A safety early warning method for elevator

Abstract

The present invention discloses a safety early warning method for an elevator, comprising the following steps: constructing a prediction model based on the basic information of an existing elevator; then using the date of the day as a time slice, extracting and preprocessing the basic information of the elevator to be predicted by the prediction model; then classifying the basic information of the elevator according to the information type to generate a feature vector of the elevator; then scoring each feature vector individually by the prediction model; then scoring and rating each individual risk index of the elevator to be predicted according to the score of each feature vector, then scoring and rating the comprehensive risk index of the elevator according to the score of each individual risk index of the elevator, and generating an elevator safety and health report including the score of the individual risk index of the elevator, the level of the individual risk index of the elevator, the score of the comprehensive risk index of the elevator and the level of the comprehensive risk index of the elevator. The present invention can effectively improve the prediction accuracy of the risk level of the elevator.

Description

Technical Field

[0001] The invention relates to a method for predicting elevator safety, in particular to a safety early warning method for an elevator. Background Art

[0002] As one of the most commonly used means of transportation, elevators are closely related to the safety of people's lives, so it is necessary to improve the operating stability of elevators as much as possible to reduce the occurrence of accidents. In addition, as a complex electromechanical device, elevators have complex structures and strong correlations. Their safety is not only related to the quality level of the elevator itself, but also closely related to multi-dimensional factors such as daily use, maintenance, and inspection of the elevator. This makes it difficult for conventional monitoring methods and early warning methods for equipment parameters to ensure the stable operation of elevators.

[0003] In response to the above problems, some manufacturers have developed a prediction model for evaluating the operation risk of elevators. The prediction model inputs the basic information and operation parameters of the elevator into the prediction model, and then the prediction model evaluates and scores the elevator risk by integrating various data, thereby identifying the potential safety hazards and failure probability of the elevator. However, the defect of the existing prediction model in the actual prediction process is that since the number of indicators based on the risk assessment process of the elevator is as high as hundreds or thousands, and the score range and importance of each indicator cannot be individually defined, the prediction model will inevitably dilute the weight of the elevator's single indicator based on the comprehensive data of various aspects. For example, when a certain important indicator of the elevator forms a high-risk value, the prediction model will calculate the indicator value and other indicators with low risk values ​​of the elevator. The final score of the elevator risk calculated will be lower than the single score of the important indicator, and even the final risk level of the elevator will be diluted from high risk to higher risk or other lower risk levels, thereby weakening the harmfulness of the high-risk indicator.

[0004] On the other hand, existing prediction models generally use multiplication to calculate the final score based on each individual indicator. This results in a serious impact on the accuracy of the final score when the score of any indicator is too low or even close to zero. Even if the manufacturer makes targeted adjustments to the weights of different indicators when designing the algorithm, a single score that is too low will inevitably cause a decrease in the final score after multiplying it with other scores, thereby reducing the accuracy of elevator predictions.

[0005] Therefore, the prediction method of the existing elevator risk prediction model has the problem of low accuracy. Summary of the invention

[0006] The purpose of the present invention is to provide a safety early warning method for an elevator, which can effectively improve the prediction accuracy of the risk level of the elevator.

[0007] The technical solution of the present invention is a safety early warning method for an elevator, comprising the following steps:

[0008] A. Build a prediction model based on the basic information of existing elevators;

[0009] B. Taking the current date as the time slice, the prediction model extracts and preprocesses the basic information of the elevator to be predicted, finds the normal information within a reasonable range and the abnormal information and missing information beyond the reasonable range, and then processes the abnormal information and missing information;

[0010] C. Classify the basic information of the elevator according to the information type, so as to classify each basic information into the subdivision type of the corresponding single risk indicator of the elevator and generate the feature vector of the elevator;

[0011] D. Each feature vector is scored individually by the prediction model;

[0012] E. Score and rate each individual risk indicator of the elevator to be predicted according to the score of each eigenvector, and then score and rate the comprehensive risk indicator of the elevator according to the score of each individual risk indicator of the elevator, and generate an elevator safety and health report including the score of the individual risk indicator of the elevator, the grade of the individual risk indicator of the elevator, the score of the comprehensive risk indicator of the elevator and the grade of the comprehensive risk indicator of the elevator. The individual risk indicators of the elevator include the elevator operation risk indicator, the maintenance risk indicator, the inspection and testing risk indicator, and the usage management risk indicator.

[0013] In the aforementioned elevator safety warning method, the scores of the elevator single risk index and the elevator comprehensive risk index in step E are quantitative values ​​within the range of 0-100, and the level of the elevator single risk index is segmented according to the score of the elevator single risk index, and the level of the elevator comprehensive risk index is segmented according to the score of the elevator comprehensive risk index, wherein a score ≥85 is high risk, a score ≥60 and <85 is relatively high risk, a score ≥30 and <60 is medium risk, and a score <30 is low risk.

[0014] In the aforementioned elevator safety warning method, the method for constructing the prediction model in step A comprises the following steps:

[0015] A1. Input all the basic information of elevators and elevator fault data into the prediction model;

[0016] A2. Take all elevators with fault data as positive samples, and use the date of the elevator fault as the time slice to collect the feature vector of the faulty elevator; at the same time, take the elevator that has not failed as a negative sample, and collect the feature vector of the elevator at the same time slice as the positive sample;

[0017] A3. Use the prediction model to score and rate the elevator single risk index and the elevator comprehensive risk index in each positive and negative sample to obtain the elevator risk profile;

[0018] A4. Predict the failure probability of the elevator based on its risk profile, then compare the predicted result with the actual result, select the prediction model with the highest accuracy and output it.

[0019] In the aforementioned elevator safety warning method, the basic elevator information corresponding to the elevator operation risk index includes basic attribute information of the elevator and IoT monitoring information;

[0020] The basic elevator information corresponding to the inspection and detection risk index is the relevant information and inspection conclusion of the elevator inspection;

[0021] The basic elevator information corresponding to the maintenance risk indicator is the elevator maintenance unit situation and the current elevator maintenance data;

[0022] The basic information of the elevator corresponding to the usage management risk indicator is the relevant data characteristics of the elevator user unit.

[0023] In the aforementioned elevator safety early warning method, when scoring the elevator single risk index, each characteristic vector in the elevator single risk index is first scored separately, and then the highest score among the elevator characteristic vectors is counted;

[0024] If the highest score is greater than the lower limit of the high risk level, the total score of the elevator single risk indicator is the sum of the scores of other feature vectors divided by the sum of the total scores of other feature vectors, then multiplied by the difference between the full score and the highest score, plus the highest score;

[0025] If the highest score is not greater than the lower limit of the high-risk level, the total score of the elevator's single risk indicator is the sum of the scores of other feature vectors divided by the sum of the total scores of other feature vectors, and then multiplied by the difference between the lower limit of the high-risk level and the highest score, plus the highest score.

[0026] In the aforementioned safety early warning method for an elevator, when scoring the comprehensive risk index of the elevator, the highest score among the individual risk indicators of each elevator is firstly calculated;

[0027] If the highest score is greater than the lower limit of the high risk level, the total score of the comprehensive risk index of the elevator is the sum of the scores of other individual risk indicators of the elevator divided by the total score of other individual risk indicators of the elevator, then multiplied by the difference between the full score and the highest score, and then added to the highest score;

[0028] If the highest score is not greater than the lower limit of the high-risk level, the total score of the elevator comprehensive risk index is the sum of the scores of other elevator individual risk indicators divided by the total score of other elevator individual risk indicators, and then multiplied by the difference between the lower limit of the high-risk level and the highest score, plus the highest score.

[0029] In the aforementioned elevator safety warning method, the method for processing abnormal information and missing information in step B is specifically as follows:

[0030] If the abnormal or missing information is a numerical feature that cannot be counted, it is processed by direct deletion or related feature replacement;

[0031] If the abnormal information or missing information is a numerical feature that can be obtained statistically, it is processed through statistical analysis;

[0032] If the abnormal information or missing information is a discrete feature with hierarchical classification, the true value conversion method is used for processing.

[0033] In the aforementioned elevator safety warning method, the direct deletion method is to directly delete the elevator single risk indicator subdivision type corresponding to the abnormal information or missing information;

[0034] The related feature replacement method is to use other related features to replace the abnormal information or missing information as the basic information of the elevator;

[0035] The statistical analysis method is to extract the same type of features of other elevators for statistical analysis, calculate the mode, median or mean of other elevators to replace the abnormal information or missing information;

[0036] The truth value conversion method is to replace the state of the abnormal information or missing information with a corresponding field.

[0037] Compared with the prior art, the present invention has the following characteristics:

[0038] (1) The present invention can predict the elevator operation risk from multiple perspectives compared with the existing calculation method by dividing each elevator single risk index and the elevator comprehensive risk index, and enables the prediction model to calculate the corresponding risk level according to different types of elevator indicators during calculation, thereby improving the prediction accuracy of the elevator risk; on this basis, by scoring each eigenvector in each elevator single risk index separately, and calculating the elevator single risk index based on the score of each eigenvector, the score of the elevator single risk index and the elevator comprehensive risk index can be calculated by combining the eigenvectors, thereby avoiding the decrease in prediction accuracy caused by the omission of some eigenvectors;

[0039] (2) By optimizing the scoring method of each elevator single risk index and the elevator comprehensive risk index, when there is an indicator of high risk level in the feature vector, the score of the elevator single risk index where the high risk index is located can be stabilized above the score of the high risk index, thereby effectively preventing the weight dilution of the high risk index caused by other low risk indicators during calculation, and when the score of some feature vectors is too low or reaches zero, the score of the elevator single risk index will not be affected by this part of the feature vector and cause the score to decrease, thereby further improving the prediction accuracy of the present invention;

[0040] (3) Since the above-mentioned scoring method will cause the final score of the elevator single risk index to be affected by the highest score of the elevator single risk index, when an information entry error or a sensor detection error occurs, causing the elevator index extracted by the prediction model to be abnormal, the abnormal information will cause the prediction model to be inaccurate during calculation. In response to the above-mentioned problem, the present invention defines the processing method for abnormal information and missing information, so that the prediction model can extract and process this part of information, thereby preventing the abnormal information from affecting the score of the elevator single risk index;

[0041] Therefore, the present invention can effectively improve the prediction accuracy of the elevator risk level. DETAILED DESCRIPTION

[0042] The present invention will be further described below in conjunction with the embodiments, but they are not intended to limit the present invention.

[0043] Embodiment: A safety warning method for an elevator comprises the following steps:

[0044] A. Build a prediction model based on the basic information of existing elevators;

[0045] B. Taking the date of the day as the time slice, the prediction model extracts the basic information of the elevator to be predicted, and identifies the basic information based on the business method, the 3σ principle and / or the box plot analysis method, finds the normal information within the reasonable range set by humans, and the abnormal information and missing information beyond the reasonable range, and then processes the abnormal information and missing information. The abnormal information is a numerical indicator that can be achieved beyond the theoretical range of the elevator;

[0046] C. Classify the basic information of the elevator according to the information type, so as to classify each basic information into the subdivision type of the corresponding single risk indicator of the elevator and generate the feature vector of the elevator;

[0047] D. Each feature vector is scored individually by the prediction model;

[0048] E. Score and rate each individual risk indicator of the elevator to be predicted according to the score of each eigenvector, and then score and rate the comprehensive risk indicator of the elevator according to the score of each individual risk indicator of the elevator, and generate an elevator safety and health report including the score of the individual risk indicator of the elevator, the grade of the individual risk indicator of the elevator, the score of the comprehensive risk indicator of the elevator and the grade of the comprehensive risk indicator of the elevator. The individual risk indicators of the elevator include the elevator operation risk indicator, the maintenance risk indicator, the inspection and testing risk indicator, and the usage management risk indicator.

[0049] The scores of the elevator single risk index and the elevator comprehensive risk index in the step E are quantitative values ​​within the range of 0-100, with 2 decimal places; the level of the elevator single risk index is segmented according to the score of the elevator single risk index, and the level of the elevator comprehensive risk index is segmented according to the score of the elevator comprehensive risk index, wherein a score ≥85 is high risk, a score ≥60 and <85 is relatively high risk, a score ≥30 and <60 is medium risk, and a score <30 is low risk.

[0050] The method for constructing the prediction model in step A comprises the following steps:

[0051] A1. Input the basic information of all elevators and elevator failure data in the historical records into the prediction model;

[0052] A2. Take all elevators with fault data as positive samples, and use the date of the elevator fault as the time slice to collect the feature vector of the faulty elevator; at the same time, take the elevator that has not failed as a negative sample, and collect the feature vector of the elevator at the same time slice as the positive sample;

[0053] A3. Use the LightGBM model based on the Boosting ensemble algorithm to score and rate the elevator single risk index and the elevator comprehensive risk index in each positive and negative sample to obtain the risk profile of the elevator;

[0054] A4. Predict the failure probability of the elevator based on the risk profile of the elevator, and then use the formula:

[0055]

[0056] Calculate the precision, recall, and F1-score and use them as evaluation indicators to evaluate the performance of the prediction model, select the prediction model with the highest accuracy and output it.

[0057] The basic information of the elevator corresponding to the elevator operation risk index includes the basic attribute information of the elevator and the Internet of Things monitoring information. The basic attribute information of the elevator specifically includes elevator manufacturing data, elevator parameters and elevator usage, among which the elevator manufacturing data includes elevator classification, elevator manufacturing time, elevator installation time, elevator commissioning time and elevator model specifications, and the elevator parameters include the number of elevator floors, the number of stations, the number of doors, the lifting height and the rated speed, and the elevator usage includes the elevator usage status and the elevator usage place, and the Internet of Things monitoring information specifically refers to the features extracted from the Internet of Things monitoring related data, such as various types of elevator operation data and various types of alarm data, and the like;

[0058] The basic elevator information corresponding to the inspection and detection risk index is the relevant information and inspection conclusion of the elevator inspection, including the inspection time, whether the inspection is overdue, inspection defects, and the relationship between the fault occurrence time and the inspection time node;

[0059] The basic information of the elevator corresponding to the maintenance risk index is the situation of the elevator maintenance unit and the current elevator maintenance data, including quantitative evaluation information of the maintenance unit, the man-machine ratio of the maintenance unit, whether the current maintenance is overdue, historical overdue maintenance, maintenance type and interval, etc.;

[0060] The basic information of the elevator corresponding to the management risk index is the relevant data characteristics of the elevator user unit, including whether there is an accident, whether there is an administrative penalty, the deployment of safety management personnel, etc.

[0061] When scoring the elevator single risk index, each characteristic vector in the elevator single risk index is scored separately first, and then the highest score among the elevator characteristic vectors is calculated;

[0062] If the highest score is greater than the lower limit of the high risk level, the total score of the elevator single risk indicator is the sum of the scores of other feature vectors divided by the sum of the total scores of other feature vectors, then multiplied by the difference between the full score and the highest score, plus the highest score;

[0063] If the highest score is not greater than the lower limit of the high-risk level, the total score of the elevator's single risk indicator is the sum of the scores of other feature vectors divided by the sum of the total scores of other feature vectors, and then multiplied by the difference between the lower limit of the high-risk level and the highest score, plus the highest score.

[0064] Take the scoring method of maintenance risk indicators as an example: For the calculation rules, each eigenvector in the maintenance risk index is scored separately, and then the highest score max_wb in the eigenvector is counted; and then the maintenance risk index is scored according to the following calculation rules:

[0065]

[0066] where wb_T i represents the total score of the ith maintenance indicator, t_high represents the lower limit of the high risk level, total_score represents the full score of the feature vector, and wb_score represents the score of the maintenance risk indicator;

[0067] If any characteristic vector in the maintenance risk index scores 90 points, and the scores of the remaining 19 characteristic vectors are 50 points, the total score of the elevator single risk index is 95 points;

[0068] If the score of any characteristic vector in the maintenance risk index is 70 points, and the scores of the remaining 19 characteristic vectors are 50 points; the total score of the elevator single risk index is 77.5 points.

[0069] When scoring the comprehensive risk index of elevators, the highest score among the individual risk indicators of each elevator is first calculated;

[0070] If the highest score is greater than the lower limit of the high risk level, the total score of the comprehensive risk index of the elevator is the sum of the scores of other individual risk indicators of the elevator divided by the total score of other individual risk indicators of the elevator, then multiplied by the difference between the full score and the highest score, and then added to the highest score;

[0071] If the highest score is not greater than the lower limit of the high-risk level, the total score of the elevator comprehensive risk index is the sum of the scores of other elevator individual risk indicators divided by the total score of other elevator individual risk indicators, and then multiplied by the difference between the lower limit of the high-risk level and the highest score, plus the highest score.

[0072] The specific method for processing abnormal information and missing information in step B is:

[0073] If the abnormal information or missing information is a numerical feature that cannot be counted, such as the time the elevator was put into use, the number of elevator floors, the type of elevator traction, etc., it will be processed by direct deletion or related feature replacement;

[0074] If the abnormal information or missing information is a numerical feature that can be obtained by statistics, such as the elevator speed, it will be processed through statistical analysis;

[0075] If the abnormal information or missing information is a discrete feature with hierarchical classification, such as the maintenance unit rating level, factory risk level, etc., the true value conversion method is used for processing;

[0076] The numerical feature ranges that cannot be counted in the basic information of the elevator, the numerical feature ranges obtained by statistics, and the discrete feature ranges with hierarchical classification are all divided manually.

[0077] The direct deletion method is to directly delete the elevator single risk indicator subdivision type corresponding to the abnormal information or missing information;

[0078] The related feature replacement method is to use other related features to replace the abnormal information or missing information as the basic information of the elevator, such as replacing the missing information with any conventional elevator traction type as the basic information of the elevator;

[0079] The statistical analysis method is to extract the same type of features of other elevators for statistical analysis, calculate the mode, median or mean of other elevators to replace the abnormal information or missing information;

[0080] The truth value conversion method is to replace the status of the abnormal information or missing information with the corresponding field, such as the missing information is uniformly replaced with "not evaluated", and the abnormal information is replaced with "new level" or "risk level".

[0081] The prediction model generates and updates samples at a fixed time every day, and extracts new feature vectors and constructs new samples based on the various types of data added that day and stores them in the training data feature table. The prediction model is regularly trained after each update, and the latest model training results are stored to improve the prediction accuracy of the prediction model. At the same time, the prediction model predicts the failure probability of all elevators based on the samples updated every day, and then generates the corresponding failure probability value and label value and stores them in the prediction result table.

[0082] Working principle of the present invention: The present invention summarizes and divides various basic information related to elevator operation stability into four categories: elevator operation risk indicators, maintenance risk indicators, inspection and testing risk indicators, and use management risk indicators, so that it can increase the extraction range of existing prediction indicators compared to the existing elevator risk prediction method, thereby improving the prediction accuracy of the prediction model for elevators. On this basis, through the algorithm optimization of the prediction model, the score of each elevator single risk indicator can be calculated with the highest value of the indicator score within the indicator range, thereby ensuring that when a high-risk indicator appears in the elevator single risk indicator, the score of the elevator single risk indicator will not be lower than the score of the high-risk indicator, that is, avoiding the weight dilution caused by other low-risk indicators in the elevator single risk indicator, and at the same time avoiding the return of its overall score to zero when the indicator with a score of 0 appears; ensuring the prediction accuracy of the prediction model.

[0083] On this basis, by processing the extracted abnormal information and missing information, it can also effectively prevent the extracted basic information from being classified as a high-risk indicator by the prediction model during calculation due to collection errors or data filling errors, thereby seriously affecting the score and level of the elevator's single risk indicator, thereby improving the accuracy of elevator risk prediction.

Claims

1. A safety warning method for an elevator, characterized in that: The following steps are involved: A. Build a prediction model based on the basic information of existing elevators; B. Taking the current date as the time slice, the prediction model extracts and preprocesses the basic information of the elevator to be predicted, finds the normal information within a reasonable range and the abnormal information and missing information beyond the reasonable range, and then processes the abnormal information and missing information; C. Classify the basic information of the elevator according to the information type, so as to classify each basic information into the subdivision type of the corresponding single risk indicator of the elevator and generate the feature vector of the elevator; D. Each feature vector is scored individually by the prediction model; E. scoring and rating each individual risk indicator of the elevator to be predicted according to the score of each characteristic vector, and then scoring and rating the comprehensive risk indicator of the elevator according to the score of each individual risk indicator of the elevator, and generating an elevator safety and health report including the score of the individual risk indicator of the elevator, the level of the individual risk indicator of the elevator, the score of the comprehensive risk indicator of the elevator and the level of the comprehensive risk indicator of the elevator, wherein the individual risk indicator of the elevator includes the elevator operation risk indicator, the maintenance risk indicator, the inspection and testing risk indicator, and the use and management risk indicator; When scoring the elevator single risk index, each characteristic vector in the elevator single risk index is scored separately first, and then the highest score among the elevator characteristic vectors is calculated; If the highest score is greater than the lower limit of the high risk level, the total score of the elevator single risk indicator is the sum of the scores of other feature vectors divided by the sum of the total scores of other feature vectors, then multiplied by the difference between the full score and the highest score, plus the highest score; If the highest score is not greater than the lower limit of the high-risk level, the total score of the elevator's single risk indicator is the sum of the scores of other feature vectors divided by the sum of the total scores of other feature vectors, and then multiplied by the difference between the lower limit of the high-risk level and the highest score, plus the highest score.

2. The elevator safety warning method according to claim 1, characterized in that: The scores of the elevator single risk index and the elevator comprehensive risk index in the step E are quantitative values ​​within the range of 0-100. The level of the elevator single risk index is segmented according to the score of the elevator single risk index, and the level of the elevator comprehensive risk index is segmented according to the score of the elevator comprehensive risk index, wherein a score ≥85 is high risk, a score ≥60 and <85 is relatively high risk, a score ≥30 and <60 is medium risk, and a score <30 is low risk.

3. The elevator safety warning method according to claim 2, characterized in that: The method for constructing the prediction model in step A comprises the following steps: A1. Input all the basic information of elevators and elevator fault data into the prediction model; A2. Take all elevators with fault data as positive samples, and use the date of the elevator fault as the time slice to collect the feature vector of the faulty elevator; at the same time, take the elevator that has not failed as a negative sample, and collect the feature vector of the elevator at the same time slice as the positive sample; A3. Use the prediction model to score and rate the elevator single risk index and the elevator comprehensive risk index in each positive and negative sample to obtain the elevator risk profile; A4. Predict the failure probability of the elevator based on its risk profile, then compare the predicted result with the actual result, select the prediction model with the highest accuracy and output it.

4. The elevator safety warning method according to claim 1, characterized in that: The basic elevator information corresponding to the elevator operation risk index includes basic attribute information and IoT monitoring information of the elevator; The basic elevator information corresponding to the inspection and detection risk index is the relevant information and inspection conclusion of the elevator inspection; The basic elevator information corresponding to the maintenance risk indicator is the elevator maintenance unit situation and the current elevator maintenance data; The basic information of the elevator corresponding to the usage management risk indicator is the relevant data characteristics of the elevator user unit.

5. The elevator safety warning method according to claim 1, characterized in that: When scoring the comprehensive risk index of elevators, the highest score among the individual risk indicators of each elevator is first calculated; If the highest score is greater than the lower limit of the high risk level, the total score of the comprehensive risk index of the elevator is the sum of the scores of other individual risk indicators of the elevator divided by the total score of other individual risk indicators of the elevator, then multiplied by the difference between the full score and the highest score, and then added to the highest score; If the highest score is not greater than the lower limit of the high-risk level, the total score of the elevator comprehensive risk index is the sum of the scores of other elevator individual risk indicators divided by the total score of other elevator individual risk indicators, and then multiplied by the difference between the lower limit of the high-risk level and the highest score, plus the highest score.

6. The elevator safety warning method according to claim 1, characterized in that: The specific method for processing abnormal information and missing information in step B is: If the abnormal or missing information is a numerical feature that cannot be counted, it is processed by direct deletion or related feature replacement; If the abnormal information or missing information is a numerical feature that can be obtained statistically, it is processed through statistical analysis; If the abnormal information or missing information is a discrete feature with hierarchical classification, the true value conversion method is used for processing.

7. The elevator safety warning method according to claim 6, characterized in that: The direct deletion method is to directly delete the elevator single risk indicator subdivision type corresponding to the abnormal information or missing information; The related feature replacement method is to use other related features to replace the abnormal information or missing information as the basic information of the elevator; The statistical analysis method is to extract the same type of features of other elevators for statistical analysis, calculate the mode, median or mean of other elevators to replace the abnormal information or missing information; The truth value conversion method is to replace the state of the abnormal information or missing information with a corresponding field.

Images