An elevator on-demand maintenance method based on dynamic prediction

Abstract

The application discloses an elevator on-demand maintenance method based on dynamic prediction, which comprehensively considers the running state data and use condition data of the elevator, utilizes an elevator health degree evaluation model based on an LSTM algorithm, realizes evaluation on the health degree of the elevator, and obtains a maintenance priority evaluation result accordingly, and further generates a maintenance strategy; considering the maintenance resource limitation, the maintenance strategy is optimized according to the availability of maintenance personnel, tools and equipment and the maintenance cost; in the process of actual execution of the maintenance strategy, the running state of the elevator, the maintenance plan and the execution condition are monitored; and according to the feedback information, the maintenance strategy is optimized by using an optimization algorithm. The method can improve the maintenance efficiency and safety of the elevator, reduce resource waste, and provide personalized and dynamic maintenance strategies.

Description

Technical Field

[0001] The present invention relates to the field of elevator on-demand maintenance, and in particular to an elevator on-demand maintenance method based on dynamic prediction. Background Art

[0002] Elevators, as an indispensable means of transportation in modern cities, undertake the crucial task of transporting people and goods. Maintenance is essential to ensure the safe and reliable operation of elevators. Traditional elevator maintenance typically involves scheduled maintenance, with maintenance and inspections performed at fixed intervals. However, this approach presents the following problems: First, the frequency of maintenance is inconsistent; sometimes maintenance may be frequent, while other times it may go unattended for extended periods, resulting in wasted resources and unstable maintenance results. Second, it is difficult to accurately assess maintenance needs. Traditional methods rely primarily on experience and fixed maintenance cycles, failing to fully consider the actual operating conditions and usage of the elevator, leading to inaccurate maintenance plans and suboptimal results. Third, maintenance costs are high, as regular maintenance requires a significant investment of manpower and material resources, resulting in unnecessary waste.

[0003] With the rapid development of sensor technology, big data processing, and artificial intelligence algorithms, maintenance methods based on data analysis are gaining increasing attention. These methods collect elevator operating status and usage data and leverage data analysis and predictive analytics algorithms to dynamically predict elevator maintenance needs and optimize maintenance plans. However, existing data-analysis-based elevator maintenance methods still suffer from several issues, including imperfect data collection and processing, low prediction accuracy, and insufficiently refined maintenance strategy generation and optimization. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention proposes an on-demand elevator maintenance method based on dynamic prediction. This method aims to address the inconsistencies, inaccuracies, and high costs inherent in traditional elevator maintenance methods. By comprehensively considering elevator operating status and usage data, and utilizing predictive analysis algorithms to establish a predictive maintenance model, this method accurately predicts elevator maintenance needs and optimizes maintenance plans, thereby improving elevator maintenance efficiency and safety. Building on existing maintenance regulations and usage scenario data, this invention develops a comprehensive solution and technical implementation plan for intelligent elevator maintenance that achieves "on-demand maintenance and on-demand inspection."

[0005] The specific technical solutions are as follows:

[0006] An elevator on-demand maintenance method based on dynamic prediction includes the following steps:

[0007] Step one: real-time acquisition of elevator maintenance data and preprocessing thereof, extraction of elevator maintenance key features from the data, and assignment of weight values to the features according to the importance of elevator maintenance; the maintenance data includes running state data and usage data; the running state data includes the running time, speed, acceleration, and temperature of the elevator; the usage data includes the running frequency and peak period of the elevator;

[0008] Step two: statistical feature extraction of elevator maintenance key features in each maintenance cycle; according to historical maintenance data, the results of each maintenance project in the same maintenance cycle are weighted and summed to obtain an elevator health score, and the elevator health score is used for classification of the elevator health, and the classification result is used as the training label of the maintenance cycle; the statistical features of all maintenance cycles are arranged corresponding to the training labels to obtain a training data set; an elevator health evaluation model based on the LSTM algorithm is constructed and trained;

[0009] Step three: after the historical maintenance data and real-time collected elevator running data are processed by S1, the trained elevator health evaluation model is inputted, and an elevator health evaluation result is outputted, and a maintenance priority evaluation result is obtained accordingly;

[0010] Step four: generating a maintenance strategy according to the elevator health evaluation result and the priority evaluation result;

[0011] Step five: considering the maintenance resource limitation, optimizing the maintenance strategy according to the availability of maintenance personnel, tools and equipment, and maintenance cost;

[0012] Step six: monitoring the elevator running state, maintenance plan and execution during the actual execution of the maintenance strategy; and according to the feedback information, an optimization algorithm is used to optimize the maintenance strategy.

[0013] Further, in step one, the preprocessing includes data cleaning, data outlier processing, data missing value processing, and classified storage of the data.

[0014] Further, in step two, the statistical feature extraction of elevator maintenance key features in each maintenance cycle is as follows:

[0015] The preprocessed elevator running data is grouped based on the maintenance cycle, and the mathematical statistical features of the elevator running data in each maintenance cycle are extracted; for binary data with only two states, the number of times of the existence of the two states in each maintenance cycle is counted; for continuous data with multiple states, mathematical statistical feature calculation is performed in each maintenance cycle, including: calculation of maximum value, minimum value, average value, and variance; all the obtained mathematical statistical features are spliced to obtain the statistical feature data of the maintenance cycle, which is used as the training data.

[0016] Furthermore, the elevator health is divided into four categories. If the elevator health score is 0-10 points, the corresponding elevator health assessment result is normal and no maintenance is required;

[0017] If the elevator health score is 11-25 points, the corresponding elevator health assessment result is "needs attention", and maintenance is determined to be required, and the maintenance priority is low;

[0018] If the elevator health score is 26-35 points, the corresponding elevator health assessment result is that it is determined to require maintenance, and the maintenance priority is medium;

[0019] If the elevator health score is 36-45 points, the corresponding elevator health assessment result is that it is determined to require maintenance, and the maintenance priority is high.

[0020] Furthermore, the elevator health assessment model includes five layers in sequence: input layer, LSTM layer 1, LSTM layer 2, fully connected layer, and output layer;

[0021] The input layer is used to accept time series data of multiple features as an input sequence;

[0022] The LSTM layer 1 includes multiple LSTM units for capturing the basic temporal relationship between features; the number of LSTM units is between 20 and 100;

[0023] The LSTM layer 2 includes multiple LSTM units for further extracting deeper temporal relationships between features; the number of LSTM units is between 20 and 100;

[0024] The fully connected layer is used to extract the potential interaction relationship between input features;

[0025] The output of the fully connected layer is connected to the output layer, which outputs the elevator health score predicted by the model.

[0026] Furthermore, the elevator health assessment model is trained using the training data set, and the elevator health assessment model is optimized using a mean square error loss function and a stochastic gradient descent optimization algorithm during the training process.

[0027] Furthermore, the maintenance strategy includes: allocation of maintenance personnel and allocation of maintenance time.

[0028] Furthermore, in step 6, the optimization algorithm is a genetic algorithm, and the specific steps of using the genetic algorithm to optimize the maintenance strategy are as follows:

[0029] (1) Modeling the maintenance optimization problem: Determine the optimization objectives and constraints. The optimization objectives include minimizing maintenance costs and ensuring the timeliness of elevator maintenance. The constraints include ensuring that the spare parts inventory meets maintenance needs and that each elevator completes maintenance once within a specified period.

[0030] The maintenance cost C is the sum of the maintenance costs of all elevators by all maintenance personnel, and is expressed as follows:

[0031]

[0032] Where c i,j The labor cost of maintenance personnel i serving elevator j; t i,j is the transportation cost of maintenance personnel i serving elevator j; p i,j The cost of spare parts when maintenance worker i services elevator j;

[0033] The timeliness of elevator maintenance is calculated using the following expression:

[0034]

[0035] Where Reward is the sum of the incentives for the incentive algorithm to prioritize the allocation of maintenance resources to high-priority elevators, and Penalty is the sum of the penalties for high-priority elevators that fail to be maintained in time; R j is the priority coefficient of elevator j maintenance, P is the penalty coefficient, threshold is the priority coefficient threshold for elevator maintenance to be judged as high priority; α is the reward coefficient;

[0036] The fitness function of elevator maintenance optimization is obtained, and the expression is as follows:

[0037] Fitness=w1*C+w2*R

[0038] Where w1 represents the weight coefficient of minimizing maintenance cost to fitness function, w2 represents the weight coefficient of maintenance timeliness to fitness function;

[0039] The constraint expression for the spare parts library to meet maintenance requirements is as follows:

[0040]

[0041] Where x ij Indicates whether elevator j has been assigned to maintenance personnel i, x ij =1 means the task has been assigned, x ij =0 means no task is assigned; q ij are the spare parts required by maintenance personnel i to perform maintenance on elevator j, and S is the total amount of spare parts in the warehouse;

[0042] The constraint expression of each elevator completing maintenance once in a specified period is as follows:

[0043]

[0044] (2) Randomly generate N elevator maintenance allocation tasks as initial solutions, and take the existing maintenance task allocation method as one of the initial solutions, N [20, 100]; after generating the initial solutions, check whether the initial solutions meet the constraint requirements, if not, discard the initial solutions and generate new initial solutions, and all the initial solutions form an initial population;

[0045] (3) According to the fitness function, the fitness of all initial solutions is calculated, and the top 10% of initial solutions with the highest fitness are reserved to the next generation population; the remaining 90% of initial solutions are allocated according to the fitness, and the probability of being selected as a parent is higher, and the probability of being selected as a parent is higher;

[0046] Randomly select two parents from the initial solutions with the allocation probability to generate offspring through crossover and random mutation, and repeat the operation of randomly selecting and generating offspring until 0.9*N offspring are generated, and the generated offspring and the top 10% of initial solutions with the highest fitness form a new population;

[0047] (4) Repeat step (3) until the highest fitness value in the population approaches 0, then terminate the cycle, and output the maintenance allocation scheme with the highest fitness value as the optimal solution.

[0048] The beneficial effects of the present application are:

[0049] The present application formulates a maintenance demand prediction optimization strategy based on continuously updated data, integrates complex factors such as aging and reconstruction of elevators into the original prediction model, optimizes the accuracy of the model, and enhances the accuracy, practicality and adaptability of the original model. At the same time, accurate and comprehensive maintenance data is opened to users to enhance their trust in personalized and dynamic maintenance strategies. BRIEF DESCRIPTION OF DRAWINGS

[0050] Figure 1 is the flow chart of the elevator on-demand maintenance method based on dynamic prediction proposed by the embodiment of the present application. DETAILED DESCRIPTION

[0051] The purpose and effect of the present application will become more apparent from the following detailed description of the drawings and preferred embodiments. The present application will be further described in detail below in combination with the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and do not limit the present application.

[0052] As Figure 1As shown, a dynamic prediction-based elevator on-demand maintenance method includes the following steps:

[0053] S1: Disassemble the elevator on-demand maintenance problem.

[0054] In this step, the formulation of the elevator maintenance strategy is disassembled in detail, including: maintenance items, maintenance standards, maintenance basis, maintenance scenarios, etc. These details will be used for subsequent data collection, analysis and prediction.

[0055] S2: Data collection.

[0056] Collect elevator historical maintenance data and elevator operation data.

[0057] Specifically, the elevator historical maintenance data includes the machine room environment, operating device status, drive host vibration, brake status, encoder status, etc.

[0058] In this embodiment, the data in the elevator quarterly maintenance and repair sheet of a certain elevator company is taken as an example of elevator historical maintenance data, and the maintenance and repair items are taken as feature data. The result of each item is divided into three cases: normal (complying with requirements, test, and test function normal), abnormal (discovered not to comply, need to adjust or replace parts), and no (the elevator has no such item or function). According to the needs of elevator maintenance, the feature value of the feature data with abnormal result is scored as 1, and the feature value of the feature data with normal or no result is scored as 0. That is, if it reaches 1 point, it means that the item needs to be maintained.

[0059] Specifically, the elevator operation data refers to the elevator operation state data and usage data collected by installing various sensors such as acceleration sensors, temperature sensors, and sound sensors in the elevator. The following are some examples of data collection:

[0060] Operation state data, including: elevator running time, speed, acceleration, temperature, etc.

[0061] Usage data, including: elevator running frequency, peak period, etc.

[0062] S3: Maintenance data analysis and feature engineering.

[0063] In this step, the data collected in S2 is preprocessed and feature engineered.

[0064] Data preprocessing: The collected elevator maintenance data and elevator operation data are respectively subjected to data cleaning, data outlier processing, and data missing value processing to ensure the accuracy and integrity of the data.

[0065] Feature Engineering: Select key features of elevator maintenance from preprocessed historical elevator maintenance data and elevator operation data, and assign weights based on the importance of these features to elevator maintenance.

[0066] In this embodiment, 45 key features of elevator maintenance were extracted, as shown in Table 1. In Table 1, the requirements for judging the "machine room and pulley room environment" numbered 1 as normal are that they are clean, the doors and windows are intact, and the lighting is normal; the requirement for the "manual emergency operation device status" numbered 2 is that the device is complete and located in the specified position; the requirement for the "drive host status" numbered 3 is that there is no abnormal vibration and abnormal sound during operation; the requirement for the "brake clearance status" numbered 5 is that there should be no friction between the brake lining and the brake wheel when the car is opened, and the clearance value meets the requirements of the manufacturer; the requirement for the "brake self-monitoring status" numbered 6 specifically refers to the self-monitoring status of the brake when it is used as the stopping subsystem of the car accidental movement protection device; The requirement for "encoder status" No. 7 is to be clean and firmly installed; the requirement for "oil cup status on guide shoe" No. 13 specifically refers to whether the oil amount in the oil cup is appropriate and whether there is any leakage; the requirement for "engagement length of floor door locking element" No. 29 is to be no less than 7mm; the requirement for "pit environment status" No. 30 is to be clean, without water seepage or accumulation, and with normal lighting; the requirement for "reducer lubricating oil status" No. 33 is to have an appropriate amount of lubricating oil and no leakage; the requirement for "traction wheel groove and suspension device cleanliness, wire rope status" No. 37 is to be free of severe grease, have uniform tension, and meet the requirements of the manufacturer.

[0067] Table 1 Elevator maintenance feature number table

[0068]

[0069] According to the importance of these key features to elevator maintenance, these key features are divided into three levels: A, B, and C. The weights are assigned to Class A features as 0.2, Class B features as 0.5, and Class C features as 1. The classification is shown in Table 2.

[0070] Table 2 Elevator maintenance feature classification and weight assignment table

[0071]

[0072] S4: Elevator health prediction and maintenance needs assessment, mainly including the following steps:

[0073] S4.1: First, build and train the elevator health assessment model. This is achieved by the following steps:

[0074] S4.1.1: Constructing training data set: The pre-processed elevator operation data is grouped based on the maintenance period, and the mathematical statistical features of the elevator operation data in each maintenance period are extracted to obtain the training data of the maintenance period. Specifically, for the elevator door opening and closing state, it belongs to binary data, and the number of opening and closing doors in each maintenance period is counted; for continuous data such as elevator running speed, elevator running acceleration, and elevator running temperature, the maximum value, minimum value, average value, variance, etc. of each maintenance period are calculated; all the mathematical statistical features obtained are spliced to obtain the statistical feature data (i.e. training data) of the maintenance period.

[0075] The historical maintenance data processed by S3 is weighted and summed for each maintenance project result in the same maintenance period to obtain the elevator health score, and the elevator health classification is performed according to the elevator health score, and the health classification result obtained is used as the training label of the maintenance period.

[0076] In this embodiment, the elevator health classification is performed according to the method shown in Table 3.

[0077] Table 3: Correspondence table of elevator health score and elevator maintenance requirement

[0078] Elevator health score Elevator maintenance needs and priorities 0-10 Normal 11-25 Attention needed, low priority for maintenance 26-35 Warning, medium priority for maintenance 36-45 Fault, high priority for maintenance

[0079] The statistical features of all maintenance periods are arranged in correspondence with the training labels to obtain the training data set.

[0080] S4.1.2: Constructing elevator health evaluation model: The elevator health evaluation model is realized based on the LSTM algorithm, and the model includes five layers in turn: input layer, LSTM layer one, LSTM layer two, fully connected layer, and output layer.

[0081] Input layer: used to accept multiple time series data of features as input sequences.

[0082] LSTM layer one: contains multiple LSTM units for capturing basic temporal relationships between features; the number of LSTM units can be set according to requirements, generally between 20 and 100.

[0083] LSTM layer two: also contains multiple LSTM units for further extracting deeper temporal relationships between features; the number of LSTM units can be the same as or different from that of LSTM layer one, and the number of LSTM units is also between 20 and 100.

[0084] Fully connected layer: used to extract potential interaction relationships between input features.

[0085] Output layer: connects the output of the fully connected layer to the output layer, and outputs the elevator health score predicted by the model.

[0086] S4.1.3: Use the training dataset constructed in S4.1.1 to train the elevator health assessment model and adopt appropriate loss functions and optimization algorithms during the training process, such as the mean squared error (MSE) loss function and the stochastic gradient descent (SGD) optimization algorithm to optimize the model.

[0087] S4.2: The historical elevator maintenance data and the real-time collected elevator operation data are used as the elevator data to be evaluated. After processing, they are input into the trained elevator health assessment model for calculation and analysis to obtain real-time elevator health assessment results.

[0088] S4.3: Maintenance Needs Assessment: Determine whether the elevator requires maintenance and its maintenance priority based on the elevator health assessment results. Specifically, if the elevator health assessment result is normal, it is determined that maintenance is not required; if the elevator health assessment result is "Attention Required," it is determined that maintenance is required and has a low maintenance priority; if the elevator health assessment result is "Warning," it is determined that maintenance is required and has a medium maintenance priority; if the elevator health assessment result is "Faulty," it is determined that maintenance is required and has a high maintenance priority.

[0089] In this step, when performing real-time evaluation of elevator operation data, a maintenance demand prediction curve for the elevator can be generated based on the elevator health assessment model to achieve real-time prediction of elevator maintenance demand. The curve reflects the future trend of changes in the elevator health score and triggers maintenance instructions when the score reaches the maintenance priority index threshold.

[0090] S5: Generate maintenance strategy.

[0091] In this step, a targeted maintenance strategy is generated based on the elevator health assessment results and maintenance priority assessment results obtained in S4. The maintenance strategy includes: maintenance personnel allocation, maintenance time allocation, etc.

[0092] In this embodiment, for normal elevators, two inspection and maintenance personnel will be dispatched every month for maintenance; for elevators requiring attention and with low maintenance priority, one inspection and maintenance personnel and one senior maintenance specialist will be dispatched to inspect them every two weeks; for elevators with warnings and with medium maintenance priority, two senior maintenance specialists will be dispatched to complete the inspection of the elevator within one week; for elevators with faults and with high maintenance priority, three senior maintenance specialists will be dispatched to complete the inspection of the elevator within 24 hours.

[0093] S6: Maintenance strategy optimization.

[0094] Maintenance Resource Optimization: Considering maintenance resource constraints, we optimize the maintenance strategies generated by S5. We adjust the scheduling and priority of maintenance strategies based on factors such as the availability and cost of maintenance personnel, tools, and equipment to maximize maintenance efficiency and resource utilization.

[0095] Maintenance Feedback and Optimization: Collect feedback from the maintenance process, including maintenance records, fault repair status, and maintenance costs. Analyze this feedback, evaluate maintenance effectiveness, and identify potential issues and areas for improvement. Based on this feedback, optimize and adjust maintenance strategies to improve maintenance quality and efficiency.

[0096] Continuous Improvement: Regularly evaluate the effectiveness and feasibility of maintenance strategies and make continuous improvements based on actual conditions. Based on data such as elevator health assessments, maintenance execution, and maintenance feedback, we continuously optimize maintenance strategies and processes to adapt to changing needs and environments.

[0097] Specifically, in this step, continuous optimization and adjustment of maintenance strategies based on actual conditions and needs ensures consistently accurate maintenance demand forecasts. Based on the passage of time and actual data feedback from maintenance execution, adjustments can be made to data collection frequency, feature extraction methods, and data analysis algorithms to optimize maintenance strategies and processes. Alternatively, optimization algorithms (such as genetic algorithms and simulated annealing) can be employed to optimize and adjust maintenance strategies to achieve the optimal allocation of maintenance resources and optimize maintenance outcomes.

[0098] This example further explains the maintenance strategy based on a genetic algorithm. The specific steps are as follows:

[0099] (7.1) Modeling of maintenance optimization problem:

[0100] Determine optimization objectives and constraints. Optimization objectives include minimizing maintenance costs and ensuring timely elevator maintenance. Constraints include ensuring the spare parts inventory meets maintenance needs and ensuring each elevator is maintained within a specified period.

[0101] The optimization objective can be modeled as follows.

[0102] ① Considering the optimization goal of minimizing maintenance cost, the maintenance cost C is the sum of the maintenance costs of all maintenance personnel for all elevators. The optimization objective function can be expressed as:

[0103]

[0104] Where c i,j The labor cost of maintenance personnel i serving elevator j; t i,j is the transportation cost of maintenance personnel i serving elevator j; pi,j The cost of spare parts when maintenance worker i services elevator j.

[0105] ② Considering the optimization goal of ensuring the timeliness of elevator maintenance, the elevator maintenance timeliness coefficient can be expressed as:

[0106]

[0107] Where Reward is the sum of the incentives for the incentive algorithm to prioritize the allocation of maintenance resources to high-priority elevators, and Penalty is the sum of the penalties for failing to maintain high-priority elevators in a timely manner.

[0108] Specifically, for elevator j:

[0109]

[0110] Where R j is the priority coefficient of elevator j maintenance, P is the penalty coefficient, which is used to adjust the penalty intensity; threshold is the priority coefficient threshold for elevator maintenance to be judged as high priority, that is, for elevator maintenance that meets R j Faults with a value greater than the threshold are considered high priority.

[0111]

[0112] Where α is the reward coefficient, which is used to indicate the reward intensity for handling high-priority maintenance issues.

[0113] All optimization objective functions are integrated according to the weights to obtain the fitness function of elevator maintenance optimization:

[0114] Fitness=w1*C+w2*R

[0115] Where w1 represents the weight coefficient of minimizing maintenance cost to the fitness function, and w2 represents the weight coefficient of maintenance timeliness to the fitness function.

[0116] Constraints can be modeled as follows:

[0117] ①Each elevator must complete maintenance once within the prescribed period:

[0118]

[0119] In the formula, the variable x ij Used to indicate whether elevator j is assigned to maintenance personnel i, x ij =1 means the task is assigned, x ij =0 means no task is assigned.

[0120] ② The spare parts library meets maintenance needs:

[0121]

[0122] Where q ij are the spare parts required by maintenance personnel i to maintain elevator j, and S is the total amount of spare parts in the spare parts warehouse.

[0123] (7.2) Randomly generate N elevator maintenance assignment tasks as initial solutions, and use the existing maintenance task assignment party as one of the initial solutions, usually N∈[20,100]. After generating the initial solution, check whether the initial solution meets the constraint requirements. If not, discard the initial solution and generate a new initial solution. Finally, all the initial solutions constitute the initial population.

[0124] (7.3) The fitness of all initial solutions is calculated according to the fitness function, and the top 10% of the initial solutions with the highest fitness are retained to the next generation population; the remaining 90% of the initial solutions are assigned the probability of being selected as the parent according to their fitness. The higher the fitness, the higher the probability of being selected as the parent.

[0125] Randomly select two parents from the initial solution with assigned probabilities to perform crossover and random mutation to generate offspring, and repeat this operation until 0.9*N offspring are generated. The generated offspring and the top 10% initial solutions with the highest fitness form a new population.

[0126] (7.4) Repeat step (7.3) until the highest comfort value in the population approaches 0, then terminate the loop and output the maintenance allocation plan with the highest comfort value as the optimal solution.

[0127] S8: Visualization of maintenance strategy execution and monitoring.

[0128] In this step, the generated maintenance plan is implemented in actual operations, and the elevator operating status, maintenance plan, and execution status are monitored and displayed through visualization technology. The maintenance plan is implemented in actual operations, and maintenance execution and monitoring are carried out.

[0129] Use data visualization, dashboards, reports and other technologies to visualize and monitor elevator operating status, maintenance plans and execution status, and assist in real-time monitoring of maintenance status.

[0130] Those skilled in the art will understand that the foregoing descriptions are merely preferred embodiments of the invention and are not intended to limit the invention. Although the invention has been described in detail with reference to the foregoing examples, those skilled in the art will still be able to modify the technical solutions described in the foregoing examples or substitute equivalents for some of the technical features therein. Any modifications, equivalent substitutions, etc. made within the spirit and principles of the invention shall be included within the scope of protection of the invention.

Claims

1. An elevator on-demand maintenance method based on dynamic prediction, characterized in that: The following steps are involved: Step 1: Real-time acquisition and preprocessing of elevator maintenance data. Key features of elevator maintenance are extracted from the data, and weights are assigned to the importance of elevator maintenance based on the features. Maintenance data includes operating status data and usage data. Operating status data includes elevator operating time, speed, acceleration, and temperature. Usage data includes elevator operating frequency and peak hours. Step 2: Extract statistical features of key features of elevator maintenance in each maintenance cycle; Based on historical maintenance data, the results of each maintenance project within the same maintenance cycle are weighted and summed to obtain the elevator health score. The elevator health score is then used to classify the elevator health. The classification result is used as the training label for the maintenance cycle. Arrange the statistical features of all maintenance cycles corresponding to the training labels to obtain the training data set; Build and train an elevator health assessment model based on the LSTM algorithm; The elevator health assessment model consists of five layers in sequence: input layer, LSTM layer 1, LSTM layer 2, fully connected layer, and output layer; The input layer is used to accept time series data of multiple features as an input sequence; The LSTM layer 1 includes multiple LSTM units for capturing the basic temporal relationship between features; the number of LSTM units is between 20 and 100; The LSTM layer 2 includes multiple LSTM units for further extracting deeper temporal relationships between features; the number of LSTM units is between 20 and 100; The fully connected layer is used to extract the potential interaction relationship between input features; The output of the fully connected layer is connected to the output layer, which outputs the elevator health score predicted by the model; Step 3: After processing the historical elevator maintenance data and real-time elevator operation data in step 1, the trained elevator health assessment model is input to output the elevator health assessment results, and the maintenance priority assessment results are obtained based on them; Step 4: Generate a maintenance strategy based on the elevator health assessment results and priority assessment results; Step 5: Consider maintenance resource constraints and optimize maintenance strategies based on the availability of maintenance personnel, tools, and equipment, as well as maintenance costs. Step 6: During the actual implementation of the maintenance strategy, monitor the elevator operating status, maintenance plan, and execution; And based on the feedback information, the optimization algorithm is used to optimize the maintenance strategy.

2. The elevator on-demand maintenance method based on dynamic prediction according to claim 1 is characterized in that: In the step 1, the preprocessing includes: data cleaning, data outlier processing, data missing value processing, and classified storage of data.

3. The elevator on-demand maintenance method based on dynamic prediction according to claim 1 is characterized in that: In the second step, statistical feature extraction is performed on the key features of elevator maintenance in each maintenance cycle, specifically: The preprocessed elevator operation data is grouped based on the maintenance cycle, and the mathematical statistical features of the elevator operation data in each maintenance cycle are extracted. For binary data with only two states, the number of times these two states exist is counted in each maintenance cycle; for continuous data with multiple states, mathematical statistical features are calculated in each maintenance cycle, including the calculation of the maximum value, minimum value, average value, and variance; all the obtained mathematical statistical features are spliced ​​together to obtain the statistical feature data of the maintenance cycle as training data.

4. The elevator on-demand maintenance method based on dynamic prediction according to claim 1 is characterized in that: The elevator health is divided into four categories. If the elevator health score is 0-10 points, the corresponding elevator health assessment result is normal and no maintenance is required; If the elevator health score is 11-25 points, the corresponding elevator health assessment result is "needs attention", and maintenance is determined to be required, and the maintenance priority is low; If the elevator health score is 26-35 points, the corresponding elevator health assessment result is that it is determined to require maintenance, and the maintenance priority is medium; If the elevator health score is 36-45 points, the corresponding elevator health assessment result is that it is determined to require maintenance, and the maintenance priority is high.

5. The elevator on-demand maintenance method based on dynamic prediction according to claim 1 is characterized in that: The elevator health assessment model is trained using the training data set, and the elevator health assessment model is optimized using a mean square error loss function and a stochastic gradient descent optimization algorithm during the training process.

6. The elevator on-demand maintenance method based on dynamic prediction according to claim 1, characterized in that: The maintenance strategy includes: maintenance personnel allocation and maintenance time allocation.

7. The elevator on-demand maintenance method based on dynamic prediction according to claim 1, characterized in that: In step 6, the optimization algorithm is a genetic algorithm. The specific steps of using the genetic algorithm to optimize the maintenance strategy are as follows: (1) Modeling of maintenance optimization problem: Determine the optimization objectives and constraints. The optimization objectives include minimizing maintenance costs and ensuring the timeliness of elevator maintenance. The constraints include ensuring that the spare parts inventory meets maintenance needs and that each elevator completes maintenance once within the specified period. The maintenance cost C is the sum of the maintenance costs of all elevators by all maintenance personnel, and is expressed as follows: ； Where, The labor cost of maintenance worker i serving elevator j; The transportation cost for maintenance worker i to serve elevator j; The cost of spare parts when maintenance worker i services elevator j; The timeliness of elevator maintenance is calculated using the following expression: ； ； ； Where Reward is the sum of the incentives for the incentive algorithm to prioritize the allocation of maintenance resources to high-priority elevators, and Penalty is the sum of the penalties for high-priority elevators that fail to be maintained in time; R j is the priority coefficient of elevator j maintenance, P is the penalty coefficient, threshold is the priority coefficient threshold for elevator maintenance to be judged as high priority; α is the reward coefficient; The fitness function of elevator maintenance optimization is obtained, and the expression is as follows: ； Where w1 represents the weight coefficient of minimizing maintenance cost to fitness function, w2 represents the weight coefficient of maintenance timeliness to fitness function; The constraint expression for the spare parts library to meet maintenance requirements is as follows: ； Where, x ij Indicates whether elevator j has been assigned to maintenance personnel i, x ij =1 means the task has been assigned, x ij =0 means no task is assigned; q ij are the spare parts required by maintenance personnel i to perform maintenance on elevator j, and S is the total amount of spare parts in the warehouse; The constraint expression for each elevator to complete maintenance within the specified period is as follows: ； (2) Randomly generate N elevator maintenance assignment tasks as the initial solution, and use the existing maintenance task assignment as one of the initial solutions. After generating the initial solution, check whether the initial solution meets the constraint requirements. If not, discard the initial solution and generate a new initial solution. All initial solutions constitute the initial population. (3) Calculate the fitness of all initial solutions according to the fitness function, and retain the top 10% of the initial solutions with the highest fitness to the next generation population; assign the probability of the remaining 90% of the initial solutions being selected as parents according to their fitness, and the higher the fitness, the higher the probability of being selected as a parent; Randomly select two parents from the initial solution with assigned probabilities, perform crossover and random mutation to generate offspring, and repeat this operation until 0.9*N offspring are generated. The generated offspring and the top 10% initial solutions with the highest fitness form a new population; (4) Repeat step (3) until the highest comfort value in the population approaches 0, then the loop is terminated and the maintenance allocation plan with the highest comfort value is output as the optimal solution.

Images