Linkage scheme prediction method, device and equipment for rail transit and storage medium

Abstract

The invention provides a linkage scheme prediction method and device for rail transit, equipment and a storage medium, and is applied to the technical field of rail transit, and the method comprises the steps: obtaining first monitoring data collected by a first sensor at a target object in the rail transit; if detecting that the first monitoring data is abnormal, acquiring abnormal monitoring data in the first monitoring data; inputting the abnormal monitoring data into a preset deep learning model for linkage scheme prediction processing, and determining at least one candidate linkage scheme corresponding to the target object; the deep learning model is obtained by performing supervised training according to a plurality of sample linkage schemes and known working conditions corresponding to each sample linkage scheme and / or performing unsupervised training according to a plurality of known working conditions in advance; and determining a target linkage scheme corresponding to the target object according to the at least one candidate linkage scheme of the target object. By adopting the technical scheme of the invention, the intelligence of the predicted linkage scheme can be improved.

Description

Technical Field

[0001] The present invention relates to the technical field of rail transit, and particularly to a method, device, equipment and storage medium for predicting a linkage scheme of rail transit. Background Art

[0002] With the rapid development of urban rail transit, the operation complexity of rail transit has been increasing continuously, and passengers' requirements for the travel experience of rail transit are also rising day by day. For example, when the platform temperature of rail transit is too high, the rail transit system needs to determine a scheme for adjusting the platform temperature based on the platform temperature, so as to cool down the platform through this scheme and ensure that the platform temperature is appropriate.

[0003] In the related art, when the rail transit system needs to set an adjustment scheme for a certain area or equipment, it usually determines the adjustment scheme suitable for this area or equipment based on the fixed correspondence between the parameters of the preset area or equipment and the adjustment scheme.

[0004] However, the adjustment scheme determined by the above technology is not intelligent enough. Summary of the Invention

[0005] The present invention provides a method, device, equipment and storage medium for predicting a linkage scheme of rail transit, which is used to solve the defect that the adjustment scheme determined by the fixed correspondence between the parameters of the preset area or equipment and the adjustment scheme in the prior art is not intelligent enough, and realizes predicting and processing the linkage scheme for the abnormal monitoring data at the target object in rail transit through a pre-trained deep learning model, so as to intelligently associate a more reasonable linkage scheme, improve the intelligence of the predicted linkage scheme and the linkage of the rail transit system, and further achieve the purpose of improving the operation efficiency, ensuring the system safety and optimizing the passenger experience.

[0006] The present invention provides a method for predicting a linkage scheme of rail transit, including: Obtaining first monitoring data collected by a first sensor at a target object in rail transit; the above target object includes a target area and / or a target vehicle; Performing anomaly detection on the first monitoring data, and if it is detected that the first monitoring data is abnormal, obtaining the abnormal monitoring data with anomalies in the first monitoring data; Input the abnormal monitoring data into a preset deep learning model for predicting and processing linkage solutions, and determine at least one candidate linkage solution corresponding to the target object; each candidate linkage solution includes a plurality of linked first candidate actions, and each first candidate action corresponds to an adjustment action for adjusting the abnormal situation existing in the target object. The above deep learning model is pre-trained in a supervised manner according to a plurality of sample linkage solutions and the known working conditions corresponding to each sample linkage solution and / or is trained in an unsupervised manner according to a plurality of known working conditions; According to at least one candidate linkage solution of the target object, determine the target linkage solution corresponding to the target object; the above target linkage solution includes a plurality of linked target actions and the control parameters of each target action.

[0007] According to a method for predicting a linkage solution of rail transit provided by the present invention, the above determining the target linkage solution corresponding to the target object according to at least one candidate linkage solution of the target object includes: According to the priorities corresponding to each first candidate action in each candidate linkage solution, determine the initial quantization values corresponding to each first candidate action in each candidate linkage solution; the priorities corresponding to each first candidate action in the candidate linkage solution are not completely the same; According to the initial quantization values corresponding to each first candidate action in each candidate linkage solution, perform quantization processing on each candidate linkage solution to determine the target quantization value corresponding to each candidate linkage solution; According to the target quantization value corresponding to each candidate linkage solution, determine the target linkage solution among the candidate linkage solutions.

[0008] According to a method for predicting a linkage solution of rail transit provided by the present invention, the above determining the target linkage solution corresponding to the target object according to at least one candidate linkage solution of the target object includes: Display at least one candidate linkage solution; Obtain the selection operation input by the user among at least one candidate linkage solution; In response to the selection operation, determine the candidate linkage solution corresponding to the selection operation as the target linkage solution.

[0009] According to a method for predicting a linkage solution of rail transit provided by the present invention, the above method further includes: Obtain the performance data of the target object and / or the performance data of the target device in the target object; According to the performance data of the target object and / or the performance data of the target device, adjust the control parameters of at least one target linkage action in the target linkage solution.

[0010] According to a method for predicting a linkage solution of rail transit provided by the present invention, the training method of the above deep learning model includes: Obtain the sample linkage schemes of multiple sample objects and the known working conditions corresponding to each sample linkage scheme; the known working conditions include the sample abnormal monitoring data at the sample objects; Input each known working condition into the initial deep learning model for linkage scheme prediction processing to determine the predicted linkage schemes corresponding to the sample objects of each known working condition; Calculate the losses between each predicted linkage scheme and the corresponding sample linkage scheme; Supervisedly train the initial deep learning model according to each loss, and / or unsupervisedly train the initial deep learning model according to each known working condition to obtain a deep learning model; the above supervised training is used to learn the mapping relationship between the known working conditions and the corresponding sample linkage schemes, and the above unsupervised training is used to explore the internal correlation relationships between various types of data in the known working conditions.

[0011] According to a method for predicting a linkage scheme of rail transit provided by the present invention, the above method further includes: Obtain the second monitoring data collected by the second sensor at the target object; The above-mentioned inputting the abnormal monitoring data into a preset deep learning model for linkage scheme prediction processing to determine at least one candidate linkage scheme corresponding to the target object includes: Input the abnormal monitoring data and the second monitoring data into a preset deep learning model for linkage scheme prediction processing to determine at least one candidate linkage scheme corresponding to the target object; each of the above candidate linkage schemes includes a first candidate action and a second candidate action, and the second candidate action is used to perform an energy-saving action on the energy consumption-related equipment in the target object.

[0012] According to a method for predicting a linkage scheme of rail transit provided by the present invention, the above method further includes: Obtain the execution result obtained after the equipment at the target object executes the first candidate action at a historical time, and obtain the third monitoring data collected by the third sensor at the target object; The above-mentioned inputting the abnormal monitoring data into a preset deep learning model for linkage scheme prediction processing to determine at least one candidate linkage scheme corresponding to the target object includes: Input the abnormal monitoring data, the execution result, and the third monitoring data into a preset deep learning model for linkage scheme prediction processing to determine at least one candidate linkage scheme corresponding to the target object.

[0013] The present invention also provides a device for predicting a linkage scheme of rail transit, including the following modules: A first monitoring data acquisition module, configured to acquire the first monitoring data collected by the first sensor at the target object in rail transit; the above target object includes a target area and / or a target vehicle; Anomaly detection module, configured to perform anomaly detection on the first monitoring data. If anomalies are detected in the first monitoring data, obtain the anomaly monitoring data with anomalies in the first monitoring data; Linkage plan prediction module, configured to input the anomaly monitoring data into a preset deep learning model for linkage plan prediction processing to determine at least one candidate linkage plan corresponding to the target object; each candidate linkage plan includes a plurality of linked first candidate actions, and each first candidate action corresponds to an adjustment action for adjusting the anomalies existing in the target object. The above deep learning model is pre-trained in a supervised manner according to a plurality of sample linkage plans and the known working conditions corresponding to each sample linkage plan and / or trained in an unsupervised manner according to a plurality of known working conditions; Target linkage plan determination module, configured to determine the target linkage plan corresponding to the target object according to at least one candidate linkage plan of the target object; the above target linkage plan includes a plurality of linked target actions and the control parameters of each target action.

[0014] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the computer program, the linkage plan prediction method for rail transit as described in any one of the above is implemented.

[0015] The present invention also provides a non-transitory computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the linkage plan prediction method for rail transit as described in any one of the above is implemented.

[0016] The present invention also provides a computer program product, including a computer program. When the computer program is executed by a processor, the linkage plan prediction method for rail transit as described in any one of the above is implemented.

[0017] The method, device, equipment and storage medium for predicting the linkage scheme of rail transit provided by the present invention obtain the first monitoring data collected by the first sensor at the target object including the target area and / or the target vehicle in the rail transit, and perform anomaly detection on the first monitoring data. If it is detected that the first monitoring data is abnormal, the abnormal monitoring data existing in the first monitoring data is obtained, and the abnormal monitoring data is input into the pre-trained deep learning model for linkage scheme prediction processing to determine at least one candidate linkage scheme corresponding to the target object, and then the target linkage scheme of the target object is determined according to at least one candidate linkage scheme of the target object; wherein, the deep learning model is pre-trained by supervised training and / or unsupervised training according to a plurality of sample linkage schemes and the known working conditions corresponding to each sample linkage scheme. Each candidate linkage scheme includes a plurality of linked first candidate actions, and each first candidate action corresponds to an adjustment action for adjusting the abnormal situation existing in the target object. The target linkage scheme includes a plurality of linked target actions and the control parameters of each target action. In this method, since the deep learning model pre-trained by the combination of supervised and unsupervised methods can be used to predict the linkage scheme of the monitoring data at the rail transit target object, compared with the adjustment scheme determined by the fixed correspondence, the linkage scheme predicted by the technical solution of the present invention is more intelligent and reasonable, so that the intelligence of the predicted linkage scheme can be improved and the linkage of the rail transit system can be improved; at the same time, since the data at the rail transit target object can be monitored for anomalies, and when the monitoring data is abnormal, the pre-trained deep learning model is used to predict the linkage scheme, the efficiency of predicting the linkage scheme is higher. Therefore, the rail transit system can respond to abnormal situations in a timely and rapid manner, ensuring the operation safety of the rail transit and improving the use experience of passengers. Description of the Drawings

[0018] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the following drawings are some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0019] Figure 1 It is one of the flow diagrams of the method for predicting the linkage scheme of rail transit provided by the present invention.

[0020] Figure 2 It is the second flow diagram of the method for predicting the linkage scheme of rail transit provided by the present invention.

[0021] Figure 3 It is the third flow diagram of the method for predicting the linkage scheme of rail transit provided by the present invention.

[0022] Figure 4 It is a schematic structural diagram of the linkage scheme prediction device for rail transit provided by the present invention.

[0023] Figure 5 It is a schematic structural diagram of the electronic device provided by the present invention. Specific embodiments

[0024] To make the objectives, technical solutions and advantages of the present invention clearer, the technical solutions in the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the present invention. Apparently, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments in the present invention without making creative efforts fall within the scope of protection of the present invention.

[0025] Currently, when a rail transit system needs to set an adjustment plan for a certain area or equipment, it usually determines the adjustment plan suitable for the area or equipment based on the fixed correspondence between the parameters of the preset area or equipment and the adjustment plan. However, the adjustment plan determined by the above technology is not intelligent enough. For example, when the platform temperature rises, only the air conditioner can be started to cool down, and other related factors such as ventilation cannot be comprehensively considered; when a device fails, the relevant systems cannot be timely linked to make a comprehensive response; during the peak passenger flow period, the evacuation measures are single, and it is difficult to efficiently divert passengers. In addition, the above technology is difficult to automatically associate and recommend more optimized linkage plans according to real-time changes, and cannot meet the requirements of efficient, safe and comfortable operation of modern rail transit. Based on this, the embodiments of the present invention provide a method, device, equipment and storage medium for predicting the linkage plan of rail transit, which can solve the above technical problems.

[0026] It should be noted that the execution subject of the embodiments of the present invention can be a linkage plan prediction device for rail transit, or an electronic device including the linkage plan prediction device for rail transit, or a rail transit system including the linkage plan prediction device for rail transit, or other devices or systems or devices, which are not specifically limited here. The following embodiments will be described by taking the rail transit system as the execution subject as an example. In addition, the above electronic device can be a terminal or a server.

[0027] Figure 1 It is one of the flow schematic diagrams of the method for predicting the linkage plan of rail transit provided by the present invention. As Figure 1 shown, the method includes the following steps: Step 102, obtaining first monitoring data collected by a first sensor at a target object in rail transit; the above target object includes a target area and / or a target vehicle.

[0028] Among them, the target objects in rail transit can include target areas or target vehicles. The target areas include target areas in rail transit, target vehicles, etc. The target areas can be, for example, platforms, areas at the entrance of stations, areas at the exit of stations, vehicle transfer channels at stations, etc. Or the target area can also be a certain sub - area in the platform area; the target vehicle can be a vehicle running on the rail transit line, such as a train, a bus, etc. One or more types of sensors can be pre - distributed or set at the key parts of any target object in the above - mentioned rail transit for collecting corresponding data. The number of each type of sensor can be one or more. The sensors here include, but are not limited to, temperature sensors, humidity sensors, light sensors, passenger flow counters, equipment status monitors, smoke sensors, position sensors, etc. Each type of sensor can collect corresponding monitoring data. These sensors can all adopt high - precision and high - reliability industrial - grade sensor components, and have the ability to collect data in real - time and continuously. Their sampling frequencies are set according to the changing characteristics of different data types. For example, the temperature and humidity sensors collect temperature data every 2 minutes, and the passenger flow counter updates the passenger flow data every 30 seconds.

[0029] Specifically, one or more first sensors at the above - mentioned target object can collect their corresponding types of monitoring data. The collected monitoring data is recorded as the first monitoring data. Here, the first monitoring data can be the monitoring data at a certain moment or the monitoring data within a certain time period.

[0030] Of course, other types of sensors at the target object can also collect their corresponding types of monitoring data. Here, both the first sensors and other types of sensors can collect data in real - time to ensure the real - time nature of the data and the real - time nature of the subsequent prediction linkage scheme.

[0031] Step 104, perform anomaly detection on the first monitoring data. If it is detected that the first monitoring data is abnormal, obtain the abnormal monitoring data that is abnormal in the first monitoring data.

[0032] In this step, after obtaining the first monitoring data collected by the first sensors at the target object, in order to facilitate the subsequent determination of when exactly the linkage scheme prediction process needs to be started, a conditional trigger mechanism can be set. The conditional trigger mechanism here can specifically be that after obtaining the first monitoring data, the first monitoring data can be detected for abnormalities through set abnormal conditions. If it is detected that the first monitoring data is abnormal, the subsequent linkage scheme prediction function is triggered, and then the subsequent linkage scheme prediction process can be continued. If it is detected that the first monitoring data is not abnormal, the subsequent linkage scheme prediction function is not triggered, and the monitoring data can be continuously obtained for anomaly detection.

[0033] When detecting whether the first monitoring data is abnormal, it can be to match the first monitoring data with the set abnormal conditions. If the match is successful, it is determined that the first monitoring data is abnormal, and the subsequent linkage scheme prediction function can be triggered. The set abnormal conditions here can include various abnormal conditions (i.e., including multiple condition trigger points). For example, it can include thresholds or threshold ranges related to the data type of the monitoring data, or it can also include the data change threshold of the monitoring data within a set time, or it can also include the failure times threshold of the monitoring data within a set time, etc. The set time can be set according to the actual situation, such as 2 minutes, 5 minutes, 10 minutes, etc. For example, taking the platform environment as an example, the temperature sensor in the middle of a certain platform detects that the temperature rises from 28 degrees to 32 degrees within 10 minutes, and the temperature rises by 4 degrees within 10 minutes, exceeding the preset data change threshold of 3 degrees / 5 minutes, that is, rising 3 degrees within 5 minutes, then it is determined that the temperature data in this platform is abnormal. Another example is a train in operation. The door system suddenly issues a fault warning, the door closing time is extended to 3 seconds and reports errors continuously 4 times. The door closing time exceeds the door closing time threshold by 2 seconds, and the number of failures 4 times exceeds the failure times threshold of 0 times, then it is determined that the door closing data of this train is abnormal. Another example is in terms of equipment operation. If the door system of a certain train monitors a fault warning, such as the door closing time is extended by more than 2 seconds or reports errors frequently more than 3 times within 10 minutes, then it is determined that the operation data of this train is abnormal, and the system immediately triggers and activates the association recommendation function / process.

[0034] When it is detected that the first monitoring data is abnormal, the monitoring data with abnormalities in the first monitoring data can be obtained and recorded as abnormal monitoring data. For example, the temperature data collected within 10 minutes can be used as abnormal monitoring data, or for another example, the temperature data at any moment within 10 minutes can be used as abnormal monitoring data, etc.

[0035] In addition, the monitoring data collected by other types of sensors at the target object can also be monitored for abnormalities in the above manner, and the result of whether there are abnormalities in other types of monitoring data can be obtained, and when there are abnormalities, the monitoring data with abnormalities in other types of monitoring data can be obtained.

[0036] Step 106, input the abnormal monitoring data into a preset deep learning model for linkage scheme prediction processing to determine at least one candidate linkage scheme corresponding to the target object; each candidate linkage scheme includes multiple linked first candidate actions, and each first candidate action corresponds to an adjustment action for adjusting the abnormal situation existing in the target object. The above deep learning model is pre-trained in a supervised manner according to multiple sample linkage schemes and the known working conditions corresponding to each sample linkage scheme and / or is trained unsupervised according to multiple known working conditions.

[0037] In this step, to facilitate more intelligent prediction of the linkage plan, a deep learning model can be pre-trained. The deep learning model can be trained using either unsupervised learning or supervised learning, or a combination of both. When training the deep learning model, supervised training can be performed on the deep learning model using multiple known working conditions and the optimal linkage plan under each known working condition, which can facilitate the model to quickly and accurately learn the mapping relationship between the working conditions and the linkage plan. At the same time, unsupervised training can be performed on the deep learning model using multiple known working conditions, which can enable the deep learning model to exert its imagination / association ability, associate possible linkage plans based on these known working conditions, and improve the intelligence and diversity of the associated plans. In addition, the architecture and type of the above deep learning model can be set according to the actual situation. For example, it can be a deep learning model constructed based on deep learning frameworks such as TensorFlow or PyTorch, which can adopt a multi-layer neural network structure, such as including an input layer, a hidden layer, and an output layer. Among them, the hidden layer can be set to 3 to 5 layers according to the complexity of data association, and the number of neurons in each layer is determined through experiments and optimization.

[0038] The optimal linkage plan under each of the above known working conditions can be recorded as a sample linkage plan. Each sample linkage plan can include multiple linked sample actions. Each sample action includes an adjustment action for adjusting the abnormal situation under the known working condition. Through a series of linked sample actions in each sample linkage plan, the abnormal situation under the known working condition can be improved, and the rail transit can be restored to normal as soon as possible. The above known working conditions can be the monitoring data of abnormal situations in the sample area or sample vehicle in the rail transit at historical times. Of course, it can also include normal monitoring data. Here, the monitoring data of the known working conditions can be collected by various sensors in the corresponding sample area or sample vehicle.

[0039] Through the above method, a deep learning model can be trained well. The input of the trained deep learning model is the working conditions such as the monitoring data of the target object, and the output is the linkage plan of the target object obtained by association / prediction. After obtaining the abnormal monitoring data at the target object, the abnormal monitoring data can be directly input into the trained deep learning model, or the abnormal monitoring data can be preprocessed and then input into the trained deep learning model, or the abnormal monitoring data can be jointly input into the deep learning model together with other types of abnormal monitoring data and the normal monitoring data at the target object. In the deep learning model, the linkage plan at the target object can be predicted based on the input monitoring data, and the predicted linkage plan can be obtained. Here, the predicted linkage plan can include one or more linkage plans, and each linkage plan can be recorded as a candidate linkage plan. Each candidate linkage plan can include multiple linked actions, recorded as the first candidate actions. Each first candidate action includes an adjustment action for adjusting the abnormal situation at the target object. Through a series of linked first candidate actions in each candidate linkage plan, the abnormal situation at the target object can be improved, so that the target object can return to normal as soon as possible. For example, taking the platform environment as an example, a temperature sensor in the middle of a certain platform detects that the temperature has risen from 28°C to 32°C within 10 minutes, and the temperature has risen by 4 degrees within 10 minutes, exceeding the preset data change threshold, there is an abnormality. Then, the linkage plan at the platform can be predicted through the trained deep learning model, which includes multiple linked actions, such as turning on the air conditioner at the platform for cooling, turning on / adjusting the wind speed and direction of the ventilation system, etc.

[0040] Step 108, determine the target linkage plan corresponding to the target object according to at least one candidate linkage plan of the target object; the above target linkage plan includes multiple linked target actions and the control parameters of each target action.

[0041] In this step, after obtaining at least one candidate linkage plan at the target object, if only one candidate linkage plan is obtained through the deep learning model, the candidate linkage plan can be directly used as the target linkage plan at the target object. If multiple candidate linkage plans are obtained through the deep learning model, one target linkage plan can be selected from the multiple candidate linkage plans according to the order of the identifiers (such as numbers) of the candidate linkage plans given by the deep learning model, the probabilities of the candidate linkage plans given, etc., or one target linkage plan can also be selected from the multiple candidate linkage plans by comprehensively analyzing each first candidate action in each candidate linkage plan and based on the comprehensive analysis results of the candidate linkage plans.

[0042] It can be understood that a selected candidate linkage solution can be recorded as the target linkage solution, and each first candidate action in the candidate linkage solution is also recorded as the target action. In other words, the target linkage solution may include multiple linked target actions. At the same time, it may also include the control parameters of each target action. Here, the control parameters may be, for example, the specific start time and cooling temperature when controlling the air conditioner to start cooling, the specific wind direction and wind speed when controlling the ventilation to start, etc.

[0043] Further, to facilitate the implementation of the technical solution of the present invention, a linkage solution prediction system for rail transit can be pre-built. This system includes three-layer architecture systems: a data acquisition layer, a data analysis and association layer, and a decision-making and linkage execution layer. The data acquisition layer can be used to collect monitoring data, that is, the above-mentioned step 102 can be executed, and the collected monitoring data is transmitted to the data analysis and association layer through a wired and wireless hybrid data transmission network (such as a combination of low-power Bluetooth, Wi-Fi, and wired Ethernet to ensure the stability and timeliness of data transmission). The data analysis and association layer can use big data storage and processing technologies to build a big data platform based on a distributed file system (such as Ceph, GlusterFS, etc.) and a distributed computing framework (such as Apache Flink, Spark, etc.), gather a large amount of monitoring data from the data acquisition layer, and perform data detection, analysis, and linkage solution prediction processing on the monitoring data, that is, the above-mentioned steps 104 and 106 can be executed. When receiving a real-time data change signal from the data acquisition layer, based on the constructed and trained deep learning model, a series of possible linkage actions and linkage solutions can be quickly and automatically associated. The decision-making and linkage execution layer is responsible for receiving the association recommendation results from the data analysis and association layer, and can make a decision on the candidate linkage solutions predicted by the deep learning model of the data analysis and association layer to select the target linkage solution, that is, the above-mentioned step 108 can be executed. Further, after the decision-making and linkage execution layer selects the target linkage solution, it can also execute the target linkage solution. Specifically, it can immediately trigger the corresponding systems and devices for linkage operations, such as controlling the air conditioning system, ventilation system, lighting system, train operation control system, broadcast system, turnstile system, etc., to achieve intelligent rail transit operation management. The above-mentioned linkage operation execution process can be achieved through industrial automation control protocols (such as Modbus, OPC UA / OPC Unified Architecture, etc.) to accurately control the devices or systems, ensuring the accuracy and timeliness of the execution of the target actions in the target linkage solution.

[0044] In this embodiment, by obtaining first monitoring data collected by a first sensor at a target object including a target area and / or a target vehicle in rail transit, performing anomaly detection on the first monitoring data, if it is detected that the first monitoring data is abnormal, obtaining abnormal monitoring data with anomalies in the first monitoring data, and inputting the abnormal monitoring data into a pre-trained deep learning model for linkage plan prediction processing to determine at least one candidate linkage plan corresponding to the target object, and then determining a target linkage plan of the target object according to at least one candidate linkage plan of the target object; wherein, the deep learning model is pre-trained by supervised training and / or unsupervised training according to a plurality of sample linkage plans and known working conditions corresponding to each sample linkage plan, each candidate linkage plan includes a plurality of linked first candidate actions, each first candidate action corresponds to an adjustment action for adjusting the abnormal situation existing in the target object, and the target linkage plan includes a plurality of linked target actions and control parameters of each target action. In this method, since the deep learning model pre-trained by a combination of supervised and unsupervised methods can be used to predict the linkage plan for the monitoring data at the rail transit target object, compared with the adjustment plan determined by a fixed correspondence, the linkage plan predicted by the technical solution of the present invention is more intelligent and reasonable, so that the intelligence of the predicted linkage plan can be improved and the linkage of the rail transit system can be improved; at the same time, since the data at the rail transit target object can be monitored for anomalies, and when the monitoring data is abnormal, the pre-trained deep learning model is used to predict the linkage plan, the efficiency of predicting the linkage plan is higher, so that the rail transit system can respond to abnormal situations in a timely and rapid manner, ensuring the operation safety of rail transit and improving the user experience of passengers.

[0045] The following embodiments illustrate an implementation manner of selecting a target linkage plan from multiple candidate linkage plans by comprehensively analyzing each candidate action in each candidate linkage plan and based on the comprehensive analysis results of each candidate linkage plan.

[0046] Figure 2 It is the second flowchart of the linkage plan prediction method for rail transit provided by the present invention. As Figure 2 shown, in step 108 above, determining the target linkage plan corresponding to the target object according to at least one candidate linkage plan of the target object may include the following steps: Step 202, determining an initial quantization value corresponding to each first candidate action in each candidate linkage plan according to the priority corresponding to each first candidate action in each candidate linkage plan; the priorities corresponding to each first candidate action in the candidate linkage plan are not completely the same.

[0047] In this step, at least one candidate linkage plan includes multiple candidate linkage plans. Each candidate linkage plan may include multiple first candidate actions for linkage. Each first candidate action can determine its priority according to factors such as rail transit operation safety, rail transit operation efficiency, and rail transit passenger experience. Among them, the priority of rail transit operation safety is higher than that of rail transit operation efficiency, and the priority of rail transit operation efficiency is higher than that of rail transit passenger experience. Generally, each first candidate action determines its priority according to the factor with the highest priority it involves. The priorities corresponding to the first candidate actions in each candidate linkage plan are not completely the same and may be partially the same or completely different. For example, if a first candidate action involves rail transit operation safety, its priority is determined as the corresponding priority of rail transit operation safety, which is the highest priority. Another example is that if a first candidate action only involves rail transit passenger experience, its priority is determined as the corresponding priority of rail transit passenger experience, which is the lowest priority.

[0048] For the above factors such as rail transit operation safety, rail transit operation efficiency, and rail transit passenger experience, these factors can be pre - processed by weight quantization. The sum of the weights corresponding to these factors is 1, and then the weights are allocated in turn according to the quantity and priority of each factor. For example, the weight corresponding to rail transit operation safety is 0.5, the weight corresponding to rail transit operation efficiency is 0.3, and the weight corresponding to rail transit passenger experience is 0.2.

[0049] After determining the priority of each first candidate action in each candidate linkage plan, the first candidate action can be quantized according to the weight of the factor corresponding to its priority to determine its corresponding initial quantization value, that is, the weight of the factor corresponding to the priority of each first candidate action is used as its initial quantization value.

[0050] Step 204: According to the initial quantization values corresponding to the first candidate actions in each candidate linkage plan, perform quantization processing on each candidate linkage plan to determine the target quantization value corresponding to each candidate linkage plan.

[0051] Among them, after obtaining the initial quantization values corresponding to the first candidate actions in each candidate linkage plan, for each candidate linkage plan, the initial quantization values of the multiple first candidate actions it includes can be summed up, and the obtained sum value can be used as the target quantization value corresponding to this candidate linkage plan. In this way, the target quantization value corresponding to each candidate linkage plan can be obtained.

[0052] Step 206: Determine the target linkage plan among the candidate linkage plans according to the target quantization value corresponding to each candidate linkage plan.

[0053] After obtaining the target quantization value corresponding to each candidate linkage solution, the target quantization values can be directly sorted, for example, sorted from largest to smallest, and the candidate linkage solution corresponding to the first target quantization value (i.e., the largest target quantization value) is used as the target linkage solution.

[0054] Optionally, after obtaining the target quantization value corresponding to each candidate linkage solution, the target quantization value corresponding to each candidate linkage solution can also be combined with the current state of the rail transit system to comprehensively determine the target linkage solution among the candidate linkage solutions. The current state of the rail transit system can, for example, include whether it is currently running, whether the current energy consumption is too high, etc. By combining the current state of the rail transit system to determine the target linkage solution, the determined linkage solution is more in line with the current actual situation of the rail transit system, and thus the subsequent control of the system is more accurate.

[0055] In this embodiment, the quantization values of each first candidate action in the candidate linkage solution are determined through the priorities of the first candidate actions, and then the quantization value of each candidate linkage solution is determined and the target linkage solution is decided. In this way, by comprehensively analyzing each candidate linkage solution through the priorities of the candidate actions in the candidate linkage solution, the analysis of each candidate linkage solution can be made more comprehensive and reasonable, and the finally decided target linkage solution is more reasonable and accurate, thereby improving the execution effect of the subsequent linkage solution.

[0056] In some embodiments, after obtaining multiple candidate linkage solutions, in order to make the target linkage solution for decision-making / screening more in line with the actual needs of users, optionally, step 108 of determining the target linkage solution corresponding to the target object according to at least one candidate linkage solution of the target object may include the following steps: Display at least one candidate linkage solution; Obtain the selection operation input by the user among at least one candidate linkage solution; In response to the selection operation, determine the candidate linkage solution corresponding to the selection operation as the target linkage solution.

[0057] Among them, after obtaining multiple candidate linkage solutions, each candidate linkage solution can be output to the device interface of the rail transit system for display for the user to view in a timely manner. Then the user can input a selection operation to select a candidate linkage solution among these multiple candidate linkage solutions. Here, the selection operation can be implemented through the selection control provided on the device interface (such as a physical button, a touch button, etc.), or can also be implemented through the user's voice input. After the rail transit system receives the user's selection operation, it can know the specific candidate linkage solution selected by the user, and then can respond to the selection operation group and use the specific candidate linkage solution selected by the user as the target linkage solution.

[0058] In this embodiment, by displaying multiple candidate linkage solutions on the device interface for the user to select the target linkage solution, the finally determined target linkage solution can better meet the actual needs of the user.

[0059] As the rail transit system operates continuously, new monitoring data will be continuously generated, and the performance of the equipment in the rail transit system will also change. In order to more accurately control the equipment of the rail transit system based on the target linkage solution, this embodiment proposes a technical solution that can further adjust the target linkage solution based on the performance data of the equipment. The following embodiments will illustrate this process.

[0060] In some embodiments, the above method may further include the following steps: Obtain the performance data of the target object and / or the performance data of the target equipment in the target object; Adjust the control parameters of at least one target linkage action in the target linkage solution according to the performance data of the target object and / or the performance data of the target equipment.

[0061] Wherein, if the target object is a target area, such as a platform, the performance data of the target equipment in the platform after running for a period of time can be obtained. The target equipment can be, for example, an air conditioner, a ventilation system, etc. The performance data of the air conditioner can be, for example, the refrigeration effect data and the heating effect data after the air conditioner has run for a period of time. The performance data of the ventilation system can be, for example, the wind speed and wind direction data after the ventilation system has run for a period of time. If the target object is a target vehicle, the performance data of the target vehicle after running for a period of time can be obtained, such as the door closing time, the number of faults, the braking distance, the starting time, and so on.

[0062] After obtaining the performance data of the target object after running for a period of time and / or the performance data of the target equipment in the target object after running for a period of time, if the target linkage solution includes a target action for adjusting the target equipment or the target vehicle at the target object, the control parameters of the target action can be adjusted through the performance data. For example, the value of the control parameter can be adjusted to be more in line with the actual situation, so that the finally adjusted target action can improve the abnormal situation at the target object faster and more accurately.

[0063] Optionally, after obtaining the performance data of the target object after running for a period of time and / or the performance data of the target equipment in the target object after running for a period of time, the abnormal conditions in the conditional trigger mechanism in the above embodiment can also be adjusted according to these performance data. For example, the original abnormal condition is to trigger the linkage solution prediction function when the temperature exceeds 30 degrees. Here, the abnormal condition can be adjusted to trigger the linkage solution prediction function when the temperature exceeds 28 degrees through the performance data.

[0064] For example, during the high-temperature period in summer, after accumulating data for a period of time, the rail transit system finds that the air-conditioning cooling effect at a certain platform is slightly insufficient during peak hours. Through the linkage scheme of automatic adjustment and association, for example, the frequency of ventilation-assisted cooling can be increased, the trigger condition / anomaly condition of ventilation-assisted cooling can be adjusted from the original temperature higher than 30 degrees to higher than 28 degrees, and the time to turn on the air conditioner can be appropriately advanced, from the original 10 minutes in advance to 15 minutes in advance, to ensure the comfort of passengers.

[0065] In addition, as the rail transit system continues to operate, new operation data will be continuously collected and fed back to data analysis and association. Here, an online learning algorithm (such as an online learning algorithm based on stochastic gradient descent, combined with a mini-batch gradient update strategy, and the deep learning model is updated once every certain number (such as 1000) of new data is collected) can be used to perform real-time update and optimization of the deep learning model, so that it can adapt to the long-term operation changes of the rail transit system, such as seasonal passenger flow fluctuations, performance changes caused by equipment aging, etc., and improve the self-optimization and adaptive adjustment ability of the rail transit system.

[0066] In this embodiment, the target linkage scheme is further adjusted through the performance data of the target object and the performance data of the target device at the target object, so that the finally adjusted target linkage scheme can improve the abnormal situation at the target object faster and more accurately, and can further improve the intelligence of the linkage scheme and the intelligence of controlling the rail transit system.

[0067] In the above embodiment, it is mentioned that multiple types of sensors can be set at the target object of the rail transit to collect data. The following embodiment will describe the content of the data collected by multiple types of sensors participating in the prediction process of the linkage scheme.

[0068] In some embodiments, the above method may further include the following steps: Obtain the second monitoring data collected by the second sensor at the target object; Correspondingly, in step 106 above, inputting the anomaly monitoring data into a preset deep learning model for linkage scheme prediction processing to determine at least one candidate linkage scheme corresponding to the target object may include the following steps: Input the anomaly monitoring data and the second monitoring data into a preset deep learning model for linkage scheme prediction processing to determine at least one candidate linkage scheme corresponding to the target object; each of the above candidate linkage schemes includes various first candidate actions and second candidate actions, and the above second candidate actions are used to perform energy-saving actions on the energy-consuming devices in the target object.

[0069] The second sensor is a sensor of a different type from the first sensor, or a sensor disposed at the target object. The second sensor may include multiple second sensors, which may be sensors of the same or different types, such as but not limited to temperature sensors, humidity sensors, light sensors, passenger flow counters, equipment status monitors, smoke sensors, position sensors, etc.

[0070] The second sensor at the target object can also monitor or collect data of the corresponding type at the target object in real time to obtain second monitoring data. The second monitoring data is monitoring data corresponding to the type of the second sensor, and its type may be different from the first monitoring data. For example, the first monitoring data is temperature data, and the second monitoring data may be humidity data, passenger flow counting data, etc.

[0071] Specifically, after obtaining the second monitoring data, the second monitoring data and the abnormal monitoring data obtained from the first monitoring data can be input into the pre-trained deep learning model together, and the two types of data are fused in the deep learning model, and the linkage scheme at the target object is predicted by the fused data, and finally a candidate linkage scheme at the target object is obtained. The candidate linkage scheme may include not only the above-mentioned first candidate action, but also a second candidate action linked with the first candidate action, wherein the first candidate action is mainly used to improve the abnormal situation at the target object, and the second candidate action is mainly to perform energy-saving actions on energy-related equipment in the target object, such as reducing the brightness of lighting in areas with less passenger flow in the target object to save energy.

[0072] For example, taking the platform environment control as an example, on a hot summer afternoon, the temperature sensor in the middle of the platform detected that the temperature rose from 28 degrees to 32 degrees within 10 minutes, triggering the associative recommendation mechanism of the rail transit system. Based on real-time data and the trained deep learning model, the data analysis and association layer quickly associates the central air conditioning on both sides of the platform with cooling, and recommends adjusting the wind speed to medium and the wind direction to blow horizontally toward the middle of the platform. At the same time, combined with the passenger flow monitored by the current passenger flow counter, it is judged that there are fewer passengers at both ends of the platform, and the lighting brightness of these two areas is automatically reduced to save energy. After receiving the recommended solution, the decision-making and linkage execution layer immediately sends instructions to the air conditioning system, ventilation system, and lighting system to realize the intelligent control of the platform environment, provide passengers with a comfortable waiting environment, and achieve the purpose of energy saving.

[0073] In this embodiment, the monitoring data collected by multiple types of second sensors at the target object and the abnormal monitoring data obtained from the data collected by the first sensor are jointly input into a deep learning model for linkage scheme prediction processing, so as to obtain actions including energy saving at the target object and actions to improve the abnormal conditions at the target object. In this way, energy consumption can be saved while improving the abnormal conditions at the target object, further enhancing the intelligence of the predicted linkage scheme.

[0074] In the above embodiment, it is mentioned that multiple types of sensors can be set at the target object of rail transit to collect data. At the same time, the rail transit system is still running and will execute the predicted linkage scheme to obtain the execution result. The following embodiment will illustrate the content in which the data collected by multiple types of sensors and the execution result obtained after executing the predicted linkage scheme are jointly involved in the subsequent linkage scheme prediction process.

[0075] In some embodiments, the above method may further include the following steps: Obtain the execution result obtained after the device at the target object executes the first candidate action at the historical time, and obtain the third monitoring data collected by the third sensor at the target object; Correspondingly, in step 106 above, inputting the abnormal monitoring data into a preset deep learning model for linkage scheme prediction processing to determine at least one candidate linkage scheme corresponding to the target object may include the following steps: Input the abnormal monitoring data, the execution result, and the third monitoring data into a preset deep learning model for linkage scheme prediction processing to determine at least one candidate linkage scheme corresponding to the target object.

[0076] Among them, the third sensor is a sensor of a different type from the first sensor and the second sensor, and can also be a sensor set at the target object. The third sensor here may include multiple third sensors, and the multiple third sensors may be sensors of the same type or different types. For example, it may include but is not limited to temperature sensors, humidity sensors, light sensors, passenger flow counters, device status monitors, smoke sensors, position sensors, etc.

[0077] The third sensor at the target object can also monitor or collect the corresponding type of data at the target object in real time to obtain the third monitoring data. The third monitoring data is the monitoring data corresponding to the type of the third sensor, and its type may be different from that of the first monitoring data and the second monitoring data. For example, the first monitoring data is temperature data, the second monitoring data may be humidity data, passenger flow counting data, etc., and the third monitoring data may be the position data of the target area in the target object collected.

[0078] Specifically, in the historical time before the current time, the data analysis and association layer deployed in the rail transit system can continuously predict candidate linkage schemes and output them to the decision-making and linkage execution layer for linkage scheme decision-making and target linkage scheme execution processing. After the decision-making and linkage execution layer executes the target linkage scheme for each decision, corresponding execution results will be obtained. In this way, the execution results after the execution of each target linkage scheme in the historical time can be obtained. Then, the above-mentioned third monitoring data, the abnormal monitoring data obtained from the first monitoring data, and the execution results after the execution of the target linkage scheme in the historical time are input into a pre-trained deep learning model together. In the deep learning model, these three types of data are fused, and the linkage scheme at the target object is comprehensively predicted through the fused data, and finally the candidate linkage scheme at the target object is obtained. The candidate linkage scheme may include each first candidate action, and each first candidate action is mainly used to improve the abnormal situation at the target object. These first candidate actions are actions determined by comprehensively considering the execution results of the historical target linkage scheme. For example, they may be actions after further adjusting their control parameters on the basis of the first candidate actions in the above step 106, so they are more accurate and meet the actual scenario requirements.

[0079] Exemplarily, taking the platform environment as an example, when the temperature sensor in a certain area detects that the temperature change exceeds the set threshold (such as rising by 3 degrees within 5 minutes), the rail transit system takes this as a trigger condition. At this time, based on data such as historical execution results and the deep learning model, the data analysis and association layer can not only quickly think of turning on the air conditioning for cooling in this area, but also recommend synchronously adjusting the wind speed and direction of the ventilation system. Specifically, if this area is close to the platform edge and past data analysis shows that the ventilation here has a greater impact on the overall comfort, it is recommended to increase the ventilation volume by 20% on the original basis and adjust the wind direction to face the center of the platform to accelerate air circulation and quickly balance the platform temperature to avoid discomfort for passengers due to stuffiness.

[0080] In this embodiment, by jointly inputting the data collected by multiple types of sensors at the target object, the abnormal monitoring data, and the execution results obtained after executing the target linkage scheme in the historical time into the deep learning model for linkage scheme prediction processing, a candidate linkage scheme for further adjusting the first candidate action is obtained. In this way, the candidate linkage scheme can be dynamically adjusted, further improving the intelligence and accuracy of the predicted linkage scheme.

[0081] The training process of the deep learning model was briefly described in the above embodiment. The following embodiment will describe the specific training process of the deep learning model.

[0082] Figure 3 is the third flow chart of the linkage scheme prediction method for rail transit provided by the present invention, as Figure 3As shown, the training method of the above deep learning model may include the following steps: Step 302: Obtain the sample linkage schemes of multiple sample objects and the known working conditions corresponding to each sample linkage scheme; the known working conditions include the sample abnormal monitoring data at the sample objects.

[0083] Among them, the sample objects are similar to the above target objects and may include sample areas and / or sample vehicles. The sample area can be a relatively large area such as a platform, or a relatively small area in the platform, etc. The known working conditions refer to the known environmental parameters or known vehicle operation parameters at the sample objects. For example, they can be the platform temperature, humidity, passenger flow data, equipment operation logs, environmental monitoring records, the number of vehicle failures, the door closing duration of the vehicle, etc. at historical times. In addition, these known environmental parameters or known vehicle operation parameters can be data under abnormal conditions. For example, they can be the temperature data when the temperature is relatively high, the passenger flow data when the passenger flow exceeds the passenger flow threshold, the failure data when the number of vehicle failures exceeds the failure threshold, the door closing data of the vehicle when the door closing duration of the vehicle exceeds the set duration, etc.

[0084] Specifically, a large amount of operation data accumulated by the rail transit system over a long time can be collected to obtain multiple known working conditions at historical times, and the optimal linkage schemes adopted under these known working conditions are collected, all of which are recorded as sample linkage schemes, and then each known working condition is bound to its corresponding sample linkage scheme.

[0085] Step 304: Input each known working condition into the initial deep learning model for linkage scheme prediction processing to determine the predicted linkage scheme corresponding to the sample object of each known working condition.

[0086] In this step, after obtaining multiple known working conditions and their corresponding sample linkage schemes, the relevant data of each known working condition can be input into the initial deep learning model (i.e., the untrained deep learning model). The initial deep learning model will predict the corresponding linkage scheme based on the data of the known working conditions, and then the optimal one can be selected as the predicted linkage scheme, or the linkage schemes output by the initial deep learning model can be used as the predicted linkage schemes.

[0087] Step 306: Calculate the loss between each predicted linkage scheme and the corresponding sample linkage scheme.

[0088] In this step, after obtaining the predicted linkage scheme corresponding to each known working condition, the predicted linkage scheme and its corresponding sample linkage scheme can be input into a loss function to calculate the loss, and the loss corresponding to each known working condition can be obtained. The loss function here can be set according to the actual situation. For example, it can be a loss function that calculates the difference between the predicted linkage scheme of the known working condition and its corresponding sample linkage scheme, such as loss functions like error, variance, mean square error, etc.

[0089] Step 308: Perform supervised training on the initial deep learning model according to each loss, and / or perform unsupervised training on the initial deep learning model according to each known working condition to obtain a deep learning model; the above-mentioned supervised training is used to learn the mapping relationship between the known working condition and the corresponding sample linkage scheme, and the above-mentioned unsupervised training is used to explore the internal correlation relationship between various types of data in the known working condition.

[0090] In this step, after obtaining the loss between the predicted linkage scheme and the sample linkage scheme of each known working condition, the parameters of the initial deep learning model can be adjusted through these losses to perform supervised training on the initial deep learning model. For example, for the linkage scheme of turning on the air conditioner and ventilation when the platform is at a high temperature, let the initial deep learning model learn the mapping relationship between the known working condition and the corresponding sample linkage scheme.

[0091] At the same time, after inputting each known working condition into the initial deep learning model, the initial deep learning model can also perform unsupervised learning. Unsupervised learning focuses on exploring hidden patterns and correlations in the data. Specifically, it can be to let the initial deep learning model perform in-depth analysis on various types of data included in the known working condition to explore the internal correlation relationship between various types of data in the known working condition. For example, learn the relationship between passenger flow at different times and platform temperature and humidity, equipment energy consumption, train departure intervals, learn the logical relationship between equipment failure modes and corresponding countermeasures, and learn / discover the potential relationship between passenger behavior preferences and equipment energy consumption under different passenger flow densities through clustering analysis, etc.

[0092] By combining the above-mentioned supervised training and unsupervised training to repeatedly train the initial deep learning model, a trained deep learning model can finally be obtained.

[0093] In addition, optionally, the relevant data of the above-mentioned known working conditions can be preprocessed before being input into the initial deep learning model, including data cleaning (removing outliers and duplicate values), normalization (unifying data of different magnitudes to the same range, such as using the Min-Max normalization method), feature engineering (extracting key features, such as extracting peak-hour and off-peak-hour features from passenger flow data), etc., to improve the training efficiency and accuracy of the model.

[0094] In this embodiment, the initial deep learning model is subjected to supervised training and unsupervised training through multiple known working conditions and their corresponding sample linkage schemes. Such a training method that combines supervised training and unsupervised training, and the model is repeatedly trained and optimized, can enable the finally trained deep learning model to have a powerful association ability, and can quickly and accurately recommend a series of reasonable linkage schemes when facing new real-time data changes, that is, it can improve the intelligence and accuracy of the predicted linkage scheme of the trained deep learning model.

[0095] To facilitate a detailed description of the technical solutions of the embodiments of the present invention, the following gives several scenario application examples of the embodiments of the present invention.

[0096] Scenario 1: Platform environment regulation In a typical platform of a certain urban rail transit, a three-layer architecture system including a data acquisition layer, a data analysis and association layer, and a decision-making and linkage execution layer of the embodiments of the present invention is deployed. On a hot summer afternoon, the temperature sensor in the middle of the platform detects that the temperature rises from 28 degrees to 32 degrees within 10 minutes, triggering the association recommendation function / mechanism of the rail transit system. The data analysis and association layer quickly associates turning on the central air-conditioning for cooling on both sides of the platform according to the real-time data and the trained deep learning model, and recommends adjusting the wind speed to medium and the wind direction to horizontally blow towards the middle of the platform. At the same time, in combination with the current passenger flow situation, it is judged that there are fewer passengers at both ends of the platform, and the lighting brightness in these two areas is automatically reduced to save energy. After receiving the recommended scheme, the decision-making and linkage execution layer immediately sends instructions to the air-conditioning system, ventilation system and lighting system to realize the intelligent regulation of the platform environment and provide a comfortable waiting environment for passengers.

[0097] Scenario 2: Response to train equipment failures For a train in operation, the door system suddenly issues a fault warning. The door closing time is extended to 3 seconds and it reports errors continuously for 4 times. After the rail transit system detects this situation, it immediately starts the Lenovo recommendation process. On the one hand, it automatically reduces the train running speed from 80 km / h to 60 km / h and issues a warning to the train driver. The warning information is presented in the form of sound and light alarms through the train internal communication system. On the other hand, it sends detailed maintenance tips to the control center, listing in detail the possible faulty components, such as the information of the door lock sensor, drive motor, controller and other components that may be faulty. The maintenance tip information is sent in the form of standardized fault codes and description texts. On the third hand, it links with the spare vehicle dispatching system. According to the current line passenger flow and operation timetable, it arranges a spare train to wait at the forward station in advance to be ready to replace the faulty train. When the faulty train arrives at the forward station, passengers quickly and orderly transfer to the spare train to ensure the continuity of operation and the safety of passengers. The spare vehicle dispatching process is as follows: First, query the location and status information of the spare vehicle, screen out the eligible spare vehicles from the vehicle management database, then determine the optimal driving route for the spare vehicle to the station where the faulty train is located according to the line planning algorithm (such as Dijkstra algorithm, calculating the optimal path by combining the line real-time passenger flow and running time cost), and finally issue a dispatching instruction to the driver of the spare vehicle to ensure that the overall operation is not greatly affected.

[0098] Scenario 3: Passenger flow peak diversion During the morning and evening rush hours on weekdays, the passenger flow monitoring equipment / sensors at the entrance of a rail transit station detect that the number of inbound passengers increases sharply within 15 minutes, reaching the set peak threshold. The rail transit system quickly starts the Lenovo recommendation mechanism. In addition to the regular opening of all turnstile channels, it also associates with the real-time passenger flow distribution on the surrounding platforms. After analysis, it is found that the passenger flow on a certain platform of an adjacent line is less, and it is recommended to adjust the departure interval of the trains on the adjacent line from the original 5 minutes to 3 minutes to increase the transport capacity. At the same time, the in-station broadcast system automatically switches to the high-frequency broadcast mode, providing passengers with real-time guidance information such as the most convenient transfer routes and waiting positions, and cooperating with the linked changes of the dynamic indicator signs to efficiently divert passengers to each waiting area, thereby effectively alleviating the platform congestion and improving the passenger passage efficiency.

[0099] Based on the descriptions of the above embodiments, the technical invention of the present invention has the following beneficial effects: (1) Improve operation efficiency: Through the intelligent Lenovo recommendation linkage plan, it can quickly respond to emergencies such as equipment failures and passenger flow peaks, rationally allocate resources, reduce train delays, and improve the overall operation efficiency of rail transit. For example, during the passenger flow peak, accurately adjust the train departure interval and the number of turnstile channels opened, effectively shortening the passenger waiting and inbound time.

[0100] (2) Ensure system safety: When the equipment fails or the environment becomes abnormal (such as a smoke alarm), the system can quickly associate and execute a series of safety guarantee measures to reduce the accident risk. For example, when the door fails, the speed is automatically reduced and a spare vehicle is allocated; when smoke appears, the fire sprinkler is activated in time and the passengers are evacuated to ensure the safety of personnel and facilities.

[0101] (3) Optimize the passenger experience: Provide a comfortable and convenient travel environment according to the real-time needs of passengers and environmental changes. For example, according to the platform temperature and passenger flow, the ventilation, lighting and other facilities are automatically adjusted to create a pleasant waiting environment for passengers; and accurate route guidance is provided during transfer to improve the passenger experience and satisfaction.

[0102] In summary, the technical solution of the present invention brings a new intelligent solution to the rail transit field through the innovative three-layer architecture design, powerful association and recommendation function, and adaptive optimization mechanism, effectively improving the operation efficiency of rail transit, ensuring the safety of system operation, and optimizing the passenger experience.

[0103] The following describes the linkage plan prediction device for rail transit provided by the present invention. The linkage plan prediction device for rail transit described below can be mutually corresponding and referred to the rail transit linkage plan prediction method described above.

[0104] Figure 4 is a schematic structural diagram of the linkage plan prediction device for rail transit provided by the present invention. Refer to Figure 4 As shown, the device may include: The first monitoring data acquisition module 410 is used to acquire the first monitoring data collected by the first sensor at the target object in the rail transit; the above target object includes the target area and / or the target vehicle; The anomaly detection module 420 is used to perform anomaly detection on the first monitoring data. If it is detected that the first monitoring data is abnormal, the abnormal monitoring data with anomalies in the first monitoring data is acquired; The linkage plan prediction module 430 is used to input the abnormal monitoring data into a preset deep learning model for linkage plan prediction processing to determine at least one candidate linkage plan corresponding to the target object; each candidate linkage plan includes a plurality of linked first candidate actions, and each first candidate action corresponds to an adjustment action for adjusting the abnormal situation existing in the target object. The above deep learning model is pre-trained in a supervised manner according to a plurality of sample linkage plans and the known working conditions corresponding to each sample linkage plan and / or trained in an unsupervised manner according to a plurality of known working conditions; The target linkage plan determination module 440 is used to determine the target linkage plan corresponding to the target object according to at least one candidate linkage plan of the target object; the above target linkage plan includes a plurality of linked target actions and the control parameters of each target action.

[0105] In some embodiments, the above-mentioned target linkage scheme determination module 440 is specifically configured to determine the initial quantization value corresponding to each first candidate action in each candidate linkage scheme according to the priority levels of the first candidate actions in each candidate linkage scheme; the priority levels of the first candidate actions in each candidate linkage scheme are not completely the same; Perform quantization processing on each candidate linkage scheme according to the initial quantization value corresponding to each first candidate action in each candidate linkage scheme, and determine the target quantization value corresponding to each candidate linkage scheme; Determine the target linkage scheme from the candidate linkage schemes according to the target quantization value corresponding to each candidate linkage scheme.

[0106] In some embodiments, the above-mentioned target linkage scheme determination module 440 is specifically configured to display at least one candidate linkage scheme; obtain the selection operation input by the user in the at least one candidate linkage scheme; in response to the selection operation, determine the candidate linkage scheme corresponding to the selection operation as the target linkage scheme.

[0107] In some embodiments, the above-mentioned device further includes: A performance data acquisition module, configured to acquire the performance data of the target object and / or the performance data of the target device in the target object; A scheme adjustment module, configured to adjust the control parameters of at least one target linkage action in the target linkage scheme according to the performance data of the target object and / or the performance data of the target device.

[0108] In some embodiments, the above-mentioned device further includes a training module, which is used to train the deep learning model. The training module is specifically configured to acquire the sample linkage schemes of multiple sample objects and the known working conditions corresponding to each sample linkage scheme; the known working conditions include the sample anomaly monitoring data at the sample objects; input each known working condition into the initial deep learning model for linkage scheme prediction processing, and determine the predicted linkage scheme corresponding to the sample object corresponding to each known working condition; calculate the loss between each predicted linkage scheme and the corresponding sample linkage scheme; perform supervised training on the initial deep learning model according to each loss, and / or perform unsupervised training on the initial deep learning model according to each known working condition to obtain the deep learning model; the supervised training is used to learn the mapping relationship between the known working conditions and the corresponding sample linkage schemes, and the unsupervised training is used to explore the internal correlation relationships between various types of data in the known working conditions.

[0109] In some embodiments, the above-mentioned device further includes: A second monitoring data acquisition module, configured to acquire the second monitoring data collected by the second sensor at the target object; The above-mentioned linkage scheme prediction module 430 is specifically configured to input the anomaly monitoring data and the second monitoring data into a preset deep learning model for linkage scheme prediction processing to determine at least one candidate linkage scheme corresponding to the target object; each of the candidate linkage schemes includes a first candidate action and a second candidate action, and the second candidate action is used to perform an energy-saving action on the energy consumption-related equipment in the target object.

[0110] In some embodiments, the above-mentioned device further includes: A third monitoring data acquisition module, configured to acquire the execution result obtained after the device at the target object executes the first candidate action at a historical time, and acquire the third monitoring data collected by a third sensor at the target object; The above-mentioned linkage scheme prediction module 430 is specifically configured to input the anomaly monitoring data, the execution result, and the third monitoring data into a preset deep learning model for linkage scheme prediction processing to determine at least one candidate linkage scheme corresponding to the target object.

[0111] It should be noted here that the above-mentioned device provided in the embodiment of the present invention can implement all the method steps implemented in the above-mentioned method embodiment and can achieve the same technical effect. The same parts and beneficial effects as those in the method embodiment will not be specifically described in this embodiment.

[0112] Figure 5 An example of a schematic physical structure diagram of an electronic device is shown in Figure 5As shown in the figure, the electronic device may include: a processor 510, a communications interface 520, a memory 530, and a communication bus 540. Among them, the processor 510, the communications interface 520, and the memory 530 complete communication with each other through the communication bus 540. The processor 510 may call logic instructions in the memory 530 to execute the prediction method for the linkage scheme of rail transit. The method includes: obtaining first monitoring data collected by a first sensor at a target object in rail transit; the above target object includes a target area and / or a target vehicle; performing anomaly detection on the first monitoring data, and if it is detected that the first monitoring data is abnormal, obtaining abnormal monitoring data that is abnormal in the first monitoring data; inputting the abnormal monitoring data into a preset deep learning model for linkage scheme prediction processing to determine at least one candidate linkage scheme corresponding to the target object; each candidate linkage scheme includes a plurality of linked first candidate actions, and each first candidate action corresponds to an adjustment action for adjusting the abnormal situation existing in the target object. The above deep learning model is pre-trained with supervision according to a plurality of sample linkage schemes and the known working conditions corresponding to each sample linkage scheme and / or obtained by unsupervised training according to a plurality of known working conditions; determining a target linkage scheme corresponding to the target object according to at least one candidate linkage scheme of the target object; the above target linkage scheme includes a plurality of linked target actions and control parameters of each target action.

[0113] In addition, when the logic instructions in the above memory 530 can be implemented in the form of software function units and sold or used as an independent product, they can be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present invention. The foregoing storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memories (ROMs, Read-Only Memories), random access memories (RAMs, Random Access Memories), magnetic disks, or optical discs that can store program codes.

[0114] On the other hand, the present invention also provides a computer program product, which includes a computer program. The computer program can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the prediction method for the linkage plan of rail transit provided by the above-mentioned various methods. The method includes: obtaining first monitoring data collected by a first sensor at a target object in rail transit; the above-mentioned target object includes a target area and / or a target vehicle; performing anomaly detection on the first monitoring data. If it is detected that the first monitoring data is abnormal, obtaining the abnormal monitoring data existing in the first monitoring data; inputting the abnormal monitoring data into a preset deep learning model for linkage plan prediction processing to determine at least one candidate linkage plan corresponding to the target object; each candidate linkage plan includes a plurality of linked first candidate actions, and each first candidate action corresponds to an adjustment action for adjusting the abnormal situation existing in the target object. The above-mentioned deep learning model is pre-trained by supervised learning according to a plurality of sample linkage plans and the known working conditions corresponding to each sample linkage plan and / or by unsupervised learning according to a plurality of known working conditions; determining a target linkage plan corresponding to the target object according to at least one candidate linkage plan of the target object; the above-mentioned target linkage plan includes a plurality of linked target actions and the control parameters of each target action.

[0115] On another aspect, the present invention also provides a non-transitory computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, it is implemented to execute the prediction method for the linkage plan of rail transit provided by the above-mentioned various methods. The method includes: obtaining first monitoring data collected by a first sensor at a target object in rail transit; the above-mentioned target object includes a target area and / or a target vehicle; performing anomaly detection on the first monitoring data. If it is detected that the first monitoring data is abnormal, obtaining the abnormal monitoring data existing in the first monitoring data; inputting the abnormal monitoring data into a preset deep learning model for linkage plan prediction processing to determine at least one candidate linkage plan corresponding to the target object; each candidate linkage plan includes a plurality of linked first candidate actions, and each first candidate action corresponds to an adjustment action for adjusting the abnormal situation existing in the target object. The above-mentioned deep learning model is pre-trained by supervised learning according to a plurality of sample linkage plans and the known working conditions corresponding to each sample linkage plan and / or by unsupervised learning according to a plurality of known working conditions; determining a target linkage plan corresponding to the target object according to at least one candidate linkage plan of the target object; the above-mentioned target linkage plan includes a plurality of linked target actions and the control parameters of each target action.

[0116] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment. Those of ordinary skill in the art can understand and implement it without creative efforts.

[0117] Through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solution, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. The computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0118] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments or equivalently replace some of the technical features. These modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A prediction method for the linkage scheme of rail transit, characterized in that, Including: Obtaining first monitoring data collected by a first sensor at a target object in rail transit; The target object includes a target area and / or a target vehicle; Performing anomaly detection on the first monitoring data. If it is detected that the first monitoring data is abnormal, obtaining abnormal monitoring data that is abnormal in the first monitoring data; Inputting the abnormal monitoring data into a preset deep learning model for prediction processing of linkage solutions, and determining at least one candidate linkage solution corresponding to the target object; each candidate linkage solution includes a plurality of linked first candidate actions, and each first candidate action corresponds to an adjustment action for adjusting the abnormal situation existing in the target object. The deep learning model is pre-trained in a supervised manner according to a plurality of sample linkage solutions and the known working conditions corresponding to each sample linkage solution and / or is unsupervised trained according to a plurality of the known working conditions; Determining a target linkage solution corresponding to the target object according to at least one candidate linkage solution of the target object; The target linkage solution includes a plurality of linked target actions and control parameters of each target action.

2. The prediction method for the linkage plan of rail transit according to claim 1, characterized in that, The determining the target linkage solution corresponding to the target object according to at least one candidate linkage solution of the target object includes: Determining an initial quantization value corresponding to each first candidate action in each candidate linkage solution according to the priority corresponding to each first candidate action in each candidate linkage solution; the priorities corresponding to the first candidate actions in each candidate linkage solution are not completely the same; Performing quantization processing on each candidate linkage solution according to the initial quantization value corresponding to each first candidate action in each candidate linkage solution, and determining a target quantization value corresponding to each candidate linkage solution; Determining the target linkage solution in the candidate linkage solutions according to the target quantization value corresponding to each candidate linkage solution.

3. The prediction method for the linkage plan of rail transit according to claim 1, wherein The determining the target linkage solution corresponding to the target object according to at least one candidate linkage solution of the target object includes: Displaying the at least one candidate linkage solution; Obtaining a selection operation input by a user in the at least one candidate linkage solution; In response to the selection operation, determining the candidate linkage solution corresponding to the selection operation as the target linkage solution.

4. The prediction method for the linkage plan of rail transit according to any one of claims 1 to 3, characterized in that, The method further includes: Obtaining performance data of the target object and / or performance data of target devices in the target object; Adjusting control parameters of at least one target linkage action in the target linkage solution according to the performance data of the target object and / or the performance data of the target devices.

5. The prediction method for the linkage plan of rail transit according to claim 1, characterized in that The training method of the deep learning model includes: Obtaining the sample linkage solutions of a plurality of sample objects and the known working conditions corresponding to each sample linkage solution; the known working conditions include sample abnormal monitoring data at the sample objects; Inputting each of the known working conditions into an initial deep learning model for prediction processing of linkage solutions, and determining a predicted linkage solution corresponding to the sample object corresponding to each of the known working conditions; Calculating the loss between each predicted linkage solution and the corresponding sample linkage solution; Supervisedly train the initial deep learning model according to each of the losses, and / or unsupervisedly train the initial deep learning model according to each of the known working conditions to obtain the deep learning model; the supervised training is used to learn the mapping relationship between the known working conditions and the corresponding sample linkage solutions, and the unsupervised training is used to explore the internal correlation relationships between various types of data in the known working conditions.

6. The prediction method for the linkage plan of rail transit according to any one of claims 1 to 3, characterized in that, The method further includes: Obtain second monitoring data collected by a second sensor at the target object. The step of inputting the abnormal monitoring data into a preset deep learning model for linkage solution prediction processing to determine at least one candidate linkage solution corresponding to the target object includes: Input the abnormal monitoring data and the second monitoring data into a preset deep learning model for linkage solution prediction processing to determine at least one candidate linkage solution corresponding to the target object; each candidate linkage solution includes each of the first candidate actions and second candidate actions, and the second candidate actions are used to perform energy-saving actions on energy consumption-related devices in the target object.

7. The prediction method for the linkage plan of rail transit according to any one of claims 1 to 3, characterized in that The method further includes: Obtain the execution result obtained after the device at the target object executes the first candidate action at historical time, and obtain third monitoring data collected by a third sensor at the target object. The step of inputting the abnormal monitoring data into a preset deep learning model for linkage solution prediction processing to determine at least one candidate linkage solution corresponding to the target object includes: Input the abnormal monitoring data, the execution result, and the third monitoring data into a preset deep learning model for linkage solution prediction processing to determine at least one candidate linkage solution corresponding to the target object.

8. A prediction device for the linkage scheme of rail transit, characterized in that, It includes: A first monitoring data acquisition module, configured to acquire first monitoring data collected by a first sensor at a target object in rail transit; The target object includes a target area and / or a target vehicle; An abnormality detection module, configured to perform abnormality detection on the first monitoring data. If it is detected that the first monitoring data is abnormal, obtain the abnormal monitoring data in the first monitoring data that is abnormal; A linkage solution prediction module, configured to input the abnormal monitoring data into a preset deep learning model for linkage solution prediction processing to determine at least one candidate linkage solution corresponding to the target object; each candidate linkage solution includes a plurality of linked first candidate actions, and each first candidate action corresponds to an adjustment action for adjusting the abnormal situation existing in the target object. The deep learning model is pre-obtained by supervised training according to a plurality of sample linkage solutions and the known working conditions corresponding to each sample linkage solution and / or unsupervised training according to a plurality of the known working conditions; A target linkage solution determination module, configured to determine a target linkage solution corresponding to the target object according to at least one candidate linkage solution of the target object; The target linkage solution includes a plurality of linked target actions and control parameters of each target action.

9. An electronic device, comprising a memory, a processor, and a computer program stored on the memory and running on the processor, characterized in that, When the processor executes the computer program, the method for predicting the linkage plan of rail transit according to any one of claims 1 to 7 is implemented.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, the method for predicting the linkage plan of rail transit according to any one of claims 1 to 7 is implemented.

Images