator fault diagnosis and early warning method based on data-driven as claimed in claim 1, is characterized in that, the generation process of classifier comprises data preprocessing module, feature extraction module and classifier generation module, wherein data preprocessing module adopts Including standardization and variance reduction steps, it is responsible for eliminating noise data such as abnormal data and redundant data in the data; the feature extraction module adopts principal component analysis and partial least square method, which is responsible for simplifying the data flow and improving training efficiency; the classifier generation module also includes Neural network, support vector machine sub-modules.