A BIM-based rail operation and maintenance path optimization method

Abstract

The present invention provides a BIM-based track operation and maintenance path optimization method, which belongs to the field of track data processing technology, including: based on the building information model, constructing a digital twin model of track equipment, and establishing an operation and maintenance space topology network by comprehensively collecting and finely preprocessing the three-dimensional coordinates, real-time status, historical data, maintenance requirements and other information of the equipment. Innovatively use deep learning technology to predict equipment operation and maintenance time, and combine with multi-objective optimization models, use improved genetic algorithms and particle swarm optimization algorithms to optimize path plans from multiple dimensions. The algorithm not only pursues minimization of operation and maintenance costs and maximization of equipment reliability, but also takes into account the workload balance of operation and maintenance personnel, and accurately generates the optimal operation and maintenance path plan through complex matrix operations and intelligent algorithms. The present invention solves the problem that the prior art lacks a method that can optimize the operation and maintenance path according to track operation and maintenance needs.

Description

Technical Field

[0001] The present invention belongs to the technical field of track data processing, and in particular, relates to a track operation and maintenance path optimization method based on BIM. Background Art

[0002] Traditional rail transit equipment maintenance mainly relies on manual inspections and fixed-cycle maintenance. Operation and maintenance personnel need to inspect and maintain various types of equipment distributed in different stations and areas according to preset inspection routes. This approach has the following prominent problems: First, maintenance route planning lacks scientific basis and is often arranged based on the experience of operation and maintenance personnel, making it difficult to achieve route optimization; second, it fails to fully consider equipment status differences and maintenance priorities, resulting in unreasonable allocation of maintenance resources; third, equipment maintenance time prediction is inaccurate, resulting in waste of human resources and increased risk of equipment failure.

[0003] In recent years, although some rail transit operators have begun to introduce intelligent operation and maintenance management systems, existing technical solutions still have obvious shortcomings. For example, the operation and maintenance path planning method based on the geographic information system (GIS) only considers the spatial position relationship of the equipment, ignoring important constraints such as equipment status and maintenance sequence. The maintenance decision-making method based on the expert system, although it introduces an equipment status evaluation mechanism, its decision-making rules are too rigid and difficult to adapt to the complex and changeable actual operation and maintenance scenarios. The path optimization algorithm based on operations research, although it can achieve path optimization to a certain extent, has high computational complexity and is difficult to meet the real-time optimization needs of large-scale rail transit networks.

[0004] In addition, existing technologies generally have data silos, with equipment status information, maintenance records, fault data, etc. scattered in different systems, making it difficult to effectively integrate and utilize data. This results in a lack of comprehensive data support for equipment maintenance decisions, affecting maintenance efficiency and quality. At the same time, due to the lack of an effective time prediction model, existing technologies are difficult to accurately estimate the time required for equipment maintenance, which in turn affects the formulation and execution of the overall maintenance plan.

[0005] Faced with the growing demand for rail transit operation and maintenance, there is an urgent need for a method that can realize the intelligent operation and maintenance path based on rail operation and maintenance needs.

[0006] A new operation and maintenance method that comprehensively considers multi-dimensional factors such as space constraints, time constraints, equipment status, etc., and can achieve intelligent path optimization. Summary of the invention

[0007] In view of this, the present invention provides a BIM-based track operation and maintenance path optimization method, which can solve the problem that the prior art lacks a method that can optimize the operation and maintenance path according to the track operation and maintenance requirements.

[0008] The present invention is achieved in that: The present invention provides a BIM-based track operation and maintenance path optimization method, comprising the following steps: S10. Use the building information model to build a digital twin model of rail equipment, collect the three-dimensional position coordinate information, real-time equipment status information, historical status data, maintenance requirement information, fault history records, and maintenance records of rail equipment, pre-process and clean the collected data, remove abnormal data and supplement missing data, so as to achieve comprehensive acquisition of rail equipment information; S20, establishing a spatial distance matrix based on the three-dimensional position coordinates of the rail equipment, constructing a constraint matrix considering the maintenance order constraint relationship between the equipment, generating a priority weight matrix in combination with the equipment maintenance priority index, and fusing the spatial distance matrix, the constraint matrix, and the priority weight matrix to construct a complete rail operation and maintenance spatial topology network; S30, perform multivariate mixed decomposition calculation on historical operation and maintenance data, extract stable components and fluctuating components of operation and maintenance time series data, establish a multidimensional correlation model of equipment state parameters, environmental influencing factors and operation and maintenance duration, use deep learning methods to train the operation and maintenance time prediction model, verify the accuracy of the operation and maintenance time prediction model through cross-validation methods, and obtain deep learning prediction deviation and historical state volatility; S40, calculating the standard operation and maintenance duration of each device based on the operation and maintenance time prediction model, establishing a time influence coefficient matrix in combination with the fluctuation component, obtaining a time error compensation amount through matrix operation, and superimposing the standard operation and maintenance duration with the time error compensation amount to obtain a corrected device operation and maintenance time matrix; S50, taking minimization of total operation and maintenance cost, maximization of equipment reliability, and balancing of workload of operation and maintenance personnel as optimization objectives, establishing a weight allocation mechanism among the optimization objectives, taking the track operation and maintenance spatial topology network and the equipment operation and maintenance time matrix as constraints, and constructing a multi-objective optimization model; S60. Design a population initialization strategy based on operation and maintenance experience, construct an adaptive crossover operator and a mutation operator, introduce an elite retention mechanism, use an improved genetic algorithm to solve the multi-objective optimization model, record the convergence speed of the genetic algorithm and the number of iterative optimizations, and obtain an optimized operation and maintenance path plan through multi-generation iterative optimization; S70, performing local search on the optimized operation and maintenance path scheme based on the particle swarm optimization algorithm, obtaining the particle swarm optimal position vector, local search iteration step, and local search fitness value, focusing on optimizing the transfer distance between devices and the equipment maintenance sequence, and generating multiple groups of candidate operation and maintenance path schemes; S80. According to the objectives of minimizing operation and maintenance costs and maximizing equipment reliability, the candidate operation and maintenance path solutions are evaluated using an operation and maintenance path evaluation equation group to obtain a final operation and maintenance path solution.

[0009] Furthermore, the operation and maintenance path evaluation equation group includes a path optimization evaluation equation, a state prediction evaluation equation, a time response evaluation equation, and a comprehensive score equation; The input of the path optimization evaluation equation includes the optimal position vector of the particle swarm, the spatial distance in the track operation and maintenance space topological network, the constraint satisfaction index, and the local search fitness value, and outputs the path optimization score; The state prediction evaluation equation inputs include the deep learning prediction deviation, the historical state volatility, the operation and maintenance time correction coefficient in the time impact coefficient matrix, and the equipment reliability index, and outputs a state prediction score; The time response evaluation equation input includes the standard operation and maintenance time, the number of iterative optimization times, the genetic algorithm convergence speed, and the local search iteration step, and outputs a time response score; The comprehensive score equation input includes the path optimization score, the state prediction score, the weight coefficient of the time response score, and the global optimal solution convergence degree of the particle swarm optimization algorithm, and outputs a comprehensive score, and selects the operation and maintenance path plan with the highest comprehensive score as the final operation and maintenance path plan.

[0010] Further, the mathematical model involved in the present invention is described in detail as follows: 1. Data preprocessing in S10: The data cleaning and outlier processing equations are specifically expressed as follows: ; In the formula, For cleaned data; is the original data; is the sample mean; is the sample standard deviation; threshold 3 is an empirical value and can be adjusted according to actual conditions.

[0011] The missing value filling equation is specifically expressed as follows: ; In the formula, is the fill value; The data before the missing value; The data is the moment after the missing value; is the historical average of this type of data; is the weight coefficient and satisfies ; is a random disturbance term, which follows a normal distribution .

[0012] 2. Matrix construction in S20: The spatial distance matrix is ​​specifically expressed as follows: ; ; In the formula, For equipment To device spatial distance; For equipment The three-dimensional coordinates of is the total number of devices.

[0013] The constraint matrix is ​​specifically expressed as follows: ; .

[0014] The priority weight matrix is ​​specifically expressed as follows: ; ; In the formula, For equipment Failure frequency index; For equipment Reliability indicators; For equipment Maintenance time urgency indicator; For equipment Importance indicators; is the weight coefficient and satisfies .

[0015] The fusion matrix is ​​specifically expressed as follows: ; In the formula, is the topological network matrix after fusion; is the fusion weight and satisfies ; is a random disturbance term, which follows a normal distribution .

[0016] 3. Timing decomposition in S30: The specific expression of multivariate mixed decomposition calculation is as follows: ; In the formula, is the original time series data; is the trend item; is the seasonal item; is a random item.

[0017] The operation and maintenance duration prediction model is specifically expressed as follows: ; ; In the formula, is the predicted operation and maintenance duration; is the input feature vector; are model parameters; is the prediction error; For the The weights of the hidden layer nodes; For the The output of hidden layer nodes; is the bias term; is the number of hidden layer nodes.

[0018] 4. Construct the time matrix for S40: The time impact coefficient matrix is ​​specifically expressed as follows: ; In the formula, is the time influence coefficient; is the fluctuation rate of equipment status; Environmental impact factors; Efficiency factor for historical maintenance; is the weight coefficient and satisfies .

[0019] The time error compensation amount is specifically expressed as follows: ; In the formula, is the time error compensation; It is the standard operation and maintenance time; is the fluctuation coefficient; is the current time; The fluctuation period.

[0020] The corrected time matrix is ​​specifically expressed as follows: ; In the formula, It is the final operation and maintenance time matrix; is a random disturbance term, which follows a normal distribution .

[0021] 5. Multi-objective optimization model for S50: The objective function is specifically expressed as follows: ; ; ; ; In the formula, is the total operation and maintenance cost; It is a negative value of equipment reliability; is the workload imbalance; is the target weight coefficient; For slave devices To device the transfer costs; For maintenance costs; is a 0-1 decision variable; For maintenance time; is the importance of the equipment; is the failure rate; For maintenance time; For maintenance personnel About equipment Maintenance time; is the total time; The number of maintenance personnel.

[0022] 6. Evaluation equations for S90: The path optimization evaluation equation is specifically expressed as follows: ; In the formula, Scoring path optimization; is the particle swarm position convergence rate; is the spatial distance; To constrain satisfaction indicators; is the local search fitness value; is the weight coefficient; is the attenuation coefficient, ranging from 0.1 to 0.5.

[0023] Construction principle: The first item adopts the sum of squares of position convergence rates to indicate the stability of path convergence; the second item adopts a negative exponential function to reflect the nonlinear attenuation effect of spatial distance on the score; the third item adopts the form of a reliability function to reflect the degree of constraint satisfaction; the fourth item adopts the square root form to reflect the marginal effect of local search.

[0024] The state prediction evaluation equation is specifically expressed as follows: ; In the formula, Score the state predictions; Predict the time-varying function of the bias for deep learning; is the maximum allowable deviation; is the historical state volatility; is the time correction factor; It is the equipment reliability index; is the weight coefficient; is the volatility influence coefficient, ranging from 0.5 to 1.0; The time period for evaluation.

[0025] Construction principle: The first item adopts the integral form to consider the cumulative impact of forecast deviations over the entire time period; the second item adopts the negative exponential function to reflect the attenuation effect of historical fluctuations on forecast accuracy; the third item adopts the sensitivity of the correction coefficient to reliability to reflect the stability of state forecasting; the fourth item adopts the logarithmic function to reflect the marginal diminishing effect of improved reliability.

[0026] The time response evaluation equation is specifically expressed as follows: ; In the formula, score the time response; It is the standard operation and maintenance time; is the maximum allowed duration; is the number of iterations for optimization; is the convergence speed of the genetic algorithm; is the reference convergence speed; It is the local search iteration step; is the maximum step length; is the weight coefficient; It is the time response sensitivity index, ranging from 1.5 to 2.5.

[0027] Construction principle: The first item adopts the form of power function to reflect the nonlinear characteristics of time response; the second item adopts the partial derivative of the number of iterations to the convergence speed to reflect the efficiency of the optimization process; the third item adopts the form of Gaussian function to describe the optimal interval characteristics of the convergence speed; the fourth item adopts the form of cube root to reflect the smoothness characteristics of step size adjustment.

[0028] The comprehensive scoring equation is specifically expressed as follows: ; In the formula, For comprehensive rating; Scoring the sensitivity of paths to the optimization variables; is the convergence degree of particle swarm algorithm; Rate of change for time scoring; is the weight coefficient; is the convergence influence coefficient, ranging from 0.8 to 1.2; is the optimization variable.

[0029] Construction principle: The first item combines the path score and its sensitivity to the optimization variable to reflect the robustness of the path optimization; the second item uses a negative exponential function to modulate the state score to reflect the impact of convergence on state prediction; the third item combines the time score and the square root of its rate of change to reflect the dynamic characteristics of the time response. The comprehensive scoring equation achieves a comprehensive evaluation of the optimization results through the nonlinear combination of the scores of each sub-item.

[0030] Compared with the prior art, the BIM-based track operation and maintenance path optimization method provided by the present invention has the following beneficial effects: 1. Comprehensively obtain equipment information. Use the BIM platform to establish a digital twin model of rail equipment, and achieve comprehensive collection and unified management of equipment information through three-dimensional coordinate collection, status monitoring, fault recording and other means. Provide a reliable data foundation for subsequent path optimization.

[0031] 2. Optimize path planning. A multi-objective optimization model is used to consider factors such as overall operation and maintenance costs, equipment reliability, and personnel workload. The spatial topological relationship and time constraint relationship between track equipment are constructed, and the improved genetic algorithm and particle swarm algorithm are used to solve the global optimal operation and maintenance path plan. The efficiency and accuracy of path planning are greatly improved.

[0032] 3. Predictive maintenance support. By integrating deep learning and time series analysis technology, a correlation model between equipment status, environmental factors and maintenance duration is established, which can accurately predict the maintenance needs and maintenance duration of equipment. This provides a basis for formulating targeted preventive maintenance strategies.

[0033] In summary, the present invention solves the problem that the prior art lacks a method for optimizing the operation and maintenance path according to the track operation and maintenance requirements. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 A flow chart of the method provided by the present invention; Figure 2 This is a timing decomposition diagram of the temperature data of device B001 in Example 2; Figure 3 This is a performance evaluation diagram of the equipment status prediction model in Example 2. DETAILED DESCRIPTION

[0035] like Figure 1 As shown in FIG. 1 , it is a flow chart of a BIM-based rail operation and maintenance path optimization method provided by the present invention. The specific implementation method of each step is described in detail below: The specific implementation method of step S10 is to first obtain the location information, real-time status information and maintenance requirement information of the rail equipment through three-dimensional coordinate acquisition technology to establish a digital twin model of the equipment. Then, a reasonable data acquisition cycle is designed to record the historical operating status data, fault records and maintenance records of the equipment to form a complete equipment status database.

[0036] In order to ensure the accuracy and completeness of the data, this method uses two preprocessing techniques. The first is outlier cleaning, which uses the 3 times standard deviation threshold method to remove abnormal data points that are beyond the normal range. Then there is missing value filling, which uses the weighted average method of the data before and after the time series to supplement the missing data points. After these preprocessing steps, comprehensive acquisition of rail equipment information can be achieved, laying the foundation for subsequent operation and maintenance path optimization.

[0037] The specific implementation method of step S20 is to first calculate the Euclidean distance between any two devices based on the three-dimensional coordinates of the devices, and establish a distance matrix that describes the spatial relationship of the devices. Secondly, analyze the maintenance order constraint relationship between the devices in combination with the equipment maintenance manual, and construct a three-value constraint matrix. Furthermore, comprehensively consider factors such as the failure frequency, reliability, maintenance time urgency and importance of the equipment, and establish an equipment maintenance priority weight matrix.

[0038] Finally, the linear weighted fusion method is used to fuse the above three matrices to construct a complete track operation and maintenance spatial topology network. This network model not only includes the spatial distance relationship between equipment, but also considers the order constraints and priority weights of equipment maintenance, providing the necessary constraints and evaluation indicators for subsequent path optimization.

[0039] The specific implementation method of step S30 is to first perform multivariate mixed decomposition on the historical operation and maintenance data to extract stable state parameters. Then, a multi-layer deep neural network model is established, and the input layer includes equipment state parameters and environmental influencing factors. The operation and maintenance time prediction model is obtained through training. In order to verify the accuracy of the model, the cross-validation method is used to calculate the prediction deviation and record the volatility of the historical state data.

[0040] The key to this step is to use the powerful modeling capabilities of deep learning to extract the inherent laws between equipment status and environmental factors and operation and maintenance time from a large amount of historical operation and maintenance data. By analyzing the prediction deviation and volatility, the applicability and robustness of the model can be better evaluated, laying the foundation for subsequent time error correction.

[0041] The specific implementation method of step S40 is to first calculate the standard operation and maintenance time of each device based on the aforementioned operation and maintenance time prediction model. Then, the time impact coefficient matrix is ​​established by combining the equipment state volatility, environmental impact factors and historical maintenance efficiency. Next, a periodic fluctuation term is introduced to calculate the compensation amount of the time error. Finally, the standard operation and maintenance time is superimposed with the time error compensation amount to obtain a corrected equipment operation and maintenance time matrix.

[0042] The purpose of this step is to fully consider the impact of equipment status changes and environmental dynamics on maintenance time, and to improve the accuracy of operation and maintenance time prediction by compensating for time errors. The construction of the time impact coefficient matrix incorporates multiple key factors and can better reflect the time dynamic characteristics of the actual operation and maintenance process.

[0043] The specific implementation method of step S50 is to first establish a linear function with minimizing the total operation and maintenance cost as the first objective. Secondly, construct an exponential function with maximizing equipment reliability as the second objective. Furthermore, establish a variance function with balancing the workload of operation and maintenance personnel as the third objective. Finally, the aforementioned constructed track operation and maintenance spatial topology network and equipment operation and maintenance time matrix are used as constraints, and these three objective functions are combined to construct a multi-objective optimization model.

[0044] The key to this step is to take into account multiple factors such as operation and maintenance costs, equipment reliability, and personnel workload through multi-objective optimization. The combination of linear, exponential, and variance functions can better describe the mathematical relationship between these objectives. The topological network and time matrix are used as constraints to ensure that the optimized path plan is feasible and meets the actual operation and maintenance needs.

[0045] The specific implementation method of step S60 is to first design a reasonable population initialization strategy based on historical operation and maintenance experience. Then, construct an adaptive crossover operator based on the operation and maintenance time window and an adaptive mutation operator based on the equipment constraint relationship. Finally, use an improved genetic algorithm to solve the aforementioned multi-objective optimization model, and record the convergence speed and number of iterations of the algorithm.

[0046] The focus of this step is to use genetic algorithms to optimize and solve multi-objective problems. By designing a population initialization strategy and adaptive crossover mutation operator that are suitable for actual operation and maintenance characteristics, the search efficiency and convergence of the algorithm can be effectively improved. At the same time, the convergence speed and number of iterations of the algorithm are recorded to provide a basis for subsequent time response evaluation.

[0047] The specific implementation method of step S70 is to first use the particle swarm algorithm to perform a local search on the aforementioned optimized operation and maintenance path plan. Secondly, record the particle swarm optimal position vector and the step size of the local search. Furthermore, calculate the fitness value of the local search. Finally, focus on optimizing the transfer distance and maintenance sequence between devices to generate multiple sets of candidate operation and maintenance path plans.

[0048] The purpose of this step is to further use the particle swarm algorithm to locally optimize the path based on the global optimization results obtained by the genetic algorithm. By recording indicators such as the optimal position of the particle swarm, search step length, and fitness value, the required data support is provided for subsequent path evaluation. At the same time, focusing on the optimization of equipment transfer distance and maintenance sequence can further improve the practicality of the operation and maintenance path.

[0049] The specific implementation method of step S80 is to first calculate the sum of squares of the particle swarm position convergence rate and establish a path optimization score. Secondly, calculate the integral value of the deep learning prediction deviation and construct a state prediction score. Furthermore, calculate the partial derivative of the number of iterations with respect to the convergence speed and establish a time response score. Finally, a nonlinear combination method is used to perform a weighted synthesis of the above three scoring results, and select the plan with the highest comprehensive score as the final operation and maintenance path plan.

[0050] The core of this step is to build a comprehensive path evaluation system. The path optimization score focuses on path convergence and spatial constraint satisfaction; the state prediction score focuses on prediction accuracy and equipment reliability; the time response score focuses on optimization efficiency and response speed. By comprehensively considering these three evaluation indicators in a nonlinear combination, the comprehensive performance of the final solution can be more accurately reflected.

[0051] Optionally, step S90 is further included. The specific implementation of step S90 is to first write the selected final operation and maintenance path plan into the building information model. Then, a real-time feedback mechanism including path length, maintenance time and equipment status is established. Furthermore, an evaluation system including indicators such as optimization efficiency, prediction accuracy and response time is constructed. Finally, the entire operation and maintenance knowledge base is dynamically updated according to the actual operation and maintenance results.

[0052] The purpose of this step is to apply the optimized operation and maintenance path plan to practice and establish a closed-loop feedback mechanism. The path execution status and equipment status changes are recorded in real time, and the actual effect of the optimization plan is evaluated regularly. By dynamically updating the knowledge base, the algorithm model and evaluation index system can be continuously improved to improve the applicability and reliability of the entire operation and maintenance optimization method.

[0053] Specifically, the principle of the present invention is to establish a digital twin model of rail equipment based on BIM, comprehensively perceive the status information, fault information and maintenance requirements of the equipment, combine historical operation and maintenance data, and use a multi-objective optimization algorithm to plan an efficient and reasonable rail equipment operation and maintenance path. The specific technical principles are as follows: 1. Construction of digital twin model. Using the BIM platform, through 3D laser scanning, RFID tags and other technical means, the 3D coordinate position, real-time status parameters, historical fault records and maintenance requirements of the rail equipment are obtained to establish a digital twin model of the equipment. At the same time, data preprocessing technology is used to clean and fill the collected information to ensure the integrity and accuracy of the data. This provides reliable basic data for subsequent path optimization.

[0054] 2. Spatial topological network construction. According to the three-dimensional coordinate information of the equipment, the spatial distance between any two devices is calculated, and a distance matrix describing the spatial relationship between the devices is constructed. Combined with the analysis of the equipment maintenance manual, a matrix describing the equipment maintenance sequence constraints is established. And the equipment maintenance priority matrix is ​​constructed by integrating indicators such as equipment failure frequency, reliability, maintenance urgency and importance. Finally, these three matrices are weighted fused to construct a complete rail operation and maintenance spatial topological network.

[0055] 3. State prediction model training. Using historical operation and maintenance data, the stable components of equipment state parameters are extracted using the multivariate time series decomposition method. Then a multi-layer deep neural network model is constructed, with the input layer including equipment state parameters and environmental influencing factors, and the output is the predicted maintenance time. The prediction accuracy of the model is evaluated through cross-validation and other methods, and the prediction deviation and historical state volatility are recorded. These indicators will serve as the basis for subsequent state prediction evaluation.

[0056] 4. Analysis of time impact factors. Based on the aforementioned state prediction model, the standard maintenance time of each device is calculated. Combining factors such as equipment state fluctuations, environmental impacts, and historical maintenance efficiency, a time impact coefficient matrix is ​​constructed. Introduce periodic fluctuation terms, calculate the compensation amount of time error, and superimpose it on the standard time to obtain the corrected equipment maintenance time matrix. This step aims to more accurately characterize the dynamic characteristics of equipment maintenance time.

[0057] 5. Multi-objective optimization modeling. With the minimization of total operation and maintenance costs, maximization of equipment reliability, and balance of personnel workload as the three optimization goals, a multi-objective optimization model is constructed. Among them, the spatial topology network and time matrix constructed above are used as constraints. This forms a complex multi-objective optimization problem.

[0058] 6. Algorithm solution and evaluation. For the above multi-objective optimization problem, a population initialization strategy based on operation and maintenance experience is designed, an adaptive genetic algorithm crossover mutation operator is constructed, and an improved genetic algorithm is used for solution. At the same time, a particle swarm algorithm is introduced for local optimization to further improve the quality of the solution. In order to comprehensively evaluate the optimization results, the present invention constructs a comprehensive evaluation system including path optimization scoring, state prediction scoring and time response scoring, and obtains the final operation and maintenance path solution through nonlinear combination.

[0059] A specific embodiment 1 of the present invention is provided below. The steps in this embodiment 1 are described in detail as follows: The specific implementation of step S10 is as follows: First, the location information, real-time status information and maintenance requirement information of the rail equipment are obtained through three-dimensional coordinate acquisition technology, and a digital twin model of the equipment is established. It can be expressed as: ;in, For equipment The three-dimensional coordinate position information, For equipment Real-time status information, For equipment maintenance requirements information.

[0060] Then, design a reasonable data collection cycle , during this period, the historical operating status data, fault records and maintenance records of the equipment are recorded to form a complete equipment status database .

[0061] To ensure the accuracy and completeness of the data, this method uses two preprocessing techniques.

[0062] The first step is to clean the outliers by using the 3 times standard deviation threshold method to clean the original data. Cleaning is performed to remove abnormal data points that are beyond the normal range, which can be expressed as: ; in, and are the mean and standard deviation of the samples respectively, and 3 is the empirical threshold.

[0063] The second is missing value filling, which uses the weighted average method of the data before and after the time series and the historical average to supplement the missing data points, which can be expressed as: ;in, , and is the weight coefficient, satisfying ; is a random disturbance term, which follows a normal distribution .

[0064] Finally, establish the equipment information database , the cleaned and filled data are classified and stored according to device number, timestamp and data type.

[0065] The specific implementation of step S20 is as follows: First, calculate the Euclidean distance between any two devices based on the device's three-dimensional coordinates , establish a distance matrix describing the spatial relationship of devices : ; ; Secondly, according to the equipment maintenance manual, the maintenance sequence constraint relationship between equipment is analyzed and a three-value constraint matrix is ​​constructed. : ; ; Furthermore, based on the frequency of equipment failure ,reliability , Time urgency and importance Factors such as the following are used to establish the equipment maintenance priority weight matrix : ; ; in, , , and is the corresponding weight coefficient, satisfying .

[0066] Finally, the above three matrices are fused using the linear weighted fusion method to construct a complete rail operation and maintenance space topology network matrix. : ; in, , and is the fusion weight coefficient, satisfying ; is a random disturbance term, which follows a normal distribution .

[0067] The specific implementation of step S30 is as follows: First, historical operation and maintenance data Perform multivariate mixed decomposition and extract stable state parameters: ;in, For trend items, For seasonal items, is a random item.

[0068] Then, a multi-layer deep neural network model is established, and the input layer includes the device state parameters and environmental factors , the operation and maintenance time prediction model is obtained through training : ; ;in, is the prediction error; and are the weights and bias terms of the neural network respectively; For the The output of hidden layer nodes; is the number of hidden layer nodes.

[0069] In order to verify the accuracy of the model, the cross-validation method was used to calculate the prediction deviation. , and record the volatility of historical status data : ; ;in, is the actual observed value, is the sample mean, is the observation time length.

[0070] The specific implementation of step S40 is as follows: First, based on the aforementioned operation and maintenance time prediction model Calculate the standard operation and maintenance time of each device .

[0071] Then, combined with the equipment status fluctuation rate 、Environmental factors and historical maintenance efficiency , establish the time influence coefficient matrix : ;in, , and is the corresponding weight coefficient, satisfying .

[0072] Next, we introduce the periodic fluctuation term and calculate the compensation amount of the time error : ;in, is the fluctuation coefficient, is the current time, The fluctuation period.

[0073] Finally, the standard operation and maintenance time Time error compensation Superimpose to obtain the corrected equipment operation and maintenance time matrix : ;in, is a random disturbance term, which follows a normal distribution .

[0074] The specific implementation of step S50 is as follows: First, establish a total operation and maintenance cost Minimize a linear function as the first optimization objective: ;in, For slave devices To device The transfer cost, For maintenance costs, and are 0-1 decision variables and maintenance time respectively.

[0075] Secondly, build on equipment reliability Maximize an exponential function as the second optimization objective: ;in, For equipment The importance of For equipment The failure rate, For equipment maintenance time.

[0076] Furthermore, establish a workload balance system for operation and maintenance personnel Variance function for the third optimization objective: ;in, For maintenance personnel About equipment Maintenance time, is the total time, The number of maintenance personnel.

[0077] Finally, the track operation and maintenance space topology network constructed above is and equipment operation and maintenance time matrix As constraints, these three objective functions are combined to construct a multi-objective optimization model: ;in, , and is the weight coefficient of the corresponding objective function.

[0078] The specific implementation of step S60 is as follows: First, design a reasonable population initialization strategy based on historical operation and maintenance experience.

[0079] Then, construct an adaptive crossover operator based on the operation and maintenance time window: ;in, and is the decision variable of the parent individual, is the crossover probability, and are the current number of iterations and the maximum number of iterations respectively, is a random number function.

[0080] Furthermore, an adaptive mutation operator based on device constraint relationships is constructed: ;in, is the decision variable of the parent individual, To change the asynchronous length, is the mutation probability, and are the current constraint satisfaction and the maximum constraint satisfaction respectively.

[0081] Finally, the improved genetic algorithm is used to solve the aforementioned multi-objective optimization model, and the convergence speed of the algorithm is recorded. and the number of iterations .

[0082] The specific implementation of step S70 is as follows: First, the particle swarm algorithm is used to perform local search on the aforementioned optimized operation and maintenance path solution, and the optimal position vector of the particle swarm is recorded. and the step size of the local search .

[0083] Then, calculate the fitness value of the local search : ;in, For particles The position convergence rate.

[0084] Finally, focus on optimizing the transfer distance between devices and maintenance timing , generate multiple sets of candidate operation and maintenance path solutions.

[0085] The specific implementation of step S80 is as follows: First, calculate the square sum of the particle swarm position convergence rate and establish the path optimization score : ;in, , , and is the weight coefficient, and is the attenuation coefficient.

[0086] Second, calculate the deep learning prediction bias The integral value of constructing the state prediction score : ; in, , , and is the weight coefficient, is the volatility influence coefficient, is the maximum allowable deviation, It is an indicator of equipment reliability.

[0087] Furthermore, calculate the number of iterations Convergence speed The partial derivative of , establishes the time response score : ;in, , , and is the weight coefficient, is the time response sensitivity index, is the maximum allowed duration, is the reference convergence speed, is the maximum step length.

[0088] Finally, a nonlinear combination method was used to weight the above three scoring results and select the comprehensive score. The highest solution is used as the final operation and maintenance path solution: ; in, , and is the weight coefficient, is the convergence degree of the particle swarm algorithm, To optimize the variables, is the convergence influence coefficient.

[0089] The specific implementation of step S90 is as follows: First, the selected final operation and maintenance path plan is written into the building information model.

[0090] Then, a real-time feedback mechanism including path length, maintenance time and equipment status is established.

[0091] Furthermore, an evaluation system is constructed that includes indicators such as optimization efficiency, prediction accuracy, and response time.

[0092] Finally, the entire operation and maintenance knowledge base is dynamically updated based on the actual operation and maintenance results.

[0093] In order to better understand and implement the present invention, the following is an example 2 of a specific application scenario of the present invention: The rail transit network of a certain city has covered the main areas of the city, with a total of 5 subway lines and a total of 150 stations. In recent years, with the continuous extension of rail transit lines and the continuous increase in the number of stations, the management department is facing increasing pressure on operation and maintenance. The traditional manual inspection and maintenance mode has been difficult to meet actual needs, and there are problems such as low efficiency and poor reliability. For this reason, the management department decided to adopt the BIM-based rail operation and maintenance path optimization method proposed in the present invention, in order to improve the level of intelligent management of rail equipment.

[0094] In the first step, the management department set up a technical research team to be responsible for the specific implementation of the method of the present invention in rail transit operation and maintenance management. The team first set out to establish a digital twin model of rail equipment.

[0095] Through 3D laser scanning technology, the team conducted comprehensive 3D coordinate measurement and collection of the main equipment in 150 rail transit stations in the city, including 18 traction substations, 45 signal equipment and 87 electromechanical equipment. Combined with RFID technology, the team also obtained the real-time operating status parameters of each device, such as temperature, current, vibration, etc. At the same time, combined with the equipment maintenance manual and historical fault records, the team established an equipment maintenance demand information database, including indicators such as the failure frequency, reliability level, maintenance urgency and importance of each device.

[0096] To ensure the accuracy of the data, the team designed a reasonable data collection cycle. For key equipment such as substation equipment and signal equipment, the collection cycle is 1 hour; for electromechanical equipment, the collection cycle is 3 hours. During the one-year data collection process, the operating status data, fault history records and maintenance records of 18 substation equipment, 45 signal equipment and 87 electromechanical equipment in 150 stations were recorded.

[0097] After completing the data collection, the team preprocessed the raw data. First, the 3 times standard deviation threshold method was used to clean the abnormal data. The results are shown in Table 1.

[0098] Table 1 Outlier cleaning results

[0099] As can be seen from Table 1, after cleaning, the proportion of valid data reached 95.3%, laying the foundation for subsequent status analysis and prediction.

[0100] For the few missing data points that still exist after cleaning, the team used the weighted average method of the data before and after the time series to fill them in. Taking a certain status parameter of equipment B001 as an example, assuming time, Moment and The data at each time are 50, X, and 60, respectively. The data X at the moment is missing. According to the formula ,in To obey the normal distribution The random disturbance term is calculated and the filling value X=54.2 is obtained. The results of the entire missing value filling process are shown in Table 2.

[0101] Table 2 Missing value filling results

[0102] It can be seen that after the filling process, the overall effective data ratio increased to 98.1%, laying a good data foundation for the subsequent operation and maintenance path optimization.

[0103] With a comprehensive equipment digital twin model, the team then proceeded to build a track operation and maintenance spatial topology network. First, the Euclidean distance between any two devices is calculated based on the three-dimensional coordinate position of the equipment to form the inter-equipment distance matrix D shown in Table 3.

[0104] Table 3. Distance matrix D between devices

[0105] Secondly, the team analyzed the maintenance sequence constraints between equipment based on the equipment maintenance manual and constructed the constraint matrix C shown in Table 4.

[0106] Table 4 Equipment maintenance sequence constraint matrix C

[0107] In the table, 1 means that device B001 must be maintained before device B002, 0 means there is no order constraint between the two devices, and -1 means that device B002 must be maintained before device B150.

[0108] Furthermore, the team comprehensively considered factors such as equipment failure frequency, reliability level, maintenance urgency and importance, and established the equipment maintenance priority weight matrix W as shown in Table 5.

[0109] Table 5 Equipment maintenance priority weight matrix W

[0110] Finally, the team used a linear weighted method to fuse the above three matrices and obtained the track operation and maintenance space topology network matrix M as shown in Table 6.

[0111] Table 6 Track operation and maintenance spatial topology network matrix M

[0112] Through the above steps, the team successfully built a digital twin model of rail equipment and established a topological network matrix that describes the spatial relationship of equipment, maintenance order constraints, and priority weights. This laid the necessary foundation for subsequent operation and maintenance path optimization.

[0113] Next, the team began to analyze and model historical operation and maintenance data.

[0114] First, the team performed multivariate time series decomposition on the historical operation and maintenance data and extracted stable state parameters. Taking the temperature collection data of a certain substation B001 as an example, Figure 2As shown in the figure, the trend term, seasonal term and random term of the device temperature can be obtained through decomposition. It can be seen that the temperature of device B001 has obvious periodic fluctuation characteristics, which is mainly affected by the ambient temperature. The random term reflects the random fluctuation of the device's own state. This information lays the foundation for the subsequent time influencing factor analysis.

[0115] Based on the above data decomposition results, the team built a multi-layer deep neural network model to predict the maintenance time of equipment. The input layer of the model includes equipment status parameters and environmental influencing factors, and the output is the predicted maintenance time. Through training on historical data, the team obtained the parameters of the prediction model and calculated the prediction deviation of the model through cross-validation method. The results are shown in Table 7.

[0116] Table 7 Prediction deviation of maintenance time prediction model

[0117] As can be seen from Table 7, the average prediction deviation of the prediction model is 7.3%, and the maximum prediction deviation is 14.0%. The prediction accuracy is high and can provide support for the subsequent time influencing factor analysis.

[0118] like Figure 3 As shown in the figure, the performance indicators of the prediction models of different types of equipment (substation equipment, signal equipment, electromechanical equipment) are displayed using bar charts, including prediction accuracy, R² score, and standardized RMSE. These indicators can be used to visually compare the performance differences of prediction models of different types of equipment. At the same time, the team also recorded the volatility of historical status data, as shown in Table 8.

[0119] Table 8 Fluctuation rate of equipment historical status data

[0120] Based on the above prediction model and state fluctuation analysis, the team constructed an influencing factor matrix for equipment maintenance time. Specifically, the standard maintenance time of each device was calculated first, and then the factors such as equipment state fluctuation, environmental impact and historical maintenance efficiency were comprehensively considered to obtain the time influence coefficient matrix K shown in Table 9.

[0121] Table 9 Equipment maintenance time influence coefficient matrix K

[0122] In the table, a coefficient greater than 1 indicates that the equipment maintenance time will be increased, and vice versa. For example, the maintenance time of equipment B001 will increase by 24% compared to the standard time, while the maintenance time of equipment B002 will increase by 19%. These coefficients reflect the impact of dynamic changes in equipment status on maintenance time.

[0123] With the above analysis results, the team began to build a multi-objective optimization model. First, with the minimization of total operation and maintenance costs as the first goal, the following linear function was established: ;in, For slave devices To device The transfer cost, For equipment maintenance costs, and are 0-1 decision variables and maintenance time respectively.

[0124] Secondly, taking the maximization of equipment reliability as the second goal, the following exponential function is constructed: ;in, For equipment The importance of For equipment The failure rate, For equipment maintenance time.

[0125] Furthermore, taking the workload balance of operation and maintenance personnel as the third goal, the following variance function is established: ;in, For maintenance personnel About equipment Maintenance time, is the total time, The number of maintenance personnel.

[0126] Finally, the track operation and maintenance space topology network matrix constructed above is and equipment maintenance time matrix As constraints, combining these three objective functions, the team constructed the following multi-objective optimization model: ; For this multi-objective optimization problem, the team used an improved genetic algorithm to solve it. First, a reasonable population initialization strategy was designed based on the experience of the operation and maintenance personnel. Then, an adaptive crossover operator based on the maintenance time window and an adaptive mutation operator based on the constraint relationship were constructed to improve the algorithm's search efficiency. Finally, during the algorithm's iterative solution process, the team recorded the algorithm's convergence speed and number of iterations.

[0127] At the same time, the team also introduced the particle swarm algorithm to conduct local search on the optimization results, focusing on optimizing the transfer distance and maintenance sequence between devices. By recording indicators such as the particle swarm's optimal position vector, local search step length, and fitness value, data support was provided for subsequent comprehensive evaluation.

[0128] After completing the optimization calculations, the team built a comprehensive evaluation system, including path optimization scoring, state prediction scoring, and time response scoring. Specifically: Path optimization score Taking into account factors such as particle swarm position convergence rate, inter-device transfer distance, constraint satisfaction and local search fitness, the expression is as follows: ; Status prediction score Taking into account factors such as deep learning prediction deviation, historical state volatility, time correction coefficient and equipment reliability, the expression is as follows: ; Time Response Scoring Considering factors such as standard maintenance time, genetic algorithm convergence speed, number of iterations and local search step length, the expression is as follows: ; Finally, the above three scores are weighted and integrated by nonlinear combination method to obtain the final comprehensive score. : ; After comprehensive evaluation, the team finally determined a set of optimized rail operation and maintenance path plans and fed them back into the BIM system to form a closed-loop management.

[0129] In the actual application process, the management department has established a feedback mechanism for real-time monitoring and regular evaluation. Key indicators such as operation and maintenance path length, maintenance time and equipment status are recorded daily, and the actual effect of the optimization plan is evaluated regularly. At the same time, based on the feedback of operation and maintenance practices, the management department continuously optimizes and improves key links such as digital twin models, prediction models and evaluation systems, so as to continuously improve the entire rail operation and maintenance management system.

[0130] It should be noted that the explanations of the variables involved in the description of the present invention are shown in Tables 10 and 11 below.

[0131] Table 10 Variable explanation table (Part I)

[0132] Table 11 Variable Explanation Table (Part 2)

[0133] The above description is only a specific implementation mode of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A BIM-based rail operation and maintenance path optimization method, characterized in that: The following steps are involved: S10. Use the building information model to build a digital twin model of rail equipment, collect the three-dimensional position coordinate information, real-time equipment status information, historical status data, maintenance requirement information, fault history records, and maintenance records of rail equipment, pre-process and clean the collected data, remove abnormal data and supplement missing data, so as to achieve comprehensive acquisition of rail equipment information; S20, establishing a spatial distance matrix based on the three-dimensional position coordinates of the rail equipment, constructing a constraint matrix considering the maintenance order constraint relationship between the equipment, generating a priority weight matrix in combination with the equipment maintenance priority index, and fusing the spatial distance matrix, the constraint matrix, and the priority weight matrix to construct a complete rail operation and maintenance spatial topology network; S30, perform multivariate mixed decomposition calculation on historical operation and maintenance data, extract stable components and fluctuating components of operation and maintenance time series data, establish a multidimensional correlation model of equipment state parameters, environmental influencing factors and operation and maintenance duration, use deep learning methods to train the operation and maintenance time prediction model, verify the accuracy of the operation and maintenance time prediction model through cross-validation methods, and obtain deep learning prediction deviation and historical state volatility; S40, calculating the standard operation and maintenance duration of each device based on the operation and maintenance time prediction model, establishing a time influence coefficient matrix in combination with the fluctuation component, obtaining a time error compensation amount through matrix operation, and superimposing the standard operation and maintenance duration with the time error compensation amount to obtain a corrected device operation and maintenance time matrix; S50, taking minimization of total operation and maintenance cost, maximization of equipment reliability, and balancing of workload of operation and maintenance personnel as optimization objectives, establishing a weight allocation mechanism among the optimization objectives, taking the track operation and maintenance spatial topology network and the equipment operation and maintenance time matrix as constraints, and constructing a multi-objective optimization model; S60. Design a population initialization strategy based on operation and maintenance experience, construct an adaptive crossover operator and a mutation operator, introduce an elite retention mechanism, use an improved genetic algorithm to solve the multi-objective optimization model, record the convergence speed of the genetic algorithm and the number of iterative optimizations, and obtain an optimized operation and maintenance path plan through multi-generation iterative optimization; S70, performing local search on the optimized operation and maintenance path scheme based on the particle swarm optimization algorithm, obtaining the particle swarm optimal position vector, local search iteration step, and local search fitness value, focusing on optimizing the transfer distance between devices and the equipment maintenance sequence, and generating multiple groups of candidate operation and maintenance path schemes; S80. According to the objectives of minimizing operation and maintenance costs and maximizing equipment reliability, the candidate operation and maintenance path solutions are evaluated using an operation and maintenance path evaluation equation group to obtain a final operation and maintenance path solution.

2. The method according to claim 1, characterized in that The step S10 specifically includes: Step 101: Collect three-dimensional coordinates of track equipment, including equipment location coordinate information, real-time status information, and maintenance requirement information; Step 102, establishing a data collection cycle, recording the running status data, fault history records, and maintenance records of the track equipment during the data collection cycle to form an equipment history status database; Step 103: Use the 3-times standard deviation threshold method to clean abnormal data, and mark and delete data points that exceed the 3-times standard deviation range; Step 104: fill in the missing data using a weighted average method of the time series data before and after, wherein the weighted average method includes a linear weighted combination of the data at the previous moment, the data at the next moment, and the historical average value; Step 105: Establish a device information database, and classify and store the cleaned and filled data according to device number, timestamp, and data type.

3. The method according to claim 1, characterized in that The step S20 specifically includes: Step 201, calculating the Euclidean distance between any two devices according to the three-dimensional position coordinates of the track device, and establishing a distance matrix representing the spatial relationship between the devices; Step 202: Analyze the maintenance sequence constraint relationship between devices according to the equipment maintenance manual and establish a three-value constraint matrix, wherein the three values ​​include must be before, no constraint, and must be after; Step 203: Establish a weight matrix based on the equipment failure frequency index, reliability index, time urgency index, and importance index; Step 204: Use a linear weighted method to fuse the distance matrix, the constraint matrix, and the weight matrix to construct a complete track operation and maintenance space topology network.

4. The method according to claim 1, characterized in that The step S30 specifically includes: Step 301: Decompose the historical operation and maintenance data into trend components, seasonal components, and random components, and extract stable state parameters; Step 302: Establish a multi-layer deep neural network, where the input layer includes device state parameters and environmental influencing factors; Step 303: Use historical operation and maintenance data to train the multi-layer deep neural network to obtain an operation and maintenance time prediction model; Step 304: Calculate the prediction deviation through a cross-validation method and record the historical state volatility.

5. The method according to claim 1, characterized in that The step S40 specifically includes: Step 401: Calculate the standard operation and maintenance duration of each device based on the operation and maintenance time prediction model; Step 402: Establish a time impact coefficient matrix by combining equipment status fluctuation rate, environmental impact factor, and historical maintenance efficiency factor; Step 403: introducing a periodic fluctuation term and calculating a time error compensation amount; Step 404: superimpose the standard operation and maintenance duration and the time error compensation amount to obtain a corrected equipment operation and maintenance time matrix.

6. The method according to claim 1, characterized in that The step S50 specifically includes: Step 501: Establish a linear objective function with minimization of total operation and maintenance cost as the first optimization goal; Step 502: establishing an exponential reliability function with maximizing equipment reliability as the second optimization objective; Step 503: Establish a variance function with workload balancing of operation and maintenance personnel as the third optimization objective; Step 504: construct a multi-objective optimization model by taking the track operation and maintenance spatial topology network and the equipment operation and maintenance time matrix as constraints.

7. The method according to claim 1, characterized in that The step S60 specifically includes: Step 601: Design a population initialization strategy based on historical operation and maintenance experience; Step 602: construct an adaptive crossover operator based on the operation and maintenance time window; Step 603: construct an adaptive mutation operator based on device constraint relations; Step 604: Use an improved genetic algorithm to solve the multi-objective optimization model, and record the algorithm convergence speed and the number of iterations.

8. The method according to claim 1, characterized in that The step S70 specifically includes: Step 701: Use a particle swarm algorithm to perform a local search on the optimized operation and maintenance path solution; Step 702, record the particle swarm optimal position vector and local search step length; Step 703: Calculate the local search fitness value; Step 704: Focus on optimizing the transfer distance and maintenance sequence between devices, and generate multiple sets of candidate operation and maintenance path solutions.

9. The method according to claim 1, characterized in that: The step S80 specifically includes: Step 801, calculating the square sum of the particle swarm position convergence rate and establishing a path optimization score; Step 802: Calculate the deep learning prediction deviation integral value and establish a state prediction score; Step 803, calculating the partial derivative of the number of iterations with respect to the convergence rate, and establishing a time response score; Step 804: Use a nonlinear combination method to weight the scoring results, and select the solution with the highest comprehensive score as the final operation and maintenance path solution.

10. The method according to claim 1, characterized in that The method further includes S90, writing the final operation and maintenance path solution into the building information model, establishing a real-time feedback mechanism for operation and maintenance data and a solution evaluation index system, and realizing dynamic updating of the operation and maintenance knowledge base.

Images