Elevator risk early warning method and device

By building a fine-grained elevator risk warning model, using fault and risk prediction sub-models for feature extraction and predictive value adjustment, the challenges in elevator fault prediction and safety management are solved, and prediction accuracy and safety are significantly improved.

CN119026722APending Publication Date: 2024-11-26HANGZHOU SPECIAL EQUIP INSPECTION & RES INST
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410868447.4
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-01
Publication Date
2024-11-26

AI Technical Summary

Technical Problem

The existing technology is difficult to effectively predict and early warning of elevator failures, resulting in frequent safety accidents, and there are challenges in designing effective elevator failure prediction and maintenance solutions.

Method used

Using an elevator risk warning method, by obtaining sample data, including maintenance cycle, risk label and fault sub-data, the fault prediction sub-model and risk prediction sub-model are used for feature extraction and predicted value adjustment, combining fine-grained feature integration and information Shadow and build a new risk warning model to improve prediction accuracy.

Benefits of technology

It significantly improves the accuracy of elevator fault prediction and the accuracy of risk warning, can output specific fault prediction results in fine-grainedness, avoid the noise caused by the difference in data distribution obtained at different times, and improves the reliability of elevator safety management.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119026722A_ABST
    Figure CN119026722A_ABST
Patent Text Reader

Abstract

The invention relates to an elevator risk early warning method and device.The elevator risk early warning method comprises the steps that sample data are obtained, the sample data comprise sample sub-data corresponding to multiple maintenance periods of a sample elevator, risk labels of the sample elevator and fault sub-data corresponding to the sample sub-data, and the sample sub-data comprise sample basic sub-data and a weather data sequence; inputting each sample sub-data into a fault prediction sub-model in the risk early warning model to obtain a fault prediction value; the fault prediction sub-model inputs the feature information corresponding to each sample sub-data into a risk prediction sub-model in a risk early warning model to obtain a risk prediction value; adjusting parameters in the fault prediction sub-model according to the difference between each fault prediction value and the corresponding fault sub-data; adjusting the parameters of the risk prediction sub-model according to the difference between the risk prediction value and the risk label until a parameter adjustment stop condition is reached; and elevator risk early warning is carried out based on the risk early warning model obtained after parameter adjustment. The embodiment of the invention improves the prediction accuracy.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of elevators, and in particular to an elevator risk early warning method and device. Background Art

[0002] With the acceleration of urbanization and economic development, the number of elevators, as an indispensable vertical transportation tool in modern buildings, is increasing rapidly. However, this is also accompanied by frequent elevator failures, safety accidents and other problems, posing a threat to people's lives and property. Due to the complexity of elevators and the influence of various factors, it is indeed a big challenge to design an effective elevator failure prediction and maintenance plan. Summary of the invention

[0003] The invention provides an elevator risk early warning method and device.

[0004] The present invention is implemented by the following technical solution: an elevator risk early warning method, comprising: Acquire sample data, the sample data including sample sub-data corresponding to multiple maintenance cycles of a sample elevator, risk labels of the sample elevator, and fault sub-data corresponding to the sample sub-data, each of the sample sub-data including sample basic sub-data and a weather data sequence; Input each of the sample sub-data into a fault prediction sub-model in a risk warning model to obtain a fault prediction value, wherein the fault prediction sub-model is used to perform the following operations: extract features from the sample sub-data to obtain corresponding feature information; and obtain the fault prediction value based on the feature information; The fault prediction sub-model inputs the feature information corresponding to each of the sample sub-data into the risk prediction sub-model in the risk warning model to obtain a risk prediction value, and the risk prediction sub-model is used to perform the following operations: horizontally splicing each of the feature information to obtain a sample comprehensive feature; by masking the sample comprehensive feature, obtaining a sample effective comprehensive feature corresponding to each of the sample sub-data, and the sample effective comprehensive feature and the sample comprehensive feature have the same number of channels; extracting features from each of the sample effective comprehensive features and each fault sub-data to obtain a feature set corresponding to the sample data; and obtaining the risk prediction value according to the feature set; Adjusting the parameters in the fault prediction sub-model according to the difference between each fault prediction value and the corresponding fault sub-data; and adjusting the parameters of the risk prediction sub-model according to the difference between the risk prediction value and the risk label until the parameter adjustment stop condition is reached; Elevator risk warning is performed based on the risk warning model obtained after parameter adjustment.

[0005] Furthermore, the maintenance cycle refers to a single maximum time interval between the time when the sample elevator is in a normal operating state and the time before the last failure occurs.

[0006] Furthermore, the sample basic sub-data has timestamp information, and the feature extraction of the sample sub-data to obtain corresponding feature information includes: Clustering each sample basic sub-data in the sample sub-data based on the timestamp information to obtain a preset number of clusters; Channel expansion is performed according to the product of the number of clusters and the data channel of the sample basic sub-data to obtain the channel space; Determine a valid channel for each cluster in the channel space, deploy sample basic sub-data of the cluster in the valid channel, and obtain sample channel sub-data corresponding to the sample basic sub-data; Use the feature extractor corresponding to each cluster to extract features from all sample channel sub-data in the cluster to obtain cluster features corresponding to the cluster; Fusion of the cluster features to obtain a first feature; Performing time series-based feature extraction on the weather data sequence of the sample sub-data to obtain a second feature; The first feature and the second feature are fused to obtain feature information corresponding to the sample sub-data.

[0007] Further, the fault prediction sub-model includes a feature network and a first prediction network, the feature network is used to obtain feature information, the first prediction network is used to obtain the fault prediction value based on the feature information, the feature network includes each of the feature extractors, a first fusion device and a second fusion device, the first fusion device is used to fuse cluster features, and the second fusion device is used to fuse the first feature and the second feature; The adjusting the parameters in the fault prediction sub-model according to the difference between each fault prediction value and the corresponding fault sub-data includes: adjusting the parameters of the feature network and the first prediction network according to the difference between each fault prediction value and the corresponding fault sub-data.

[0008] Furthermore, the sample effective comprehensive feature corresponding to each of the sample sub-data is obtained by performing information masking on the sample comprehensive feature, including: for each of the sample sub-data, a copy of the sample comprehensive feature is copied, and the feature channels in the sample comprehensive feature that do not belong to the sample sub-data are set to blank to obtain the corresponding sample effective comprehensive feature.

[0009] Furthermore, the risk prediction sub-model includes: a multi-layer perceptron, a third fuser, a convolutional network, a fourth fuser and a second prediction network corresponding to each sample sub-data, each of the multi-layer perceptrons is connected to the third fuser, the convolutional network and the third fuser are connected to the fourth fuser, and the fourth fuser is connected to the second prediction network; The feature extraction is performed on each effective comprehensive feature of the sample and each fault sub-data to obtain a feature set corresponding to the sample data, including: Based on the multi-layer perceptron corresponding to each sample sub-data, feature extraction is performed on the corresponding sample effective comprehensive feature to obtain the corresponding third feature; fusing the third features based on the third fuser to obtain a fourth feature; Extracting features of each fault sub-data based on the convolutional network to obtain a fifth feature; The feature set is obtained by fusing the fourth feature and the fifth feature based on the fourth fuser.

[0010] Further, obtaining the risk prediction value according to the feature set includes: The second prediction network outputs the risk prediction value according to the feature set; The adjusting the parameters of the risk prediction sub-model according to the difference between the risk prediction value and the risk label comprises: According to the difference between the risk prediction value and the risk label, the parameters of the multilayer perceptron, the third fuser, the convolutional network, the fourth fuser and the second prediction network are adjusted.

[0011] The present application also discloses an elevator risk warning device, the elevator risk warning device comprising: A training module that performs the following operations: Acquire sample data, the sample data including sample sub-data corresponding to multiple maintenance cycles of a sample elevator, risk labels of the sample elevator, and fault sub-data corresponding to the sample sub-data, each of the sample sub-data including sample basic sub-data and a weather data sequence; Input each of the sample sub-data into a fault prediction sub-model in a risk warning model to obtain a fault prediction value, wherein the fault prediction sub-model is used to perform the following operations: extract features from the sample sub-data to obtain corresponding feature information; and obtain the fault prediction value based on the feature information; The fault prediction sub-model inputs the feature information corresponding to each of the sample sub-data into the risk prediction sub-model in the risk warning model to obtain a risk prediction value, and the risk prediction sub-model is used to perform the following operations: horizontally splicing each of the feature information to obtain a sample comprehensive feature; by masking the sample comprehensive feature, obtaining a sample effective comprehensive feature corresponding to each of the sample sub-data, and the sample effective comprehensive feature and the sample comprehensive feature have the same number of channels; extracting features from each of the sample effective comprehensive features and each fault sub-data to obtain a feature set corresponding to the sample data; and obtaining the risk prediction value according to the feature set; Adjusting the parameters in the fault prediction sub-model according to the difference between each fault prediction value and the corresponding fault sub-data; and adjusting the parameters of the risk prediction sub-model according to the difference between the risk prediction value and the risk label until the parameter adjustment stop condition is reached; The application module is used to provide elevator risk warning based on the risk warning model obtained after parameter adjustment.

[0012] The present application also discloses a computer device, which includes a processor and a memory, wherein the memory stores at least one instruction, at least one program, a code set or an instruction set, and the at least one instruction, the at least one program, the code set or the instruction set are loaded and executed by the processor to implement the aforementioned elevator risk warning method.

[0013] The present application also discloses a computer-readable storage medium, in which at least one instruction, at least one program, a code set or an instruction set is stored. The at least one instruction, the at least one program, the code set or the instruction set is loaded and executed by a processor to implement the aforementioned elevator risk warning method.

[0014] The embodiment of the present application designs a new network structure of a risk warning model, which can realize fine-grained feature extraction and feature integration by period and time clustering, and can avoid the noise caused by the distribution differences between data acquired at different times. Combined with the fault sub-data corresponding to the maintenance cycle, fine-grained risk prediction error correction is realized, thereby significantly improving the accuracy of risk prediction, and specific fault prediction results can also be output in a fine-grained manner. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 1 is a flow chart of an elevator risk early warning method according to an embodiment of the present invention; Figure 2 is a flow chart of a feature information extraction method according to an embodiment of the present invention; Figure 3 is a flow chart of a feature set acquisition method according to an embodiment of the present invention; Figure 4 It is a block diagram of an elevator risk warning device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0016] In order to make the objectives, technical solutions and advantages of the present invention more clear, the present invention will be further described in detail below with reference to the accompanying drawings.

[0017] The present application provides an elevator risk early warning method.

[0018] In one embodiment, Figure 1 As shown, the elevator risk warning method comprises the following steps: S1. Obtain sample data, wherein the sample data includes sample sub-data corresponding to multiple maintenance cycles of a sample elevator, risk labels of the sample elevator, and fault sub-data corresponding to the sample sub-data. Each of the sample sub-data includes sample basic sub-data and a weather data sequence.

[0019] In the embodiment of the present application, data related to the sample elevator can be continuously collected for multiple times and in different time periods for the sample elevator, so as to obtain the sample data required for training, and each sample elevator corresponds to a set of sample data. The sample data includes sample sub-data corresponding to multiple maintenance cycles, risk labels of the sample elevators, and fault sub-data corresponding to the sample sub-data.

[0020] In the embodiment of the present application, the above-mentioned maintenance cycle refers to the single maximum time interval between the moment when the above-mentioned sample elevator is in normal operation and the moment before the most recent failure. For example, the elevator fails on January 1 and is restored to normal through maintenance, and fails again on January 10, then January 2 to January 9 is a maintenance cycle. The failure that occurred on January 10 is the fault sub-data corresponding to the maintenance cycle from January 2 to January 9, that is, the fault sub-data is the fault content of the fault that caused the corresponding maintenance cycle to end. The sample sub-data is the data generated within the maintenance cycle. This application does not limit the fault. Elevator faults can be divided into elevator entrapment, open door operation, continuous opening and closing of elevator doors, elevator overspeed, etc.

[0021] The risk label is a label set by professional elevator operation and maintenance personnel based on the operating performance of the sample elevator and related fault sub-data during multiple maintenance cycles. The label indicates the risk level of the sample elevator in the future. If the risk level is high, the elevator may fail soon or have a serious failure. If the risk level is low, the elevator may run smoothly for a long time.

[0022] Each of the above sample sub-data includes sample basic sub-data and weather data sequence. The weather data sequence refers to the weather time series changes within the maintenance period corresponding to the sample sub-data. Sample basic sub-data refers to the general term for data useful for elevator fault prediction and risk prediction. The specific content of the data is not limited in the embodiments of this application and does not constitute an implementation obstacle. For example, the following content may be included: the type of place where the elevator is located, the content of elevator-related events that occurred in the place, the type of elevator, usage status, number of resident status, elevator operation status, usage, maintenance information, etc.

[0023] S2, input each of the sample sub-data into the fault prediction sub-model in the risk warning model to obtain a fault prediction value, wherein the fault prediction sub-model is used to perform the following operations: extract features from the sample sub-data to obtain corresponding feature information. Based on the feature information, the fault prediction value is obtained.

[0024] The data acquired in different time periods will lead to differences in the information distribution of the data in these time periods. In actual use, it is found that this phenomenon will reduce the prediction accuracy, that is, noise related to the distribution difference is introduced when performing indifferent data processing. In order to improve the prediction accuracy, the embodiment of the present application eliminates this type of noise, thereby improving the prediction accuracy.

[0025] Each sample sub-data includes multiple sample basic sub-data, that is, the sample basic sub-data can be obtained multiple times during the maintenance cycle, so as to fully extract relevant data and improve the prediction accuracy. The timestamp information is configured for the sample basic sub-data each time it is extracted. Please refer to Figure 2 , which shows a flow chart of a feature information extraction method. The feature extraction of the sample sub-data is performed to obtain corresponding feature information, including: S21. Clustering each sample basic sub-data in the above sample sub-data based on the timestamp information to obtain a preset number of clusters.

[0026] The fault prediction sub-model includes a feature network and a first prediction network, wherein the feature network is used to obtain feature information, and the first prediction network is used to obtain the fault prediction value based on the feature information, and the feature network includes feature extractors. Each feature extractor corresponds to a cluster, that is, the feature network includes the preset number of feature extractors.

[0027] If the sample sub-data includes 10 sample basic sub-data, and the preset number is 3, the clustering result can obtain 3 clusters, and cluster 1 includes sample basic sub-data 1, sample basic sub-data 4, and sample basic sub-data 9. Cluster 2 includes sample basic sub-data 2, sample basic sub-data 3, and sample basic sub-data 5. Cluster 3 includes sample basic sub-data 6, sample basic sub-data 7, sample basic sub-data 8, and sample basic sub-data 10.

[0028] S22. Perform channel expansion according to the product of the number of clusters and the data channel of the sample basic sub-data to obtain the channel space.

[0029] The number of channels of each sample basic sub-data is the same. Assuming that they all have 4 channels, the channel space is channel 1 to channel 12, with a total of 12 channels.

[0030] S23. Determine the effective channel of each cluster in the channel space, deploy the sample basic sub-data of the cluster in the effective channel, and obtain the sample channel sub-data corresponding to the sample basic sub-data.

[0031] For example, the valid channels of cluster 1 are channels 1 to 4, the valid channels of cluster 2 are channels 5 to 8, and so on. Then the four channels of sample basic sub-data 1 are arranged in channels 1 to 4 respectively, and the other channels are set to zero, so as to obtain the sample channel sub-data corresponding to the above sample basic sub-data.

[0032] S24. Use the feature extractor corresponding to each cluster to perform feature extraction on all sample channel sub-data in the above cluster to obtain cluster features corresponding to the above cluster.

[0033] S25. Fuse the above cluster features to obtain the first feature.

[0034] S26. Perform time-series-based feature extraction on the weather data sequence of the sample sub-data to obtain a second feature.

[0035] S27. The first feature and the second feature are combined to obtain feature information corresponding to the sample sub-data.

[0036] The feature network further includes a first fuser and a second fuser, wherein the first fuser is used to fuse cluster features, and the second fuser is used to fuse the first feature and the second feature.

[0037] The embodiments of the present application specifically state that the relevant logical components used in the operations such as feature extraction and feature fusion in the embodiments of the present application, such as feature extractors and fusers, can refer to the relevant achievements of neural networks or artificial intelligence, and do not constitute an implementation obstacle to the embodiments of the present application. In other words, the specific processing methods for data processing in the present application, in addition to the contents disclosed in the present application, the undisclosed contents can use the existing technology and do not constitute an implementation obstacle to the embodiments of the present application.

[0038] The invention of this application mainly lies in the improvement of the overall network structure, that is, how to build an overall network architecture of a risk warning model with specific functions based on the detailed network layer, and improve the accuracy of the prediction results through the design of fine-grained data processing and network training process, rather than other detailed contents of the invention, such as the specific form of a specific network layer, which can use existing technologies and do not constitute an obstacle to the implementation of the embodiments of this application.

[0039] S3. The above-mentioned fault prediction sub-model inputs the characteristic information corresponding to each of the above-mentioned sample sub-data into the risk prediction sub-model in the above-mentioned risk warning model to obtain the risk prediction value. The above-mentioned risk prediction sub-model is used to perform the following operations: horizontally splice each of the above-mentioned characteristic information to obtain the sample comprehensive feature. By masking the above-mentioned sample comprehensive feature, the sample effective comprehensive feature corresponding to each of the above-mentioned sample sub-data is obtained, and the above-mentioned sample effective comprehensive feature has the same number of channels as the above-mentioned sample comprehensive feature. Feature extraction is performed on each of the above-mentioned sample effective comprehensive features and each fault sub-data to obtain the feature set corresponding to the above-mentioned sample data. The above-mentioned risk prediction value is obtained according to the above-mentioned feature set.

[0040] Horizontal splicing is channel expansion horizontal splicing, which is an existing data splicing method, and the embodiments of this application will not be described in detail. The above-mentioned effective comprehensive features of the samples corresponding to each of the above-mentioned sample sub-data are obtained by information masking of the above-mentioned sample comprehensive features, including: for each of the above-mentioned sample sub-data, a copy of the sample comprehensive features is copied, and the feature channels in the above-mentioned sample comprehensive features that do not belong to the sample sub-data are emptied to obtain the corresponding effective comprehensive features of the samples.

[0041] The above-mentioned risk prediction sub-model includes: a multilayer perceptron, a third fuser, a convolutional network, a fourth fuser and a second prediction network corresponding to each sample sub-data, each of the above-mentioned multilayer perceptrons is connected to the third fuser, the above-mentioned convolutional network and the above-mentioned third fuser are connected to the above-mentioned fourth fuser, and the above-mentioned fourth fuser is connected to the above-mentioned second prediction network.

[0042] Please refer to Figure 3 , which shows a flow chart of the feature set acquisition method. The above-mentioned effective comprehensive features of each sample and each fault sub-data are extracted to obtain the feature set corresponding to the sample data, including: S31. Based on the multi-layer perceptron corresponding to each sample sub-data, feature extraction is performed on the corresponding sample effective comprehensive feature to obtain the corresponding third feature.

[0043] S32. Based on the third fuser, the third features are fused to obtain a fourth feature.

[0044] S33. Based on the above convolutional network, feature extraction is performed on each fault sub-data to obtain the fifth feature.

[0045] S34. Based on the fourth fuser, the fourth feature and the fifth feature are fused to obtain the feature set.

[0046] S4, adjusting the parameters in the fault prediction sub-model according to the difference between each fault prediction value and the corresponding fault sub-data. Also, adjusting the parameters in the risk prediction sub-model according to the difference between the risk prediction value and the risk label, until the parameter adjustment stop condition is reached.

[0047] The adjusting of the parameters in the fault prediction sub-model according to the difference between each fault prediction value and the corresponding fault sub-data includes: adjusting the parameters of the feature network and the first prediction network according to the difference between each fault prediction value and the corresponding fault sub-data.

[0048] The step of obtaining the risk prediction value according to the feature set includes: the second prediction network outputs the risk prediction value according to the feature set. The step of adjusting the parameters of the risk prediction sub-model according to the difference between the risk prediction value and the risk label includes: adjusting the parameters of the multi-layer perceptron, the third fuser, the convolutional network, the fourth fuser and the second prediction network according to the difference between the risk prediction value and the risk label.

[0049] The embodiments of the present application do not set specific data forms for various prediction values, fault sub-data, and risk labels, as long as the differences can be quantified. The difference quantification method is also not limited, for example, a common loss function for neural network training can be used. Parameter adjustment and parameter adjustment stop conditions are also not elaborated, and existing technologies can be used.

[0050] S5. Perform elevator risk warning based on the risk warning model obtained after parameter adjustment.

[0051] The fault prediction submodel trained in the embodiment of the present application can predict future faults based on the basic data and weather sequence of the elevator over a period of time, while the risk prediction submodel can predict future risk levels based on the basic data, weather sequence, and related fault data of the elevator in multiple maintenance cycles. The fault prediction submodel and the risk prediction submodel can be used separately or in combination, that is, the components of each submodel obtained by training can be intelligently combined to achieve the purpose of risk warning. Of course, the data content in the basic data and the sample basic subdata is based on the same inventive concept, the data content in the weather sequence and the weather data sequence is based on the same inventive concept, and the data content in the fault data and the fault subdata is based on the same inventive concept.

[0052] For example, if the basic data and weather sequence of the target elevator over a period of time are obtained, the fault prediction sub-model can be used directly to predict its faults. The characteristic information calculated by the fault prediction model can also be input into the risk prediction sub-model for risk prediction. At this time, if there is no information in the dimension of fault data, it can be left blank, and the risk prediction sub-model can still output the risk prediction results.

[0053] For another example, the sub-data corresponding to the multiple maintenance cycles of the target elevator, the fault data corresponding to the above sub-data, and the sub-data corresponding to the latest maintenance cycle can be directly obtained. The latest maintenance cycle has not ended yet, so there is no corresponding fault data. Each of the above sub-data includes basic sub-data and weather data sequence. The sub-data corresponding to the multiple maintenance cycles and the sub-data corresponding to the latest maintenance cycle are directly input into the fault prediction sub-model to obtain the feature information corresponding to each sub-data, and the fault prediction result corresponding to the latest maintenance cycle. Each feature information, the fault prediction result, and each fault data are input into the risk prediction sub-model to obtain the risk prediction result.

[0054] Another embodiment of the present application provides an elevator risk warning device, such as Figure 4 As shown, the elevator risk warning device includes: The training module 401 is used to perform the following operations: Sample data is obtained, wherein the sample data includes sample sub-data corresponding to multiple maintenance cycles of the sample elevator, risk labels of the sample elevator, and fault sub-data corresponding to the sample sub-data, and each of the sample sub-data includes sample basic sub-data and a weather data sequence.

[0055] Each of the sample sub-data is input into the fault prediction sub-model in the risk warning model to obtain a fault prediction value, and the fault prediction sub-model is used to perform the following operations: extract features from the sample sub-data to obtain corresponding feature information. The fault prediction value is obtained based on the feature information.

[0056] The above-mentioned fault prediction sub-model inputs the characteristic information corresponding to each of the above-mentioned sample sub-data into the risk prediction sub-model in the above-mentioned risk warning model to obtain the risk prediction value. The above-mentioned risk prediction sub-model is used to perform the following operations: horizontally splicing each of the above-mentioned characteristic information to obtain the sample comprehensive feature. By masking the above-mentioned sample comprehensive feature, the sample effective comprehensive feature corresponding to each of the above-mentioned sample sub-data is obtained, and the above-mentioned sample effective comprehensive feature has the same number of channels as the above-mentioned sample comprehensive feature. Feature extraction is performed on each of the above-mentioned sample effective comprehensive features and each fault sub-data to obtain the feature set corresponding to the above-mentioned sample data. The above-mentioned risk prediction value is obtained according to the above-mentioned feature set.

[0057] According to the difference between each fault prediction value and the corresponding fault sub-data, the parameters in the fault prediction sub-model are adjusted. And, according to the difference between the risk prediction value and the risk label, the parameters of the risk prediction sub-model are adjusted until the parameter adjustment stop condition is reached.

[0058] Application module 402 is used to perform elevator risk warning based on the risk warning model obtained after parameter adjustment.

[0059] Furthermore, the maintenance cycle refers to a single maximum time interval between the time when the sample elevator is in a normal operating state and the time before the last failure.

[0060] Furthermore, the sample basic sub-data has timestamp information, and the training module 401 is used to perform the following operations: Each sample basic sub-data in the above sample sub-data is clustered based on the timestamp information to obtain a preset number of clusters.

[0061] Channel expansion is performed according to the product of the number of clusters and the data channel of the sample basic sub-data to obtain the channel space.

[0062] Determine the effective channel of each cluster in the channel space, deploy the sample basic sub-data of the cluster in the effective channel, and obtain the sample channel sub-data corresponding to the sample basic sub-data.

[0063] The feature extractor corresponding to each cluster is used to extract features from all sample channel sub-data in the above cluster to obtain cluster features corresponding to the above cluster.

[0064] The above cluster features are combined to obtain the first feature.

[0065] The weather data sequence of the sample sub-data is subjected to time series-based feature extraction to obtain a second feature.

[0066] The first feature and the second feature are combined to obtain feature information corresponding to the sample sub-data.

[0067] Furthermore, the above-mentioned fault prediction sub-model includes a feature network and a first prediction network, the above-mentioned feature network is used to obtain feature information, the above-mentioned first prediction network is used to obtain the above-mentioned fault prediction value based on the above-mentioned feature information, the above-mentioned feature network includes each of the above-mentioned feature extractors, the first fuser and the second fuser, the above-mentioned first fuser is used to fuse cluster features, and the above-mentioned second fuser is used to fuse the above-mentioned first features and the above-mentioned second features.

[0068] The training module 401 is used to perform the following operations: adjusting the parameters of the feature network and the first prediction network according to the difference between each fault prediction value and the corresponding fault sub-data.

[0069] Furthermore, the training module 401 is used to perform the following operations: for each of the sample sub-data, copy a sample comprehensive feature, and empty the feature channels in the sample comprehensive feature that do not belong to the sample sub-data, to obtain the corresponding sample effective comprehensive feature.

[0070] Furthermore, the above-mentioned risk prediction sub-model includes: a multilayer perceptron, a third fuser, a convolutional network, a fourth fuser and a second prediction network corresponding to each sample sub-data, each of the above-mentioned multilayer perceptrons is connected to the third fuser, the above-mentioned convolutional network and the above-mentioned third fuser are connected to the above-mentioned fourth fuser, and the above-mentioned fourth fuser is connected to the above-mentioned second prediction network.

[0071] The training module 401 is used to perform the following operations: Based on the multi-layer perceptron corresponding to each sample sub-data, feature extraction is performed on the corresponding sample effective comprehensive feature to obtain the corresponding third feature.

[0072] The fourth feature is obtained by fusing the above third features based on the above third fuser.

[0073] Based on the above convolutional network, feature extraction is performed on each fault sub-data to obtain the fifth feature.

[0074] The fourth feature and the fifth feature are fused based on the fourth fuser to obtain the feature set.

[0075] Furthermore, the training module 401 is used to perform the following operations: The second prediction network outputs the risk prediction value based on the feature set.

[0076] The training module 401 is used to perform the following operations: According to the difference between the risk prediction value and the risk label, the parameters of the multilayer perceptron, the third fuser, the convolutional network, the fourth fuser and the second prediction network are adjusted.

[0077] The present application also provides a computer device, which includes a processor and a memory, wherein the memory stores at least one instruction, at least one program, a code set or an instruction set, and the at least one instruction, the at least one program, the code set or the instruction set are loaded and executed by the processor to implement the aforementioned elevator risk warning method.

[0078] The present application also provides a computer-readable storage medium, in which at least one instruction, at least one program, a code set or an instruction set is stored. The at least one instruction, the at least one program, the code set or the instruction set is loaded and executed by a processor to implement the aforementioned elevator risk warning method.

[0079] The embodiment of the present application designs a new network structure of a risk warning model, which can realize fine-grained feature extraction and feature integration by period and time clustering, and can avoid the noise caused by the distribution differences between data acquired at different times. Combined with the fault sub-data corresponding to the maintenance cycle, fine-grained risk prediction error correction is realized, thereby significantly improving the accuracy of risk prediction, and specific fault prediction results can also be output in a fine-grained manner.

[0080] The above disclosure is only the preferred embodiment of the present invention, which certainly cannot be used to limit the scope of the present invention. Therefore, equivalent changes made according to the claims of the present invention are still within the scope of the present invention.

Claims

1. An elevator risk early warning method, characterized in that: The elevator risk early warning method comprises: Acquire sample data, the sample data including sample sub-data corresponding to multiple maintenance cycles of a sample elevator, risk labels of the sample elevator, and fault sub-data corresponding to the sample sub-data, each of the sample sub-data including sample basic sub-data and a weather data sequence; Input each of the sample sub-data into a fault prediction sub-model in a risk warning model to obtain a fault prediction value, wherein the fault prediction sub-model is used to perform the following operations: extract features from the sample sub-data to obtain corresponding feature information; and obtain the fault prediction value based on the feature information; The fault prediction sub-model inputs the feature information corresponding to each of the sample sub-data into the risk prediction sub-model in the risk warning model to obtain a risk prediction value, and the risk prediction sub-model is used to perform the following operations: horizontally splicing each of the feature information to obtain a sample comprehensive feature; by masking the sample comprehensive feature, obtaining a sample effective comprehensive feature corresponding to each of the sample sub-data, and the sample effective comprehensive feature and the sample comprehensive feature have the same number of channels; extracting features from each of the sample effective comprehensive features and each fault sub-data to obtain a feature set corresponding to the sample data; and obtaining the risk prediction value according to the feature set; Adjusting the parameters in the fault prediction sub-model according to the difference between each fault prediction value and the corresponding fault sub-data; and adjusting the parameters of the risk prediction sub-model according to the difference between the risk prediction value and the risk label until the parameter adjustment stop condition is reached; Elevator risk warning is performed based on the risk warning model obtained after parameter adjustment.

2. The elevator risk warning method according to claim 1, characterized in that: The maintenance cycle refers to a single maximum time interval between the time when the sample elevator is in a normal operating state and the time before the last failure.

3. The elevator risk warning method according to claim 1 or 2, characterized in that: The sample basic sub-data has timestamp information, and the feature extraction of the sample sub-data to obtain corresponding feature information includes: Clustering each sample basic sub-data in the sample sub-data based on the timestamp information to obtain a preset number of clusters; Channel expansion is performed according to the product of the number of clusters and the data channel of the sample basic sub-data to obtain the channel space; Determine a valid channel for each cluster in the channel space, deploy sample basic sub-data of the cluster in the valid channel, and obtain sample channel sub-data corresponding to the sample basic sub-data; Use the feature extractor corresponding to each cluster to extract features from all sample channel sub-data in the cluster to obtain cluster features corresponding to the cluster; Fusion of the cluster features to obtain a first feature; Performing time-series-based feature extraction on the weather data sequence of the sample sub-data to obtain a second feature; The first feature and the second feature are fused to obtain feature information corresponding to the sample sub-data.

4. The elevator risk warning method according to claim 3 is characterized in that: The fault prediction sub-model includes a feature network and a first prediction network, the feature network is used to obtain feature information, the first prediction network is used to obtain the fault prediction value based on the feature information, the feature network includes each of the feature extractors, a first fusion unit and a second fusion unit, the first fusion unit is used to fuse cluster features, and the second fusion unit is used to fuse the first feature and the second feature; The adjusting the parameters in the fault prediction sub-model according to the difference between each fault prediction value and the corresponding fault sub-data includes: adjusting the parameters of the feature network and the first prediction network according to the difference between each fault prediction value and the corresponding fault sub-data.

5. The elevator risk warning method according to claim 4, characterized in that: The method of obtaining the effective sample comprehensive feature corresponding to each of the sample sub-data by performing information masking on the sample comprehensive feature includes: for each of the sample sub-data, copying a sample comprehensive feature, and setting the feature channels in the sample comprehensive feature that do not belong to the sample sub-data to blank, to obtain the corresponding effective sample comprehensive feature.

6. The elevator risk warning method according to claim 5, characterized in that: The risk prediction sub-model includes: a multi-layer perceptron, a third fuser, a convolutional network, a fourth fuser, and a second prediction network corresponding to each sample sub-data, each of the multi-layer perceptrons is connected to the third fuser, the convolutional network and the third fuser are connected to the fourth fuser, and the fourth fuser is connected to the second prediction network; The feature extraction is performed on each effective comprehensive feature of the sample and each fault sub-data to obtain a feature set corresponding to the sample data, including: Based on the multi-layer perceptron corresponding to each sample sub-data, feature extraction is performed on the corresponding sample effective comprehensive feature to obtain the corresponding third feature; fusing the third features based on the third fuser to obtain a fourth feature; Extracting features of each fault sub-data based on the convolutional network to obtain a fifth feature; The feature set is obtained by fusing the fourth feature and the fifth feature based on the fourth fuser.

7. The elevator risk warning method according to claim 6, characterized in that: The step of obtaining the risk prediction value according to the feature set includes: The second prediction network outputs the risk prediction value according to the feature set; The adjusting the parameters of the risk prediction sub-model according to the difference between the risk prediction value and the risk label comprises: According to the difference between the risk prediction value and the risk label, the parameters of the multilayer perceptron, the third fuser, the convolutional network, the fourth fuser and the second prediction network are adjusted.

8. An elevator risk warning device, characterized in that: The elevator risk warning device comprises: A training module that performs the following operations: Acquire sample data, the sample data including sample sub-data corresponding to multiple maintenance cycles of a sample elevator, risk labels of the sample elevator, and fault sub-data corresponding to the sample sub-data, each of the sample sub-data including sample basic sub-data and a weather data sequence; Input each of the sample sub-data into a fault prediction sub-model in a risk warning model to obtain a fault prediction value, wherein the fault prediction sub-model is used to perform the following operations: extract features from the sample sub-data to obtain corresponding feature information; and obtain the fault prediction value based on the feature information; The fault prediction sub-model inputs the feature information corresponding to each of the sample sub-data into the risk prediction sub-model in the risk warning model to obtain a risk prediction value, and the risk prediction sub-model is used to perform the following operations: horizontally splicing each of the feature information to obtain a sample comprehensive feature; by masking the sample comprehensive feature, obtaining a sample effective comprehensive feature corresponding to each of the sample sub-data, and the sample effective comprehensive feature and the sample comprehensive feature have the same number of channels; extracting features from each of the sample effective comprehensive features and each fault sub-data to obtain a feature set corresponding to the sample data; and obtaining the risk prediction value according to the feature set; Adjusting the parameters in the fault prediction sub-model according to the difference between each fault prediction value and the corresponding fault sub-data; and adjusting the parameters of the risk prediction sub-model according to the difference between the risk prediction value and the risk label until the parameter adjustment stop condition is reached; The application module is used to provide elevator risk warning based on the risk warning model obtained after parameter adjustment.

9. A computer device, characterized in that: The computer device includes a processor and a memory, wherein the memory stores at least one instruction, at least one program, a code set or an instruction set, and the at least one instruction, the at least one program, the code set or the instruction set is loaded and executed by the processor to implement the elevator risk warning method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: The storage medium stores at least one instruction, at least one program, a code set or an instruction set, and the at least one instruction, the at least one program, the code set or the instruction set is loaded and executed by the processor to implement the elevator risk warning method as described in any one of claims 1 to 7.