Railway machine room equipment health degree prediction model training method and device, equipment and medium

Abstract

The invention relates to the technical field of railway machine room equipment management, and discloses a training method and device of a railway machine room equipment health degree prediction model, equipment and a medium. The parameter value time sequence data of the operation parameters of the target equipment in the railway machine room can be obtained. And normalizing the parameter value time sequence data of the target equipment operation parameters to obtain normalized parameter value time sequence data. And inputting the normalized parameter value time sequence data into a trained deep residual shrinkage network DRSN to perform attention mechanism screening, and obtaining screened parameter value time sequence data and taking the screened parameter value time sequence data as an equipment health index. And smoothing the equipment health index to obtain a processed equipment health index. And taking the processed equipment health indexes as training data, and training a to-be-trained equipment health degree prediction model based on the training data and a grey wolf optimization algorithm. The trained target equipment health degree prediction model can effectively predict the health degree of the railway machine room equipment.

Description

Technical Field

[0001] The present invention relates to the technical field of railway machine room equipment management, and in particular to a training method, device, equipment and medium for a railway machine room equipment health prediction model. Background Art

[0002] The railway machine room is the core part of the railway communication network. It has the characteristics of wide distribution, large number and high security requirements. Its normal operation and maintenance are crucial to ensuring railway safety services.

[0003] With the continuous improvement of railway informatization, the types and number of systems carried by railway machine rooms are increasing, which also makes operation and maintenance more complicated, with greater workload and management difficulty. It is very important to determine the operation status of railway machine room equipment.

[0004] However, the relevant technology cannot effectively determine the operating status of railway machine room equipment. Summary of the invention

[0005] The present invention provides a training method, device, equipment and medium for a railway machine room equipment health prediction model, which is used to solve the defect that the operation status of railway machine room equipment cannot be effectively determined in the related technology, and effectively realize the prediction of the health of railway machine room equipment.

[0006] In a first aspect, the present invention provides a method for training a railway machine room equipment health prediction model, comprising: Obtaining the time series data of the parameter values ​​of the target equipment operating parameters in the railway machine room; Normalizing the parameter value time series data of the target device operating parameter to obtain normalized parameter value time series data; Input the normalized parameter value time series data into the trained deep residual shrinkage network DRSN for attention mechanism screening, and obtain the screened parameter value time series data as the equipment health indicator; Smoothing the equipment health indicator to obtain a processed equipment health indicator; The processed equipment health index is used as training data, and the equipment health prediction model to be trained is trained based on the training data and the Grey Wolf Optimization Algorithm to obtain a trained target equipment health prediction model.

[0007] Optionally, the target device operating parameters include one or more device operating parameters, and the parameter value time series data of the target device operating parameters includes parameter value time series data of one or more device operating parameters.

[0008] Optionally, the equipment health prediction model is a bidirectional long short-term memory network BiLSTM model.

[0009] Optionally, the training of the equipment health prediction model to be trained based on the training data and the Grey Wolf Optimization Algorithm to obtain a trained target equipment health prediction model includes: Determine multiple hyperparameters to be optimized in the equipment health prediction model, and determine multiple initial parameters of the gray wolf optimization algorithm; wherein the multiple initial parameters include a search space, a maximum number of iterations, a dimension, and a gray wolf population size, and the search space includes a value range of each hyperparameter to be optimized; Initializing the position vector of each gray wolf in the population in the gray wolf optimization algorithm according to the search space, the dimension and the size of the gray wolf population, and assigning the position vector of each gray wolf in the population as the parameter value of the multiple hyperparameters to be optimized; Determine the fitness value of each of the position vectors according to the training data and the parameter values ​​of the multiple hyperparameters to be optimized, and determine the three smallest fitness values; Determine the position vectors corresponding to the three smallest fitness values ​​and use them as Wolf, Wolf and The wolf's position vector is based on Wolf, Wolf and The wolf's position vector updates the position vector of each of the gray wolves; The updated position vector of each gray wolf is used as the parameter value of the multiple hyperparameters to be optimized until the number of iterations reaches the maximum number of iterations, and the latest Wolf; The latest The wolf assigns final parameter values ​​of the multiple hyperparameters to be optimized to obtain the target device health prediction model.

[0010] Optionally, the multiple hyperparameters to be optimized include hidden layer nodes, number of training times and initial learning rate.

[0011] In a second aspect, the present invention provides a training device for a railway machine room equipment health prediction model, comprising: An acquisition unit, used to acquire parameter value time series data of operating parameters of target equipment in a railway machine room; A normalization unit, used to normalize the parameter value time series data of the target device operation parameter to obtain normalized parameter value time series data; A screening unit, used for inputting the normalized parameter value time series data into the trained deep residual shrinkage network DRSN for attention mechanism screening, obtaining the screened parameter value time series data as a device health indicator; A processing unit, used for smoothing the equipment health indicator to obtain a processed equipment health indicator; As a unit, used to use the processed equipment health indicator as training data; The training unit is used to train the equipment health prediction model to be trained based on the training data and the Grey Wolf Optimization Algorithm to obtain a trained target equipment health prediction model.

[0012] Optionally, the training unit is further used for: Determine multiple hyperparameters to be optimized in the equipment health prediction model, and determine multiple initial parameters of the gray wolf optimization algorithm; wherein the multiple initial parameters include a search space, a maximum number of iterations, a dimension, and a gray wolf population size, and the search space includes a value range of each hyperparameter to be optimized; Initializing the position vector of each gray wolf in the population in the gray wolf optimization algorithm according to the search space, the dimension and the size of the gray wolf population, and assigning the position vector of each gray wolf in the population as the parameter value of the multiple hyperparameters to be optimized; Determine the fitness value of each of the position vectors according to the training data and the parameter values ​​of the multiple hyperparameters to be optimized, and determine the three smallest fitness values; Determine the position vectors corresponding to the three smallest fitness values ​​and use them as Wolf, Wolf and The wolf's position vector is based on Wolf, Wolf and The wolf's position vector updates the position vector of each of the gray wolves; The updated position vector of each gray wolf is used as the parameter value of the multiple hyperparameters to be optimized until the number of iterations reaches the maximum number of iterations, and the latest Wolf; The latest The wolf assigns final parameter values ​​of the multiple hyperparameters to be optimized to obtain the target device health prediction model.

[0013] Optionally, the multiple hyperparameters to be optimized include hidden layer nodes, number of training times and initial learning rate.

[0014] In a third aspect, the present invention provides a computer device, comprising: a memory and a processor, the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the training method of the railway machine room equipment health prediction model of the above-mentioned first aspect or any corresponding embodiment thereof by executing the computer instructions.

[0015] In a fourth aspect, the present invention provides a computer-readable storage medium having computer instructions stored thereon, the computer instructions being used to enable a computer to execute the training method for a railway machine room equipment health prediction model of the above-mentioned first aspect or any corresponding embodiment thereof.

[0016] The training method, device, equipment and medium of the railway machine room equipment health prediction model provided by the present invention can obtain the parameter value time series data of the target equipment operating parameters in the railway machine room. The parameter value time series data of the target equipment operating parameters are normalized to obtain the normalized parameter value time series data. The normalized parameter value time series data is input into the trained deep residual shrinkage network DRSN for attention mechanism screening, and the screened parameter value time series data is obtained as the equipment health indicator. The equipment health indicator is smoothed to obtain the processed equipment health indicator. The processed equipment health indicator is used as training data, and the equipment health prediction model to be trained is trained based on the training data and the gray wolf optimization algorithm to obtain a trained target equipment health prediction model. The trained target equipment health prediction model of the present invention can predict the equipment health of the railway machine room according to the parameter value time series data of the equipment operating parameters in the railway machine room, and effectively realize the prediction of the equipment health of the railway machine room. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0018] Figure 1 A flowchart of a method for training a railway machine room equipment health prediction model provided by an embodiment of the present invention; Figure 2 A schematic diagram of the structure of a deep residual shrinkage network provided by an embodiment of the present invention; Figure 3 A schematic diagram of a related technology for fault prediction provided by an embodiment of the present invention; Figure 4 A schematic diagram of the structure of a training device for a railway machine room equipment health prediction model provided by an embodiment of the present invention; Figure 5 A schematic diagram of the structure of a computer device provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0019] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0020] Combine the following Figure 1-Figure 3 The present invention describes a training method for a railway machine room equipment health prediction model.

[0021] like Figure 1 As shown, this embodiment proposes a first method for training a railway machine room equipment health prediction model, which may include the following steps: S101. Obtain parameter value time series data of operating parameters of target equipment in a railway machine room.

[0022] Optionally, the target device operating parameters include one or more device operating parameters, and the parameter value time series data of the target device operating parameters includes parameter value time series data of one or more device operating parameters.

[0023] Specifically, the device operating parameters may be voltage, current, resistance and temperature.

[0024] S102: Normalize the parameter value time series data of the target device operating parameters to obtain normalized parameter value time series data.

[0025] It should be noted that in order to make the features have the same measurement scale and the neural network can converge quickly, the device data (voltage, current, resistance, etc.) collected by the intelligent computer room system is first normalized to divide the training set and test set samples of the health indicator extraction model. Specifically, this embodiment can use the Min-max normalization method for normalization.

[0026] Specifically, this embodiment can normalize each data in the parameter value time series data to obtain normalized parameter value time series data.

[0027] S103. Input the normalized parameter value time series data into the trained Deep Residual Shrinkage Networks (DRSN) for attention mechanism screening, and obtain the screened parameter value time series data as the device health indicator.

[0028] Specifically, this embodiment can build a DRSN, input the normalized parameter value time series data into the DRSN for attention mechanism screening, and obtain the screened parameter value time series data.

[0029] like Figure 2 The schematic diagram of the DRSN structure shown in the figure includes batch normalization + ReLU function, convolution layer, global average pooling layer, fully connected layer and Sigmoid function. Figure 2 C and W in are the relevant parameters of the input data. For network parameters, the residual network ResNet introduces an identity mapping layer in the convolutional network to form an inter-layer residual skip connection structure, which greatly alleviates the gradient explosion and degradation problems of deep networks and reduces the difficulty of training deep networks. Based on the ResNet network structure, DRSN adds attention mechanism and soft thresholding on the basis of deep residual network. DRSN replaces the nonlinear layer of the SENet structure with a soft threshold function and then integrates it into the ResNet network structure. Its essence is to replace the weighted operation with the soft thresholding of each feature channel in SENet, so that it can retain more useful features of the data set, thereby improving the feature learning ability of DRSN on complex data sets and finally obtaining a better model.

[0030] S104: Smoothing the equipment health index to obtain a processed equipment health index.

[0031] The processed equipment health index is data obtained after smoothing the equipment health index.

[0032] Specifically, after obtaining the equipment health index, this embodiment may perform smoothing processing on the equipment monitoring index to obtain a processed equipment health index.

[0033] S105: Use the processed equipment health indicators as training data.

[0034] Specifically, in this embodiment, the processed equipment health index can be used as training data for a railway machine room equipment health prediction model.

[0035] S106: Train the equipment health prediction model to be trained based on the training data and the Grey Wolf Optimization Algorithm to obtain a trained target equipment health prediction model.

[0036] Specifically, this embodiment can perform model training on the equipment health prediction model based on the processed equipment health index and the Grey Wolf Optimization Algorithm, and update the model parameters to obtain a target equipment health prediction model whose performance meets the requirements.

[0037] It can be understood that the target equipment health prediction model trained in this embodiment can predict the equipment health of the railway machine room based on the parameter value time series data of the equipment operating parameters in the railway machine room, thereby effectively realizing the prediction of the equipment health of the railway machine room.

[0038] The training method of the railway machine room equipment health prediction model proposed in this embodiment can obtain the parameter value time series data of the target equipment operating parameters in the railway machine room. The parameter value time series data of the target equipment operating parameters are normalized to obtain the normalized parameter value time series data. The normalized parameter value time series data is input into the trained deep residual shrinkage network DRSN for attention mechanism screening, and the screened parameter value time series data is obtained as the equipment health indicator. The equipment health indicator is smoothed to obtain the processed equipment health indicator. The processed equipment health indicator is used as training data, and the equipment health prediction model to be trained is trained based on the training data and the gray wolf optimization algorithm to obtain a trained target equipment health prediction model. The trained target equipment health prediction model in this embodiment can predict the equipment health of the railway machine room according to the parameter value time series data of the equipment operating parameters in the railway machine room, and effectively realize the prediction of the equipment health of the railway machine room.

[0039] It should be noted that the railway communication room is the core part of the railway communication network. It has the characteristics of wide distribution range, large number, many services and high security requirements. Its normal operation and maintenance are crucial to ensure railway safety services. With the continuous improvement of the railway informatization, the types and number of systems carried by the railway communication room are increasing, which also makes the operation and maintenance work more complicated, the workload is greater, and the management difficulty is increased. How to ensure the convenience and efficiency of operation and maintenance work has also become an urgent problem to be solved. At present, most of the communication rooms in the railway system are unmanned rooms. The operation and maintenance work at the station level is relatively scattered, and the operation and maintenance control capabilities are insufficient. When emergencies occur, they cannot be handled in time. Therefore, the daily operation and maintenance of the room is very important. Due to the rapid increase in the integration, complexity and intelligence of the railway communication room, the traditional fault diagnosis and maintenance guarantee technology is gradually difficult to adapt to the new requirements. If the monitoring data of the room can be statistically analyzed and the ideal value of the equipment operation status can be given, it will assist the operation and maintenance personnel to judge whether the detection indicators are healthy, help the operation and maintenance personnel eliminate a large amount of redundant information, and reduce the operation and maintenance costs. Alternatively, by modeling the health status of the server and issuing predictive alerts when there are signs of server failure, you can prevent problems before they occur and minimize the losses caused by equipment failure in the computer room.

[0040] The railway communication power supply and environment monitoring system is a system developed by our company to realize the automated operation and maintenance of the railway communication room. The system uses various sensors deployed in the monitoring station to monitor the room environment (temperature and humidity, door magnets, smoke sensors, water immersion, infrared, glass breakage, etc.). The monitoring unit is responsible for uploading various types of data collected in the room to the monitoring center via Ethernet, and receiving and executing various query control instructions issued by the monitoring center. The monitoring center mainly completes the analysis, storage, and forwarding of various types of data, and the monitoring terminal mainly completes the interactive operation with the user. However, the monitoring unit can only store and transmit the sensor data in a simple manner, but cannot make good use of these data to make a process prediction of the health of the equipment to achieve intelligent maintenance of the equipment in the room.

[0041] The goal of intelligent maintenance is to solve problems that arise in production and manufacturing, including equipment downtime, component damage, poor quality, and low energy utilization. To solve these problems, we cannot start from the appearance of these problems, but we must focus on the causes of these appearances. We generally call these causes invisible problems. Invisible problems include wear, corrosion, leakage, human operation and even environmental factors of parts. These factors have a variety of representations. These factors are likely to couple and cross each other and work together to produce these failures. For this situation, this embodiment can avoid these problems from the source, using monitoring, analysis and even decision support methods to solve these problems. Avoid the occurrence of invisible problems through prediction.

[0042] The evolution of the maintenance strategy is carried out in the following steps: Traditional Reactive Maintenance (RM): Waiting for equipment problems to occur before repairing; Preventive Maintenance (PM) for high-reliability equipment: For high-reliability equipment, the original traditional reactive maintenance cannot meet the reliability requirements. Therefore, we need to do preventive maintenance. This preventive maintenance is purely based on time and frequently replaces parts that have not been damaged (still can be used for a period of time). This maintenance method is very costly; Condition-Based Maintenance (CBM): This type of maintenance no longer relies solely on time, but detects early signs of failure through physical quantities measured from the equipment, thereby avoiding serious failures. Prognostics and Health Management (PHM): predicting the service life of equipment based on condition monitoring; Prognostics and Health Management (PHM) is a key technology for modern equipment to achieve automatic logistics (AL) and reduce the cost of the entire life cycle. The emergence of this technology has changed people's traditional maintenance methods for complex systems, evolving maintenance strategies from post-maintenance or regular maintenance to condition-based maintenance (CBM), and realizing the transition from traditional manual fault diagnosis to intelligent system fault prediction, thus providing a technical foundation for the transition from traditional "firefighting" repairs to proactive maintenance.

[0043] In the entire system of fault prediction and health management technology, realizing the prediction of health status change or remaining service life of PHM system is the core challenge of realizing the object system. The classification of related technologies on fault prediction methods can be divided into the following three categories: prediction methods based on physical models, fault prediction methods based on reliability models, and fault prediction methods based on data-driven. The differences between these three methods are roughly as follows: Figure 3 shown.

[0044] The inventor of the present invention has found that the fault prediction method based on the physical model is the most primitive and the method with the longest development time. This method requires that the mathematical model of the prediction object is thoroughly studied, and it is necessary to deeply explore the essential characteristics of the physical level of the research object, such as evaluating the degree of loss and cumulative effect of faults of each key component of the object system under different working conditions, and then modeling through expert knowledge and random process principles to obtain a prediction model for the remaining useful life (RUL) of the object system. This method needs to start with the study of the physical laws of the object system and find a physical mathematical model that can reflect the law of system performance change for prediction. Its advantage is that it does not need to collect a lot of data to achieve accurate prediction, but its disadvantage is that this method is extremely dependent on the accuracy of physical laws and mathematical models. For systems such as computer room servers and equipment that combine software and hardware, the degradation process and mechanism of their health status are not clear, and it is difficult to establish a more accurate physical model for their complex and changeable characteristics, so this method is not suitable for railway computer rooms.

[0045] If a large number of sample reliability experiments can be conducted on the research object, or if there is a large amount of historical fault data, it is more appropriate to use a fault prediction method based on a reliability model. This method does not require the establishment of a particularly accurate mathematical or physical model. Its working principle is to use the data information in the probability density function (Probability Density Functions, PDFs) related to the fault event for prediction. Among them, PDFs are obtained from statistical analysis of reliability experiments and historical data, so compared with the other two methods, this method requires a smaller information dimension. In the scenario of railway machine room equipment failure prediction, this method is not a suitable method because it cannot take into account many additional factors, such as historical status data, health degradation characteristics, system usage environment, etc. At the same time, the accuracy of the reliability model is generally low, and the false alarm rate of faults is relatively high.

[0046] The method of fault prediction based on data collection or experimental data is called a data-driven fault prediction method. This method uses the monitoring data of the research object to learn or identify the normal or abnormal mode of the object system from historical data, and converts the mode and data into a prediction model to achieve the prediction of the future of the system. Because the data-driven fault prediction method uses test and status monitoring data to predict the trend of the health state change of the system, it does not require prior knowledge of the object system, thus avoiding the shortcomings of prediction technology based on physical models and reliability models. Since there is an intelligent computer room operation and maintenance management system in the railway computer room, it can realize remote telemetry, remote signaling, remote control, and remote adjustment of various equipment in the computer room (such as power supply and distribution system, air conditioning system, access control system, lightning protection system, etc.), and real-time collection and monitoring of the operating status and environmental data of the system and equipment, which can provide sufficient data support for the prediction method. Therefore, this method is suitable for the detection environment of the railway communication computer room and can effectively utilize the monitoring data of the intelligent computer room system.

[0047] At present, the core task of the power environment monitoring system in the communication room is to provide real-time feedback on the operating status of the power equipment. It lacks the ability of comprehensive analysis and decision-making support, especially the insufficient use of the collected equipment data. It can only reflect abnormal data and abnormal indicators, and cannot perform coherent analysis and fault prediction based on previous historical data. Therefore, it is necessary to select the corresponding prediction scheme and specific algorithm based on the data characteristics collected by the intelligent room system.

[0048] The research object of fault prediction is mostly time series data. The general practice is to analyze historical time series data to obtain the development trend of future data. Therefore, the reasonable memory, extraction and application of historical data have become the key to the quality of time series data prediction. Recurrent Neural Network (RNN) is a network that uses time series data as network input, recursively in the direction of sequence evolution, and links nodes to a deep neural network. Since the network can realize the memory function of historical data, it is widely used in the field of remaining useful life (RUL) prediction. Although RNN provides a better solution for processing time series data, due to the existence of the gradient vanishing problem, RNN has "limited memory ability" in practical applications and can often only learn dependencies within a short time step of the network.

[0049] In the second training method of the railway machine room equipment health prediction model proposed in this embodiment, the equipment health prediction model is a bidirectional long short-term memory network (Bi-directional Long Short-Term Memory, BiLSTM) model.

[0050] Step S106 may include: Determine multiple hyperparameters to be optimized in the equipment health prediction model, and determine multiple initial parameters of the gray wolf optimization algorithm; wherein the multiple initial parameters include a search space, a maximum number of iterations, a dimension, and a gray wolf population size, and the search space includes a value range of each hyperparameter to be optimized; According to the search space, dimension and the size of the gray wolf population, the position vector of each gray wolf in the population is initialized in the gray wolf optimization algorithm, and the position vector of each gray wolf in the population is assigned to the parameter value of multiple hyperparameters to be optimized; Determine the fitness value of each position vector according to the training data and the parameter values ​​of multiple hyperparameters to be optimized, and determine the three smallest fitness values; Determine the position vectors corresponding to the three smallest fitness values ​​and use them as Wolf, Wolf and The wolf's position vector is based on Wolf, Wolf and The wolf's position vector updates the position vector of each gray wolf; The updated position vector of each gray wolf is used as the parameter value of multiple hyperparameters to be optimized until the number of iterations reaches the maximum number of iterations to determine the latest Wolf; The latest Wolf assigns final parameter values ​​of multiple hyperparameters to be optimized to obtain the target equipment health prediction model.

[0051] Optionally, the multiple hyperparameters to be optimized include hidden layer nodes, number of training times, and initial learning rate.

[0052] Specifically, this embodiment can optimize the BiLSTM model based on the Gray Wolf Optimization Algorithm. The Gray Wolf Optimization Algorithm is used to optimize the number of neurons, learning rate, and number of training times of the BiLSTM network to obtain the optimal prediction model.

[0053] Specifically, this embodiment can determine the hyperparameters that need to be optimized for the BiLSTM algorithm: hidden layer nodes, number of training times, and initial learning rate. Then determine the four initial parameters of the Gray Wolf Algorithm, search space, maximum number of iterations, and , dimension D, and the size of the gray wolf population N. The search space is the value range of all hyperparameters of the BiLSTM algorithm. Initialize the position vector of each gray wolf in the population . Assume that the current population is , represents the j-th dimension of the i-th individual in the population, j=1,2,…,D.

[0054] Afterwards, the position vector of each gray wolf in the population can be assigned to the three parameters that need to be optimized by BiLSTM, and the fitness value of the position vector can be calculated according to the following formula.

[0055]

[0056] in, For fitness, is the individual serial number, The total number of individuals. Represents the predicted value of the BiLSTM model; Indicates the expected output value.

[0057] Filter out the gray wolf position vectors corresponding to the three smallest fitness values ​​in the population and assign them to , , , update the position vector , , , so that it always keeps the fitness minimum. Determine whether the maximum number of iterations has been reached. If so, end the update. If not, continue to search for the best. , , for Wolf, Wolf and The wolf's position vector.

[0058] After the iteration, the optimal The position vector is assigned to the hidden layer nodes, training times, and initial learning rate of the BiLSTM neural network, and the model is trained and predicted according to these parameters.

[0059] Specifically, in this embodiment, three indicators, namely, root mean square error, mean absolute percentage error and mean absolute error, may be selected to evaluate the performance of the model.

[0060] It should be noted that PHM technology is a maintenance strategy that is based on the real-time status and historical data of machines, equipment or systems. Through technical means such as data collection, analysis, diagnosis and prediction, it predicts and judges their health status, thereby improving their reliability and availability.

[0061] This embodiment can analyze the sensor parameters of the monitoring equipment in the intelligent computer room system in real time, and use a large amount of historical data and prediction models to monitor and evaluate the health status of machines, equipment or systems, so as to predict the occurrence of failures in advance and take corresponding maintenance measures in time to avoid the sudden shutdown and failure of equipment from bringing catastrophic effects to the enterprise. In addition, by monitoring and analyzing the real-time status and operation data of machines, equipment or systems, their health status, including existing problems and potential risks, can be diagnosed, so as to timely eliminate faults and optimize equipment maintenance plans.

[0062] This embodiment can use the PHM concept of fault prediction and health management to achieve accurate operation and maintenance of equipment and improve the efficiency of equipment operation and maintenance. The equipment status information and alarm information data generated during the operation of the communication room are processed and the BiLSTM network model is used to predict the health status of the communication room equipment. This embodiment can transform the health management of railway room equipment from the "after-the-fact" management method in related technologies to the "prepare for a rainy day" management method, and achieve zero downtime and continuous and reliable operation of equipment through predictive maintenance, reducing maintenance workload and safety hazards, and meeting the needs of railway development.

[0063] like Figure 4 As shown, this embodiment proposes a training device for a railway machine room equipment health prediction model, which may include: The acquisition unit 401 is used to acquire the parameter value time series data of the operating parameters of the target equipment in the railway machine room; A normalization unit 402 is used to normalize the parameter value time series data of the target device operation parameter to obtain normalized parameter value time series data; A screening unit 403 is used to input the normalized parameter value time series data into the trained deep residual shrinkage network DRSN for attention mechanism screening, and obtain the screened parameter value time series data as a device health indicator; The processing unit 404 is used to perform smoothing processing on the device health indicator to obtain a processed device health indicator; As unit 405, for using the processed device health indicator as training data; The training unit 406 is used to train the equipment health prediction model to be trained based on the training data and the Grey Wolf Optimization Algorithm to obtain a trained target equipment health prediction model.

[0064] It should be noted that the processing of the acquisition unit 401, the normalization unit 402, the screening unit 403, the processing unit 404, the serving unit 405 and the training unit 406 and the beneficial effects thereof can be respectively referred to in Figure 1 Steps S101 to S106 in the above are not described in detail.

[0065] Optionally, the target device operating parameters include one or more device operating parameters, and the parameter value time series data of the target device operating parameters includes parameter value time series data of one or more device operating parameters.

[0066] Optionally, the device health prediction model is a bidirectional long short-term memory network BiLSTM model.

[0067] Optionally, the training unit 406 is further configured to: Determine multiple hyperparameters to be optimized in the equipment health prediction model, and determine multiple initial parameters of the gray wolf optimization algorithm; wherein the multiple initial parameters include a search space, a maximum number of iterations, a dimension, and a gray wolf population size, and the search space includes a value range of each hyperparameter to be optimized; According to the search space, dimension and the size of the gray wolf population, the position vector of each gray wolf in the population is initialized in the gray wolf optimization algorithm, and the position vector of each gray wolf in the population is assigned to the parameter value of multiple hyperparameters to be optimized; Determine the fitness value of each position vector according to the training data and the parameter values ​​of multiple hyperparameters to be optimized, and determine the three smallest fitness values; Determine the position vectors corresponding to the three smallest fitness values ​​and use them as Wolf, Wolf and The wolf's position vector is based on Wolf, Wolf and The wolf's position vector updates the position vector of each gray wolf; The updated position vector of each gray wolf is used as the parameter value of multiple hyperparameters to be optimized until the number of iterations reaches the maximum number of iterations to determine the latest Wolf; The latest Wolf assigns final parameter values ​​of multiple hyperparameters to be optimized to obtain the target equipment health prediction model.

[0068] Optionally, the multiple hyperparameters to be optimized include hidden layer nodes, training times, and initial learning rate.

[0069] The training device of the railway machine room equipment health prediction model proposed in this embodiment can obtain the parameter value time series data of the target equipment operating parameters in the railway machine room. The parameter value time series data of the target equipment operating parameters are normalized to obtain the normalized parameter value time series data. The normalized parameter value time series data is input into the trained deep residual shrinkage network DRSN for attention mechanism screening, and the screened parameter value time series data is obtained as the equipment health indicator. The equipment health indicator is smoothed to obtain the processed equipment health indicator. The processed equipment health indicator is used as training data, and the equipment health prediction model to be trained is trained based on the training data and the gray wolf optimization algorithm to obtain a trained target equipment health prediction model. The trained target equipment health prediction model in this embodiment can predict the equipment health of the railway machine room according to the parameter value time series data of the equipment operating parameters in the railway machine room, and effectively realize the prediction of the equipment health of the railway machine room.

[0070] The training device of the railway machine room equipment health prediction model in this embodiment is presented in the form of a functional unit, where the unit refers to an ASIC (Application Specific Integrated Circuit) circuit, a processor and memory that executes one or more software or fixed programs, and / or other devices that can provide the above functions.

[0071] The embodiment of the present invention also provides a computer device having the above Figure 4 The training device of the railway machine room equipment health prediction model is shown.

[0072] See also Figure 5, a schematic diagram of the structure of a computer device provided by an optional embodiment of the present invention, the computer device includes: one or more processors 10, a memory 20, and interfaces for connecting various components, including high-speed interfaces and low-speed interfaces. The various components are connected to each other using different buses for communication, and can be installed on a common motherboard or installed in other ways as needed. The processor can process instructions executed in the computer device, including instructions stored in or on the memory to display graphical information of the GUI on an external input / output device (such as a display device coupled to the interface). In some optional embodiments, if necessary, multiple processors and / or multiple buses can be used together with multiple memories and multiple memories. Similarly, multiple computer devices can be connected, and each device provides part of the necessary operations (for example, as a server array, a group of blade servers, or a multi-processor system). Figure 5 A processor 10 is taken as an example.

[0073] The processor 10 may be a central processing unit, a network processor or a combination thereof. The processor 10 may further include a hardware chip. The hardware chip may be a dedicated integrated circuit, a programmable logic device or a combination thereof. The programmable logic device may be a complex programmable logic device, a field programmable gate array, a general purpose array logic or any combination thereof.

[0074] The memory 20 stores instructions executable by at least one processor 10, so that at least one processor 10 executes the method shown in the above embodiment.

[0075] The memory 20 may include a program storage area and a data storage area, wherein the program storage area may store an operating system, an application required for at least one function. The data storage area may store data created according to the use of the computer device, etc. In addition, the memory 20 may include a high-speed random access memory, and may also include a non-transient memory, such as at least one disk storage device, a flash memory device, or other non-transient solid-state storage devices. In some optional embodiments, the memory 20 may optionally include a memory remotely arranged relative to the processor 10, and these remote memories may be connected to the computer device via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0076] The memory 20 may include a volatile memory, such as a random access memory. The memory may also include a non-volatile memory, such as a flash memory, a hard disk or a solid state drive. The memory 20 may also include a combination of the above-mentioned types of memory.

[0077] The computer device further comprises a communication interface 30 for the computer device to communicate with other devices or a communication network.

[0078] The embodiment of the present invention also provides a computer-readable storage medium. The method according to the embodiment of the present invention can be implemented in hardware, firmware, or can be implemented as a computer code that can be recorded in a storage medium, or can be implemented as a computer code that is originally stored in a remote storage medium or a non-temporary machine-readable storage medium and will be stored in a local storage medium through a network download, so that the method described herein can be stored in such software processing on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. Among them, the storage medium can be a magnetic disk, an optical disk, a read-only storage memory, a random access memory, a flash memory, a hard disk or a solid-state hard disk, etc.; further, the storage medium can also include a combination of the above types of memories. It can be understood that a computer, a processor, a microprocessor controller, or programmable hardware includes a storage component that can store or receive software or computer code. When the software or computer code is accessed and executed by a computer, a processor, or hardware, the method shown in the above embodiment is implemented.

[0079] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A training method for a railway machine room equipment health prediction model, characterized in that: include: Obtaining the time series data of the parameter values ​​of the target equipment operating parameters in the railway machine room; Normalizing the parameter value time series data of the target device operating parameter to obtain normalized parameter value time series data; Input the normalized parameter value time series data into the trained deep residual shrinkage network DRSN for attention mechanism screening, and obtain the screened parameter value time series data as the equipment health indicator; Smoothing the equipment health indicator to obtain a processed equipment health indicator; The processed equipment health index is used as training data, and the equipment health prediction model to be trained is trained based on the training data and the Grey Wolf Optimization Algorithm to obtain a trained target equipment health prediction model.

2. The method according to claim 1, characterized in that The target device operating parameters include one or more device operating parameters, and the parameter value time series data of the target device operating parameters includes parameter value time series data of one or more device operating parameters.

3. The method according to claim 1, characterized in that: The equipment health prediction model is a bidirectional long short-term memory network BiLSTM model.

4. The method according to claim 1, characterized in that The training of the equipment health prediction model to be trained based on the training data and the gray wolf optimization algorithm to obtain a trained target equipment health prediction model includes: Determine multiple hyperparameters to be optimized in the equipment health prediction model, and determine multiple initial parameters of the gray wolf optimization algorithm; wherein the multiple initial parameters include a search space, a maximum number of iterations, a dimension, and a gray wolf population size, and the search space includes a value range of each hyperparameter to be optimized; Initializing the position vector of each gray wolf in the population in the gray wolf optimization algorithm according to the search space, the dimension and the size of the gray wolf population, and assigning the position vector of each gray wolf in the population as the parameter value of the multiple hyperparameters to be optimized; Determine the fitness value of each of the position vectors according to the training data and the parameter values ​​of the multiple hyperparameters to be optimized, and determine the three smallest fitness values; Determine the position vectors corresponding to the three smallest fitness values ​​and use them as Wolf, Wolf and The wolf's position vector is based on Wolf, Wolf and The wolf's position vector updates the position vector of each of the gray wolves; The updated position vector of each gray wolf is used as the parameter value of the multiple hyperparameters to be optimized until the number of iterations reaches the maximum number of iterations, and the latest Wolf; The latest The wolf assigns final parameter values ​​of the multiple hyperparameters to be optimized to obtain the target device health prediction model.

5. The method according to claim 4, characterized in that The multiple hyperparameters to be optimized include hidden layer nodes, training times and initial learning rate.

6. A training device for a railway machine room equipment health prediction model, characterized in that: include: An acquisition unit, used to acquire parameter value time series data of operating parameters of target equipment in a railway machine room; A normalization unit, used to normalize the parameter value time series data of the target device operation parameter to obtain normalized parameter value time series data; A screening unit, used for inputting the normalized parameter value time series data into the trained deep residual shrinkage network DRSN for attention mechanism screening, obtaining the screened parameter value time series data as a device health indicator; A processing unit, used for smoothing the equipment health indicator to obtain a processed equipment health indicator; As a unit, used to use the processed equipment health indicator as training data; The training unit is used to train the equipment health prediction model to be trained based on the training data and the Grey Wolf Optimization Algorithm to obtain a trained target equipment health prediction model.

7. The device according to claim 6, characterized in that The training unit is further used for: Determine multiple hyperparameters to be optimized in the equipment health prediction model, and determine multiple initial parameters of the gray wolf optimization algorithm; wherein the multiple initial parameters include a search space, a maximum number of iterations, a dimension, and a gray wolf population size, and the search space includes a value range of each hyperparameter to be optimized; Initializing the position vector of each gray wolf in the population in the gray wolf optimization algorithm according to the search space, the dimension and the size of the gray wolf population, and assigning the position vector of each gray wolf in the population as the parameter value of the multiple hyperparameters to be optimized; Determine the fitness value of each of the position vectors according to the training data and the parameter values ​​of the multiple hyperparameters to be optimized, and determine the three smallest fitness values; Determine the position vectors corresponding to the three smallest fitness values ​​and use them as Wolf, Wolf and The wolf's position vector is based on Wolf, Wolf and The wolf's position vector updates the position vector of each of the gray wolves; The updated position vector of each gray wolf is used as the parameter value of the multiple hyperparameters to be optimized until the number of iterations reaches the maximum number of iterations, and the latest Wolf; The latest The wolf assigns final parameter values ​​of the multiple hyperparameters to be optimized to obtain the target device health prediction model.

8. The device according to claim 7, characterized in that The multiple hyperparameters to be optimized include hidden layer nodes, training times and initial learning rate.

9. A computer device, characterized in that: include: A memory and a processor, wherein the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the training method for the railway machine room equipment health prediction model according to any one of claims 1 to 5 by executing the computer instructions.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a computer to execute the training method for the railway machine room equipment health prediction model according to any one of claims 1 to 5.

Images