A shared device operation and maintenance management method, system and device based on big data

Abstract

This application relates to the technical field of shared device management, and discloses a method, system and device for operation and maintenance management of shared devices based on big data. The method includes: classifying shared devices into mobile devices and non-mobile devices, and the shared devices collect their own device data and send it to the central server; the central server integrates the device data of mobile devices within the first time period into a first data set, and integrates the device data of non-mobile devices into a second data set; analyzes the first data set based on a first prediction model and analyzes the second data set based on a second prediction model to determine the health status, and classifies the shared devices into healthy devices, critical devices and faulty devices based on the health status; the central server shortens the analysis interval of the device data of critical devices, and generates a personnel inspection path based on the ant colony algorithm and the location information of faulty devices. Through the present invention, the health status of shared devices is tracked at a low cost, so that potential faults can be discovered in time and repaired.

Description

Technical Field

[0001] The present application relates to the technical field of shared devices, and particularly to a method, system and device for operation and maintenance management of shared devices based on big data. Background Art

[0002] In recent years, the sharing model has developed rapidly under the promotion of Internet of Things and mobile Internet technologies. Shared devices have been widely used in fields such as transportation, energy, and public facilities. However, since shared devices need to be rotated and used among multiple users, in order to ensure the maximum efficiency of the devices, it is necessary to schedule and maintain them in a timely manner according to the device usage situation.

[0003] For this reason, in the prior art, a Chinese patent document with the publication number CN119180475A discloses a scheduling method and server for shared vehicles. After the operation and maintenance personnel open the mobile terminal application, they can select scheduling information that is relatively close and has a suitable quantity on the electronic map according to their location, establish a scheduling task, and then go to the appropriate vehicle pulling point to pull the vehicle and deliver it to the vehicle dropping point according to the vehicle pulling point location and the number of vehicles shown on the electronic map, so as to be able to quickly and effectively guide the operation and maintenance personnel to schedule shared vehicles. Another example is a Chinese patent document with the publication number CN118982179A, which discloses a shared device scheduling method, device and electronic device. By showing the geographical locations where the number of surrounding devices reaches the device acquisition condition or the device return condition to the user when the user acquires or returns the device, this method can guide the user to acquire the device from the geographical location that meets the device acquisition condition and return the device to the geographical location that meets the device return condition, thereby reducing the device scheduling cost and improving the device scheduling efficiency.

[0004] In addition to scheduling, it is also necessary to track the health status of shared devices. Currently, the operation and maintenance of shared devices rely on manual inspections or user-initiated repairs, and potential device failures cannot be predicted, resulting in an increase in downtime and maintenance costs, and affecting the user experience. Summary of the Invention

[0005] To solve the problems raised in the above background art, the present application provides a method, system and device for operation and maintenance management of shared devices based on big data.

[0006] To achieve the above invention objective, the present invention proposes a method for operation and maintenance management of shared devices based on big data, including:

[0007] Dividing the shared devices into mobile devices and non-mobile devices, and the shared devices collect their own device data and send it to the central server;

[0008] Based on the historical data of the mobile devices and the non-mobile devices, establish a first prediction model and a second prediction model respectively;

[0009] The central server distinguishes the device types of the shared devices according to the device identifiers in the received device data, and integrates the device data of the mobile devices within the first time period into a first data set, and integrates the device data of the non-mobile devices into a second data set;

[0010] Analyze the first data set based on the first prediction model and analyze the second data set based on the second prediction model to determine the health of the shared devices, and classify the shared devices into healthy devices, critical devices, and faulty devices based on the health;

[0011] The central server shortens the analysis interval of the device data of the critical devices, and generates a personnel inspection path based on the ant colony algorithm and the location information of the faulty devices.

[0012] Furthermore, the steps for the mobile device to send the device data include the following:

[0013] Set a first transmission mode and a second transmission mode in the mobile device. The device data includes the location information of the mobile device. The mobile device sends the device information to the central server based on the first transmission mode in the default state;

[0014] After the central server collects the device data within the second time period, analyze the included location information to determine the high-frequency locations used by the mobile device, deploy relay nodes at the high-frequency locations. The central server combines the location information and the deployment locations of the relay nodes. When it is determined that the mobile device enters the coverage range of the relay node, command the mobile device to use the second transmission mode to send the device data to the relay node, and the relay node adaptively compresses the device data and then sends it to the central server.

[0015] Furthermore, the steps for adaptively compressing the device data include the following:

[0016] Deploy a micro model in the relay node. The micro model is a simplified model of the first prediction model. Based on the micro model and the device data, classify the mobile devices into the healthy devices and the faulty devices, aggregate and compress the device data of the healthy devices received within the third time period into a whole block of data and then send it to the central server, and insert a fault mark into the device data of the faulty devices and then send it to the central server.

[0017] Furthermore, the steps for establishing the first prediction model include the following:

[0018] The historical data includes the location of the mobile device, the corresponding occurrence time, and the fault type. Based on the historical data, the movement characteristics of each user during the use of the mobile device are calculated. The movement characteristics include the average movement speed, the maximum movement speed, the minimum movement speed, the number of braking times, and the usage duration. The first prediction model is established based on a fully connected neural network. The movement characteristics are used as the input features of the first prediction model, and the fault type and occurrence probability are used as the output features. The first prediction model is trained and established by combining the historical data.

[0019] Further, establishing the second prediction model includes the following steps:

[0020] The historical data includes the usage characteristics of the non-mobile device and the fault type. A plurality of the usage characteristics that appear before each fault occurrence are integrated into a fault conduction chain for this fault in chronological order. The fault type is marked in the fault conduction chain, and adjacent and identical usage characteristics in the fault conduction chain are merged to optimize the fault conduction chain into a streamlined conduction chain.

[0021] Starting from the first usage characteristic in the streamlined conduction chain, each streamlined conduction chain is recursively split into multiple sub-conduction chains based on chronological order. Based on the historical data, the association probabilities of various fault types after the appearance of the sub-conduction chains are statistically calculated. The sub-conduction chains with association probabilities greater than the first threshold are selected as the first sensitive conduction chains. After screening the first sensitive conduction chains, the second sensitive conduction chains are obtained. The second sensitive conduction chains, the corresponding fault types, and the association probabilities are used as the training set, and the second prediction model based on a time series neural network is established based on the training set.

[0022] Further, screening the first sensitive conduction chains includes the following steps:

[0023] The first sensitive conduction chains originating from the same streamlined conduction chain are defined as homologous conduction chains. The homologous conduction chains are sorted from large to small based on the length feature, and the difference in the association probabilities of the homologous conduction chains at adjacent positions after sorting is calculated. If the difference is greater than the first threshold, the homologous conduction chain with a longer length is retained. If the difference is less than the second threshold, the homologous conduction chain with a shorter length is retained. The retained homologous conduction chains are continuously compared with the homologous conduction chains at subsequent positions, and this step is repeated until all the first sensitive conduction chains are traversed, and the retained ones are defined as the second sensitive conduction chains.

[0024] Further, calculating the health degree of the non-mobile device includes the following steps:

[0025] Integrate the device data of each user ID in the first dataset into an actual feature chain, input each actual feature chain into the first prediction model, obtain the failure type with the highest correlation probability with the actual feature chain, calculate the weighted average of the correlation probabilities of all the failure types, calculate the health degree based on the correlation probability with the largest average value among them, and output it jointly with the corresponding failure type.

[0026] Further, determining the high-frequency locations includes the following steps:

[0027] Divide the preset area into a fixed number of grid areas, perform spatio-temporal density clustering on the mobile devices based on the locations and corresponding occurrence times of the mobile devices, screen out the dense areas from the grid areas based on the clustering results, and use the grid areas where the dense areas are located as the high-frequency locations.

[0028] A shared device operation and maintenance management system based on big data, used to implement the above-mentioned shared device operation and maintenance management method based on big data, characterized in that

[0029] The terminal module includes shared devices, divides the shared devices into mobile devices and non-mobile devices, and the shared devices collect their own device data and send it to the central server;

[0030] The front-end module is set in the central server. The front-end module respectively establishes a first prediction model and a second prediction model based on the historical data of the mobile devices and the non-mobile devices, distinguishes the device types of the shared devices according to the device identifiers in the received device data, and integrates the device data of the mobile devices within the first time period into a first dataset, and integrates the device data of the non-mobile devices into a second dataset;

[0031] The classification module is set in the central server. The classification module analyzes the first dataset based on the first prediction model and analyzes the second dataset based on the second prediction model to determine the health degree of the shared devices, and divides the shared devices into healthy devices, critical devices, and faulty devices based on the health degree;

[0032] The policy module is set in the central server. The policy module shortens the analysis interval of the device data of the critical devices and generates a personnel inspection path based on the ant colony algorithm and the location information of the faulty devices.

[0033] This application also provides a shared device operation and maintenance management device based on big data, 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, it implements the above-mentioned method.

[0034] Beneficial effects;

[0035] The present invention subdivides shared devices into mobile devices and non-mobile devices, and respectively establishes prediction models according to their characteristics and combined with historical data. This classification processing helps to more accurately predict the device status and improve the operation and maintenance efficiency. The shared devices are divided into healthy devices, critical devices, and faulty devices according to their health status. For critical devices, by shortening the analysis interval of the device data analysis frequency, problems can be discovered more timely, thereby accelerating the response speed and avoiding the further development of device failures; for faulty devices, the ant colony algorithm is used to generate the personnel inspection path, which can ensure that maintenance personnel can efficiently and orderly handle faulty devices. This intelligent path planning method helps to reduce the travel time of maintenance personnel and improve the maintenance efficiency. Therefore, through the present invention, the health status of shared devices is tracked at a low cost, so that potential failures can be discovered in time and repaired. Brief description of the drawings

[0036] Figure 1 It is a step flowchart of a method for operation and maintenance management of shared devices based on big data according to this application;

[0037] Figure 2 It is a schematic diagram of the principle for optimizing the fault conduction chain according to this application;

[0038] Figure 3 It is a schematic diagram of the principle for splitting and streamlining the conduction chain according to this application;

[0039] Figure 4 It is a schematic diagram of the structure of a system for operation and maintenance management of shared devices based on big data according to this application. Detailed implementation manners

[0040] In order to make the objectives, technical solutions and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0041] It can be understood that the terms "first", "second", etc. used in this application can be used herein to describe various elements, but unless otherwise specified, these elements are not limited by these terms. These terms are only used to distinguish the first element from another element. For example, without departing from the scope of this application, the first xx script can be called the second xx script, and similarly, the second xx script can be called the first xx script.

[0042] As Figure 1 shown, a method for operation and maintenance management of shared devices based on big data includes:

[0043] S1: Divide shared devices into mobile devices and non-mobile devices. The shared devices collect their own device data and send it to the central server.

[0044] The shared devices in this embodiment include shared bicycles, shared electric vehicles, shared power banks, shared massage chairs, etc. Mobile vehicles such as shared bicycles and shared electric vehicles are previously divided into mobile devices, and the rest are divided into non-mobile devices. All shared devices collect their own device data and send it to the central server. The device data includes device location, start time of use, end time of use, etc.

[0045] S2: Establish a first prediction model and a second prediction model based on the historical data of mobile devices and non-mobile devices respectively.

[0046] S3: The central server distinguishes the device types of shared devices according to the device identifiers in the received device data, and integrates the device data of mobile devices in the first time period into a first data set, and integrates the device data of non-mobile devices into a second data set.

[0047] S4: Analyze the first data set based on the first prediction model and analyze the second data set based on the second prediction model to determine the health status of the shared devices. Divide the shared devices into healthy devices, critical devices, and faulty devices based on the health status.

[0048] The first prediction model identifies whether a mobile device has a device failure according to the user behavior record. The second prediction model is used to identify whether a non-mobile device has a device failure. The specific prediction rules will be introduced later. After the central server receives the device data, it determines the device type of the shared device according to the device identifier received together. The device types include shared bicycles, shared electric vehicles, shared power banks, etc. The first time period is from 00:00 to 23:00 every day. For example, after the central server receives the device data sent by a shared bicycle between 00:00 and 23:00, it organizes it into a first data set. Similarly, it organizes the device data of non-mobile devices into a second data set. Then, the first data set is input into the first prediction model, and the second data set is input into the second prediction model, so that the first prediction model outputs the health status of the corresponding mobile device, and the second prediction model outputs the health status of the corresponding non-mobile device.

[0049] This embodiment divides into three ranges. The first range is 0 - 40%, the second range is 40 - 70%, and the third range is 70 - 100%. When the health status of a shared device is within the first range, it is divided into a faulty device. When it is within the second range, it is divided into a critical device, indicating that the device may be about to fail. When it is within the third range, it is divided into a healthy device.

[0050] S5: The central server shortens the analysis interval of the critical device's device data and generates a personnel inspection path based on the ant colony algorithm and the location information of the faulty device.

[0051] For critical devices, the monitoring frequency is increased by shortening the analysis interval. For example, when the shared device is in a healthy state, the analysis frequency is once a day, and when it is in a critical state, the analysis frequency is adjusted to once every 8 hours, that is, the device is analyzed every time 8 hours of data is collected. By this method, problems can be detected more promptly when a critical device fails.

[0052] For mobile devices, when the number of mobile devices judged to be faulty in the area exceeds a certain number, these mobile devices are tested uniformly. If a fault is determined after the test, they are repaired; otherwise, the fault report is eliminated in the system. For non-mobile devices, the ant colony algorithm is used to generate a personnel inspection path, which is the optimal path, enabling the maintenance personnel responsible for the area to efficiently test and repair the faulty devices at each location in sequence.

[0053] In the present invention, shared devices are subdivided into mobile devices and non-mobile devices, and prediction models are respectively established according to their characteristics and combined with historical data. This classification processing helps to more accurately predict the device status and improve the operation and maintenance efficiency. The shared devices are classified into healthy devices, critical devices, and faulty devices according to their health status. For critical devices, by shortening the analysis interval of the device data analysis frequency, problems can be discovered more promptly, thus accelerating the response speed and preventing the further development of device failures; for faulty devices, the ant colony algorithm is used to generate a personnel inspection path, which can ensure that the maintenance personnel can efficiently and orderly handle the faulty devices. This intelligent path planning method helps to reduce the travel time of the maintenance personnel and improve the maintenance efficiency.

[0054] It should be particularly noted that the present invention tracks the health status of shared devices at a low cost, so that potential faults can be discovered in time and repaired.

[0055] In this embodiment, the steps for the mobile device to send usage information include:

[0056] Set a first transmission method and a second transmission method in the mobile device. The device data includes the location information of the mobile device. In the default state, the mobile device sends the device information to the central server based on the first transmission method.

[0057] After the central server collects the device data within the second time period, it analyzes the included location information to determine the high-frequency locations where the mobile devices are used, deploys relay nodes at the high-frequency locations, and combines the location information with the deployment locations of the relay nodes. When it determines that the mobile device enters the coverage range of the relay node, it commands the mobile device to use the second transmission method to send the device data to the relay node, and the relay node adaptively compresses the device data and then sends it to the central server.

[0058] In this embodiment, the first transmission method is cellular network communication, and the second transmission method is Bluetooth communication. In other embodiments, the second transmission method can also be local area network communication.

[0059] In the default state, the mobile device directly sends its own device data to the central server in the first transmission method. The time length of the second time period is one month. For example, after the central server collects the device data of each mobile device in a certain area within one month, it starts to analyze and determine the high-frequency usage locations of the mobile devices. More mobile devices need to be placed at the high-frequency usage locations to ensure meeting the usage needs of users. Since the presence of more mobile devices at the same location will increase the transmission load in this area, to achieve faster data transmission, this embodiment sets up relay nodes at the high-frequency locations, and the relay nodes are edge servers.

[0060] After the relay nodes are set up, when the central server detects that the mobile device enters the coverage range of a certain relay node, such as the coverage range is 100m, it sends a conversion instruction to the mobile device, so that the mobile device uses the second transmission method to transmit the device data to the relay node. After the relay node aggregates the device data of multiple mobile devices, it compresses and packages them and then sends them to the central server, thereby reducing the data transmission volume and reducing the receiving frequency of the central server, thus accelerating the data transmission speed.

[0061] In this embodiment, the adaptive compression of the device data includes the following steps:

[0062] A micro model is deployed in the relay node. The micro model is a simplified model of the first prediction model. Based on the micro model and the device data, the mobile devices are divided into healthy devices and faulty devices. The device data of the healthy devices received within the third time period is aggregated and compressed into a whole block of data and then sent to the central server, and the device data of the faulty devices is inserted with a fault mark and then sent to the central server.

[0063] Specifically, the micro model is a simplified version of the first prediction model. In this embodiment, the first prediction model is a 3-layer BP neural network, and the corresponding micro model is a neural network with reduced input features. By streamlining the structure of the neural network, the relay node, as an edge computing node, can also quickly evaluate the status of the mobile device, realizing the pre-screening of the mobile device status. Particularly, the reference range for dividing the mobile device by the relay node remains the same as before. After being identified by the relay node, for healthy devices, they are aggregated and compressed into a whole block of data and then sent to the central server. After decompression by the central server, the complete data can be obtained. For the data of faulty devices, a fault mark is inserted and then sent. The central server can directly use it without decompression after receiving it. Through this method, not only can the health status of the mobile device be obtained more quickly, but also the communication times between the mobile device and the central server can be reduced, and the communication load of the central server can be lowered.

[0064] In this embodiment, establishing the first prediction model includes the following steps:

[0065] The historical data includes the location of the mobile device, the corresponding occurrence time, and the fault type. Based on the historical data, the movement characteristics of each user during the use of the mobile device are calculated. The movement characteristics include the average movement speed, the maximum movement speed, the minimum movement speed, the number of brakes, and the usage duration. The first prediction model is established based on a fully connected neural network. The movement characteristics are used as the input features of the first prediction model, and the fault type and occurrence probability are used as the output features. And the first prediction model is trained and established in combination with the historical data.

[0066] The historical data is a large amount of historical fault data collected in advance. The historical data includes the fault type of the data, as well as the location of the mobile device every 3 s, the occurrence time of the location, and the location is recorded in the form of longitude and latitude. After the user returns the mobile device, the average movement speed is calculated based on the usage duration and the movement path of the mobile device. The interval speed between two adjacent locations is calculated based on the distance and time difference between each adjacent location, and the maximum movement speed and the minimum movement speed are selected from them. If the interval speed between two adjacent locations drops from a larger value to a smaller value, and the difference between the two is greater than a preset value, it is considered that the user has made a brake. The first prediction model is a 3-layer BP neural network. The input features are the above-mentioned movement characteristics, and the output features are various fault types and their corresponding occurrence probabilities. The original probability output by the neural network is normalized to between 0 and 1 by using the Softmax function to obtain the occurrence probability.

[0067] In this embodiment, establishing the second prediction model includes the following steps:

[0068] Historical data includes the usage characteristics and failure types of non-mobile devices. Multiple usage characteristics that appear before each failure occurrence are integrated in chronological order to form a failure conduction chain for this failure. The failure type is marked in the failure conduction chain, and adjacent and identical usage characteristics in the failure conduction chain are merged to optimize the failure conduction chain into a streamlined conduction chain.

[0069] The historical data is a large amount of historical failure data collected in advance. For example, when a non-mobile device has failure type 1, multiple usage characteristics before the occurrence of failure type 1 are integrated in time to form a failure conduction chain for failure type 1. Failure type 1 is, for example, that the massage chair cannot heat up. The usage characteristics that appeared before include turning on the heating, raising the temperature by 1°C, raising the temperature by 1°C, turning off the heating, etc. The above-mentioned usage characteristics that have appeared are sorted in time to form a failure conduction chain for failure type 1. Similarly, the above processing is performed on the historical data of all non-mobile devices, and multiple different failure conduction chains for the same failure type can be obtained. Refer to the schematic diagram as Figure 2 shown. In order to streamline the length of the conduction chain, consecutive occurrences of usage characteristics are merged. For example, if the temperature is raised by 1°C consecutively twice, they are merged into one, and the merge count is marked, which is 2 times here.

[0070] Starting from the first usage characteristic in the streamlined conduction chain, each streamlined conduction chain is recursively split into multiple sub-conduction chains based on chronological order. Based on historical data, the association probabilities of various failure types after the appearance of the sub-conduction chains are statistically calculated. The sub-conduction chains with association probabilities greater than the first threshold are selected as the first sensitive conduction chains. After screening the first sensitive conduction chains, the second sensitive conduction chains are obtained. The second sensitive conduction chains, the corresponding failure types, and the association probabilities are used as the training set, and a second prediction model based on a time-series neural network is established based on the training set.

[0071] Refer to the schematic diagram as Figure 3 shown. For example, if the streamlined conduction chain is turn on heating - raise temperature by 1°C (5 times) - turn off heating, it is recursively split into three sub-conduction chains: turn on heating, turn on heating - raise temperature by 1°C (5 times), turn on heating - raise temperature by 1°C (5 times) - turn off heating. For example, there is also a streamlined conduction chain turn on heating - raise temperature by 5°C (5 times) - turn off heating, which is recursively split into turn on heating, turn on heating - raise temperature by 5°C (5 times), turn on heating - raise temperature by 5°C (5 times) - turn off heating. Repeat this step to split each streamlined conduction chain.

[0072] Then, the association probability between each sub-conduction chain and the failure type is calculated based on the first formula. The first formula is , where is the association probability between the i-th sub-conduction chain and the j-th failure type, is the occurrence times of the $i$-th sub-conduction chain under the $j$-th fault type. is the number of all sub-conduction chains included under the $j$-th fault type. For example, through calculation, it is determined that the association probability between fault type 1 and sub-conduction chain A is 40%, and the set first threshold is 35%. Then, sub-conduction chain A is taken as the first sensitive conduction chain. The higher the association probability, the more likely it is that the device has a fault of type 1 after sequentially presenting the usage characteristics of sub-conduction chain A. Then, the first sensitive conduction chain is refined to obtain the second sensitive conduction chain, and the specific calculation method will be introduced later.

[0073] Finally, the second sensitive conduction chain, the corresponding fault type, and the association probability are used as a training set to establish an LSTM time series neural network model as the first prediction model.

[0074] The screening of the first sensitive conduction chain in this embodiment includes the following steps:

[0075] The first sensitive conduction chains derived from the same refined conduction chain are defined as homologous conduction chains. Based on the length feature, the homologous conduction chains are sorted from large to small, and the difference in the association probability of adjacent homologous conduction chains after sorting is calculated. If the difference is greater than the first threshold, the homologous conduction chain with a longer length is retained. If the difference is less than the second threshold, the homologous conduction chain with a shorter length is retained. Then, the retained homologous conduction chain is continuously compared with the homologous conduction chain at the subsequent position, and this step is repeated until all the first sensitive conduction chains are traversed, and the retained ones are defined as the second sensitive conduction chains.

[0076] For the sake of description, the refined conduction chain is defined as A - B - C - D. The homologous conduction chains derived from this refined conduction chain include A - B - C - D, B - C - D, and C - D, and their association probabilities with fault type 1 are 54%, 40%, and 36% respectively. The above homologous conduction chains have been sorted from large to small according to their length features. The difference in the association probability between homologous conduction chain A - B - C - D and B - C - D is 14%, and the first threshold is 10%. Then, homologous conduction chain A - B - C - D is retained, and homologous conduction chain B - C - D is deleted, indicating that the longer homologous conduction chain has a closer relationship with fault type 1. Another example is that the association probabilities of the above three homologous conduction chains with fault type 1 are 36%, 59%, and 60% respectively. Then, the difference in the association probability between homologous conduction chain A - B - C - D and B - C - D is -13%, and the second threshold is -10%. Then, homologous conduction chain B - C - D is retained, and homologous conduction chain A - B - C - D is deleted. In particular, if the difference is between the first threshold and the second threshold, the homologous conduction chain with a shorter length is retained. For example, in the above example of 36%, 59%, and 60%, the difference between B - C - D and C - D is between the first threshold and the second threshold, so homologous conduction chain C - D is retained.

[0077] The steps for calculating the health degree of non-mobile devices in this embodiment include the following:

[0078] Integrate the device data of each user ID in the first dataset into an actual feature chain, input each actual feature chain into the first prediction model to obtain the failure type with the highest associated probability with the actual feature chain, perform weighted averaging on the associated probabilities of all failure types, calculate the health degree based on the associated probability with the largest average value among them, and output it jointly with the corresponding failure type.

[0079] For example, there are 3 users using the device in sequence within the first time period. According to the user IDs of the 3 users, 3 corresponding actual feature chains are generated respectively, and the actual feature chains are refined. Then, input the actual feature chains into the first prediction model to obtain the following results: The associated probability between actual feature chain 1 and failure type 1 is the largest, which is 16%. The associated probability between actual feature chain 2 and failure type 2 is the largest, which is 30%. The associated probability between actual feature chain 3 and failure type 1 is the largest, which is 40%. Then, perform weighted averaging on the failure types according to the time of the actual feature chains. The closer to the current time, the greater the weight it has. For example, the weights of the 3 actual feature chains are 0.3, 0.5, and 0.7 respectively. The average value of failure type 1 is (16% * 0.3 + 40% * 0.7) / 2 = 16.4%. The average value of failure type 2 is 30% * 0.5 = 15%. Then the health degree is 1 - 16.4% = 83.6%.

[0080] For the health degree of mobile devices, the first prediction model will also output the occurrence probability of the corresponding failure type according to the movement characteristics of each user, screen out the failure type with the largest occurrence probability, and then perform weighted summation on the occurrence probabilities of the same type of failure types according to the above method. The specific calculation process is the same as above and will not be elaborated here.

[0081] In this embodiment, the steps for determining high-frequency locations include the following:

[0082] Divide the preset area into a fixed number of grid areas, perform spatio-temporal density clustering on the mobile devices based on their positions and corresponding appearance times, screen out the dense areas from the clustering results, and use the grid areas where the dense areas are located as high-frequency locations.

[0083] Specifically, determine high-frequency locations through the following steps. First, divide the preset area into grid cells of 50m × 50m, and use ST-DBSCAN to perform spatio-temporal density clustering on the positions of the mobile devices from two dimensions of position and time to determine the hot grid where the devices appear. For example, if after clustering, the position information of 100 mobile devices appears at a certain location within 30 minutes, then the grid area where this location is located is used as a high-frequency location.

[0084] Such asFigure 4 As shown in Figure 4 , a shared device operation and maintenance management system based on big data is used to implement the above-mentioned method for operating and maintaining a shared device based on big data. It is characterized in that

[0085] The terminal module includes shared devices, which are divided into mobile devices and non-mobile devices. The shared devices collect their own device data and send it to the central server.

[0086] The front-end module is set in the central server. The front-end module respectively establishes a first prediction model and a second prediction model based on the historical data of mobile devices and non-mobile devices, distinguishes the device types of shared devices according to the device identifiers in the received device data, and integrates the device data of mobile devices within the first time period into a first data set and the device data of non-mobile devices into a second data set.

[0087] The classification module is set in the central server. The classification module analyzes the first data set based on the first prediction model and analyzes the second data set based on the second prediction model to determine the health status of the shared devices, and divides the shared devices into healthy devices, critical devices, and faulty devices based on the health status.

[0088] The policy module is set in the central server. The policy module shortens the analysis interval of the device data of critical devices and generates a personnel inspection path based on the ant colony algorithm and the location information of faulty devices.

[0089] The present application also provides a shared device operation and maintenance management device based on big data, 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 above-mentioned method is implemented.

[0090] It should be understood that the various technical features of the above embodiments can be combined arbitrarily. For the sake of brevity of description, not all possible combinations of the various technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, it should be considered as falling within the scope described in this specification.

[0091] The above is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent replacements, and improvements made within the spirit and principle of the present invention should be included within the protection scope of the present invention.

Claims

1. A shared equipment operation and maintenance management method based on big data, characterized in that: Classifying shared devices into mobile devices and non-mobile devices, wherein the shared devices collect their own device data and send them to a central server; Establishing a first prediction model and a second prediction model based on the historical data of the mobile device and the non-mobile device respectively; The central server distinguishes the device type of the shared device according to the device identifier received in the device data, and integrates the device data of the mobile device in the first time period into a first data set and integrates the device data of the non-mobile device into a second data set; Analyze the first data set based on a first prediction model and analyze the second data set based on a second prediction model to determine the health of the shared device, and classify the shared device into healthy devices, critical devices, and faulty devices based on the health; The central server shortens the analysis interval of the device data of the critical device, and generates a personnel inspection path based on the ant colony algorithm and the location information of the faulty device; Establishing the first prediction model comprises the following steps: The historical data includes the location of the mobile device, the corresponding occurrence time and the fault type. Based on the historical data, the movement characteristics of each user in the process of using the mobile device are calculated, and the movement characteristics include the average movement speed, the maximum movement speed, the minimum movement speed, the number of brakes and the use time. The first prediction model is established based on a fully connected neural network, and the movement characteristics are used as input features of the first prediction model, and the fault type and occurrence probability are used as output features. The first prediction model is established in combination with the historical data training; Establishing the second prediction model comprises the following steps: The historical data includes the usage characteristics of the non-mobile device and the fault type, a plurality of the usage characteristics that appear before each fault occurs are integrated into a fault conduction chain for the fault in the order of occurrence time, the fault type is marked in the fault conduction chain, and adjacent and identical usage characteristics in the fault conduction chain are merged to optimize the fault conduction chain into a streamlined conduction chain; Starting with the first usage feature in the streamlined conduction chain, each streamlined conduction chain is recursively split into multiple sub-conduction chains based on time sequence, and the association probability of various fault types caused by the appearance of the sub-conduction chain is statistically calculated based on the historical data, and the sub-conduction chain with the association probability greater than a first threshold is screened as the first sensitive conduction chain. After screening the first sensitive conduction chain, a second sensitive conduction chain is obtained, and the second sensitive conduction chain, the corresponding fault type and the association probability are used as a training set, and the second prediction model based on the time series neural network is established based on the training set.

2. The method according to claim 1, characterized in that The mobile device sending the device data comprises the following steps: A first transmission mode and a second transmission mode are set in the mobile device, the device data includes location information of the mobile device, and the mobile device sends the device data to the central server based on the first transmission mode in a default state; After collecting the device data within the second time period, the central server analyzes the location information included therein to determine high-frequency locations used by the mobile device, and deploys relay nodes at the high-frequency locations. The central server combines the location information and the deployment location of the relay nodes, and when it is determined that the mobile device has entered the coverage range of the relay node, commands the mobile device to use the second transmission method to send the device data to the relay node. The relay node adaptively compresses the device data and sends it to the central server.

3. The method according to claim 2, characterized in that Adaptively compressing the device data comprises the following steps: A micro-model is deployed in the relay node, where the micro-model is a simplified model of the first prediction model. The mobile device is divided into the healthy device and the faulty device based on the micro-model and the device data. The device data received from the healthy device in a third time period is aggregated and compressed into a whole block of data and then sent to the central server. The device data of the faulty device is inserted into a fault mark and then sent to the central server.

4. The method according to claim 1, characterized in that Screening the first sensitive conductive chain includes the following steps: The first sensitive conduction chain derived from the same streamlined conduction chain is defined as a homologous conduction chain. The homologous conduction chains are sorted from large to small based on length characteristics, and the difference in the association probabilities of the homologous conduction chains at adjacent positions after sorting is calculated. If the difference is greater than a third threshold, the homologous conduction chain with a longer length is retained. If the difference is less than a second threshold, the homologous conduction chain with a shorter length is retained, and the retained homologous conduction chain continues to be compared with the homologous conduction chain at a subsequent position. This step is repeated until all the first sensitive conduction chains are traversed and the retained one is defined as the second sensitive conduction chain.

5. The method according to claim 4, characterized in that Calculating the health of the non-mobile device comprises the following steps: The device data of each user ID in the second data set are integrated into an actual feature chain, each actual feature chain is input into the second prediction model, the fault type with the highest probability of association with the actual feature chain is obtained, the association probabilities of all the fault types are weighted and averaged, the health degree is calculated based on the association probability with the largest average value, and the health degree is output jointly with the corresponding fault type.

6. The method according to claim 2, characterized in that Determining the high frequency location comprises the following steps: The preset area is divided into a fixed number of grid areas, and the mobile devices are clustered in time and space density based on the locations of the mobile devices and the corresponding appearance times. Based on the clustering results, dense areas are screened out from the grid areas, and the grid areas where the dense areas are located are used as the high-frequency locations.

7. A shared equipment operation and maintenance management system based on big data, used to implement a shared equipment operation and maintenance management method based on big data as described in any one of claims 1 to 6, characterized in that: The terminal module includes a shared device, which divides the shared device into a mobile device and a non-mobile device, and the shared device collects its own device data and sends it to the central server; A front-end module is arranged in a central server. The front-end module establishes a first prediction model and a second prediction model based on the historical data of the mobile device and the non-mobile device, respectively. The historical data includes the location of the mobile device, the corresponding occurrence time and the fault type. The movement characteristics of each user in the process of using the mobile device are calculated based on the historical data. The movement characteristics include the average movement speed, the maximum movement speed, the minimum movement speed, the number of brakes and the use time. The first prediction model is established based on a fully connected neural network. The movement characteristics are used as input characteristics of the first prediction model, and the fault type and occurrence probability are used as output characteristics. The first prediction model is established in combination with the historical data training. The historical data includes the usage characteristics and the fault type of the non-mobile device. The multiple usage characteristics that appear before each fault occur are integrated into a fault conduction chain for this fault in the order of occurrence time. In the fault conduction chain Marking the fault type, merging the adjacent and identical usage features in the fault conduction chain to optimize the fault conduction chain into a simplified conduction chain, starting with the first usage feature in the simplified conduction chain, recursively splitting each simplified conduction chain into multiple sub-conduction chains based on time sequence, statistically calculating the association probability of various fault types caused by the appearance of the sub-conduction chain based on the historical data, screening the sub-conduction chain with the association probability greater than a first threshold as a first sensitive conduction chain, obtaining a second sensitive conduction chain after screening the first sensitive conduction chain, taking the second sensitive conduction chain, the corresponding fault type and the association probability as a training set, establishing the second prediction model based on the time series neural network based on the training set, distinguishing the device type of the shared device according to the device identifier in the received device data, and integrating the device data of the mobile device in the first time period into a first data set and the device data of the non-mobile device into a second data set; a classification module, arranged in the central server, wherein the classification module analyzes the first data set based on a first prediction model and analyzes the second data set based on a second prediction model to determine the health of the shared device, and classifies the shared device into healthy devices, critical devices, and faulty devices based on the health; A strategy module is provided in the central server, wherein the strategy module shortens the analysis interval of the device data of the critical device and generates a personnel inspection path based on the ant colony algorithm and the location information of the faulty device.

8. A shared equipment operation and maintenance management device based on big data, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the method according to any one of claims 1 to 6 is implemented.

Images