Shared electric bicycle riding abnormity safety early warning method and system based on group data, medium and processor

Abstract

The invention provides a shared electric bicycle riding abnormity safety early warning method based on group data. The method comprises the following steps: S1, carrying out data acquisition on a vehicle-mounted terminal; s2, summarizing the vehicle-mounted terminal data, obtaining corresponding rider data, region type data, traffic flow grade data and meteorological data during riding, and packaging to form group data; s3, performing cleaning preprocessing such as verification, format unification, noise filtering and missing value processing on the group data; s4, performing feature extraction and coding on the cleaned group data, and training an MLP learning model; s5, obtaining new riding data, extracting corresponding data preprocessing, feature vector extraction and classification coding, and inputting the data into the MLP learning model for group feature matching; and S6, judging whether the behavior of the current rider accords with the normal riding mode of the group where the current rider is located or not, and if not, determining that riding is abnormal. Compared with the prior art, the abnormal riding behavior can be identified more accurately.

Description

Technical Field

[0001] The present invention relates to the technical field of abnormal riding recognition of electric bicycles, and particularly relates to a method, a system, a medium and a processor for abnormal riding safety warning of shared electric bicycles based on group data. Background Art

[0002] In the field of modern urban transportation, shared electric bicycles have developed vigorously and become a popular choice for people's short-distance travel. From the perspective of well-known technologies, currently, shared electric bicycles generally incorporate global positioning system (GPS) technology. The operation platform can thereby obtain the geographical location information of the vehicles in real time, achieve a macroscopic control of the vehicle distribution, and reasonably arrange vehicle placement and dispatching to meet the travel needs of different regions. At the same time, the sensor technology built into the vehicles is also widely used. For example, an acceleration sensor can monitor the start and stop, acceleration and deceleration states of riding, a gyroscope sensor helps to judge the attitude stability of the vehicle, and a power sensor can provide real-time feedback on the remaining power to ensure that the riding process will not break down halfway due to power exhaustion.

[0003] Delving deeper into the research significance of the shared bicycle monitoring and control system, safety comes first. Abnormal riding is often closely associated with safety risks. Behaviors such as speeding downhill on a steep slope, frequently braking suddenly and turning, or even riding with one hand or both hands off the handlebars during riding are extremely likely to cause accidents such as falls and collisions, which not only threaten the life and health of the rider himself, but may also affect surrounding pedestrians and other vehicle drivers. According to the statistical data of the traffic department, the number of traffic accidents caused by abnormal riding of shared electric bicycles has been on the rise every year, resulting in a large number of casualties and property losses. Therefore, it is urgent to establish an early warning system.

[0004] Furthermore, from the perspective of operation and management, accurately identifying abnormal riding helps shared electric bicycle enterprises reduce operation and maintenance costs. Abnormal riding behaviors are likely to accelerate the wear of vehicle parts. Early awareness and intervention can effectively reduce the frequency of repairs and extend the service life of the vehicle. Moreover, by preventing situations such as vehicle theft and illegal parking through the early warning system, the enterprise can avoid unnecessary economic losses, optimize resource allocation, and improve operation efficiency.

[0005] In addition, from the perspective of comprehensive urban traffic governance, regulating the riding order of shared electric bicycles is a key link in building a harmonious traffic environment. If abnormal riding runs wild, it will disrupt the normal traffic flow, reduce road traffic efficiency, and exacerbate traffic congestion. By using the early warning system to restrict and guide riding behaviors, it can promote shared electric bicycles to better integrate into the urban comprehensive traffic system and achieve the coordinated development of multiple travel modes.

[0006] At present, certain achievements have been made in the research on the supervision of shared electric bicycle riding. However, by using simple rule algorithms and relying on a single parameter feedback from the vehicle's built-in sensors, such as the absolute value of speed to determine speeding, once the preset speed limit is exceeded, an alarm will be triggered immediately. Some other studies have introduced limited machine learning models, collected a small number of representative riding samples, learned and trained the common abnormal behavior characteristics, and constructed a preliminary discrimination model, such as distinguishing normal and abnormal riding trajectory forms based on the decision tree algorithm.

[0007] However, the existing research has revealed significant defects and deficiencies. First, the bicycle data is viewed in isolation, resulting in a high false alarm rate for the alarm.

[0008] Second, the existing algorithm models have poor universality. On the one hand, most models are trained based on limited samples in specific cities and specific regions, and it is difficult to directly promote and apply them to other regions. The topographical features, traffic rules, and humanistic environments of different cities vary greatly, and fixed models cannot adapt to diverse changes. For example, in mountainous Chongqing, the roads have large undulations, many steep slopes and curves, and the models trained based on the data of plain cities almost fail here. On the other hand, the model update lags behind, making it difficult to cope with newly emerging abnormal riding patterns. As the usage scenarios of shared electric bicycles continue to expand, some emerging abnormal behaviors such as "group racing" and "using electric bicycles to carry goods" have gradually emerged, and the existing models lack a self-learning and updating mechanism and cannot identify them in a timely manner.

[0009] Third, it is difficult to balance real-time performance and accuracy. Some alarm systems sacrifice judgment accuracy by simplifying the algorithm process in order to pursue fast response; while some high-precision models are time-consuming in data processing due to complex calculations, and the accident may have already occurred when the alarm is issued, so they cannot truly play a role in early prevention and are difficult to meet the actual traffic management needs.

[0010] In view of this, a method, system, medium, and processor for abnormal safety warning of shared electric bicycle riding based on group data are needed. Summary of the Invention

[0011] Aiming at the problem of high false alarm rate in the existing technology, the present invention provides a method, system, medium, and processor for abnormal safety warning of shared electric bicycle riding based on group data, which can combine bicycle data with condition data such as groups and regions to improve the alarm accuracy. The specific technical solutions are as follows:

[0012] A method for abnormal safety warning of shared electric bicycle riding based on group data includes the following steps:

[0013] S1: Collect data from in-vehicle terminals;

[0014] S2: Aggregate the in-vehicle terminal data, obtain the corresponding rider data, regional type data, traffic flow level data, and meteorological data during cycling, and package them to form group data;

[0015] S3: Perform cleaning and preprocessing on the group data, such as verification, format unification, noise filtering, and missing value processing;

[0016] S4: Extract features, encode the group data after cleaning, and use it for training the MLP learning model;

[0017] S5: Obtain new cycling data, extract the corresponding data preprocessing, feature vector extraction, and classification encoding, and input them into the MLP learning model for group feature matching;

[0018] S6: Determine whether the current rider's behavior conforms to the normal cycling mode of their group. If not, it is considered an abnormal cycling.

[0019] Further, in step S4, the following steps are included:

[0020] S41: For the data after cleaning, perform normalization processing and feature extraction on numerical data, and encode categorical data to form a historical dataset;

[0021] S42: Construct an MLP learning model and train and validate it with the historical cycling data in the historical dataset, so that the MLP learning model can learn the cycling characteristics of different groups.

[0022] Further, in step S5, the following steps are included:

[0023] S51: Collect, preprocess new cycling data, extract feature vectors, and encode categorical data;

[0024] S52: Input the feature vectors and classification encodings extracted from the new cycling data into the trained MLP learning model, and calculate the probability distribution of the group categories to which the current rider belongs;

[0025] S53: Select the group category with the highest probability as the matching group for the current rider.

[0026] Further, in step S6, the following steps are included:

[0027] S61: Dynamically adjust the anomaly threshold according to the characteristics of the current rider, road conditions, and environmental conditions;

[0028] S62: According to the anomaly threshold, use the pre-trained Hidden Markov Model (HMM) to determine whether the cycling behavior conforms to the normal cycling mode of their group. If not, it is determined as an abnormal cycling.

[0029] Further, the abnormal threshold calculation formula is as follows:

[0030]

[0031] where y is the abnormal threshold, and w i is the weight of the i-th factor; x i is the value of the i-th factor after normalization; n is the number of factors affecting cycling, including 5 factors: age factor, road condition factor, regional type factor, traffic flow level factor, and meteorological factor.

[0032] Further, it also includes the following steps:

[0033] S7: Determine the type and weight of cycling anomalies, and select the corresponding warning level according to the corresponding weight to generate a warning decision instruction;

[0034] S8: Select the corresponding warning channel according to the warning decision instruction to issue a warning.

[0035] Further, the step S7 includes the following steps:

[0036] S71: Determine the type of cycling anomalies and calculate the corresponding weights;

[0037] S72: Divide different warning levels according to the calculated abnormal behavior weights, and generate corresponding decision instructions.

[0038] A shared e-bike cycling anomaly safety warning system based on group data, which is applied to the above-mentioned shared e-bike cycling anomaly safety warning method based on group data, includes:

[0039] An acquisition module, which is used to collect data from in-vehicle terminals;

[0040] A communication management module, which is used to summarize the in-vehicle terminal data and obtain the corresponding cyclist data, regional type data, traffic flow level data, and meteorological data during cycling, and package them to form group data;

[0041] A data processing center, which is used to perform cleaning preprocessing such as verifying, unifying the format, filtering noise, and processing missing values on the group data; perform feature extraction, encoding on the cleaned group data, and use it for MLP learning model training; obtain new cycling data and extract the corresponding data preprocessing, feature vector extraction, and classification encoding, and input them into the MLP learning model for group feature matching; determine whether the behavior of the current cyclist conforms to the normal cycling mode of the group he belongs to, and if not, it is a cycling anomaly.

[0042] A computer-readable storage medium, the computer-readable storage medium including a stored program, wherein when the program runs, it controls the device where the computer-readable storage medium is located to execute the above-mentioned shared electric bicycle riding anomaly safety warning method based on group data.

[0043] A processor, the processor being used to run a program, wherein when the program runs, it executes the above-mentioned shared electric bicycle riding anomaly safety warning method based on group data.

[0044] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0045] 1. In terms of riding safety guarantee, by using technologies such as group data matching, dynamic threshold mechanism, and HMM pattern recognition, compared with the prior art, it can more accurately identify abnormal riding behaviors. The prior art may only rely on simple speed or position threshold judgments, which are prone to misjudgments. This solution fully considers the characteristics of different rider groups (such as age, gender, riding time period, region, etc.) and riding scenarios (such as road conditions, weather, etc.), reduces misjudgments and missed judgments, discovers and prevents riding safety hazards in a timely manner, and guarantees the safety of riders. The system performs warning push through multiple channels such as vehicle central control voice reminder, App, text message, traffic management department, etc. The prior art may only rely on a single reminder method, and it is easy for riders not to receive warning information in a timely manner. This solution ensures that riders, operation platforms, and traffic management departments can all obtain abnormal information in a timely manner. Riders can immediately correct their behaviors, operation platforms can take measures such as remotely locking the vehicle, power off, and speed reduction in a timely manner, and traffic management departments can enforce the law in a timely manner, comprehensively guaranteeing riding safety.

[0046] 2. In terms of data processing and transmission, the data transmission preferentially uses a 5G link management module. Compared with the 4G or lower-speed networks that may be used in the prior art, the 5G network has the characteristics of high speed and low latency, ensuring that data is quickly and stably transmitted from the in-vehicle terminal to the data processing center. And when the 5G signal is poor, the network can be switched to avoid untimely warnings caused by data transmission delays and ensure the coherence of data transmission. The data processing center uses technologies such as CRC check, noise data filtering, and ARIMA model data missing processing, and reduces data redundancy through the MQTT protocol. The prior art may lack these data processing means, which are prone to inaccurate and incomplete data and low data processing efficiency. This solution guarantees the accuracy and integrity of data, improves data processing and storage efficiency, reduces data processing costs, and efficiently processes a large amount of riding data.

[0047] 3. In terms of operation management, the early warning decision-making module determines the weights of abnormal behaviors and generates decision instructions. Existing technologies may lack refined analysis and decision-making for abnormal behaviors. According to the instructions of this solution, the operation platform can intelligently manage vehicles and users. For example, it can perform early maintenance on frequently abnormal vehicles, restrict the use of users with many violations, optimize operation strategies, and improve the efficiency and scientific nature of operation management. Through accurate early warning and multi-channel push, the operation platform can targetedly manage vehicles and users, reducing unnecessary on-site inspections and manual interventions. At the same time, the power module's reasonable strategy when the battery level is low (such as reducing the sampling frequency of non-critical sensors) extends the device's battery life, reduces the battery replacement frequency, and lowers the operation and maintenance costs, showing significant advantages in operation and maintenance costs compared to existing technologies.

[0048] 4. In terms of urban traffic order, the information of abnormal riding events with serious violations is promptly pushed to the traffic management department. Existing technologies may not be able to promptly transmit the violation information of shared e-bikes to the traffic management department, resulting in untimely traffic law enforcement. This solution helps traffic police promptly grasp the violation situation and enforce the law, reduce the violation behaviors of shared e-bikes, maintain urban traffic order, and ensure public traffic safety. The large amount of riding data accumulated by the system (including normal and abnormal behavior data) can provide valuable data support for urban traffic planning and management. Existing technologies may not be able to make full use of the riding data. By analyzing this data, the present invention can understand the travel habits of riders, popular routes, high-incidence areas of violations, etc., providing a reference basis for optimizing urban traffic facilities (such as setting dedicated bike lanes, traffic signs, etc.) and formulating traffic management policies. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] In order to more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the following will briefly introduce the drawings required for use in the description of the specific embodiments or the prior art. In all the drawings, similar elements or parts are generally identified by similar reference numerals. In the drawings, the elements or parts do not necessarily draw according to the actual scale.

[0050] Figure 1 It is a schematic flow chart of a method for abnormal riding safety warning of shared e-bikes based on group data. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0051] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0052] It should be understood that when used in this specification and the appended claims, the terms "comprising" and "including" indicate the presence of the described features, wholes, steps, operations, elements, and / or components, but do not preclude the presence or addition of one or more other features, wholes, steps, operations, elements, components, and / or their combinations.

[0053] It should also be understood that the terms used in the specification of the present invention are merely for the purpose of describing specific embodiments and are not intended to limit the present invention. As used in the specification of the present invention and the appended claims, unless the context clearly indicates otherwise, the singular forms "a", "an", and "the" are intended to include the plural forms.

[0054] It should be further understood that the term "and / or" used in the specification of the present invention and the appended claims refers to any combination and all possible combinations of one or more of the associated listed items, and includes these combinations.

[0055] Existing research has revealed significant defects and deficiencies. First, it overly relies on the isolated data of individual bicycles and ignores the rich information contained in group cycling data. The user group of shared e-bikes is large and complex, and there are significant differences among different groups in terms of cycling habits, preferred routes, travel times, etc. For example, young office workers have a faster cycling rhythm and relatively fixed routes during the morning and evening rush hours on weekdays; while the elderly tend to cycle slowly and travel short distances during their leisure time to explore the surrounding areas. Looking at bicycle data in isolation cannot accurately capture these group characteristics, resulting in a high early warning misjudgment rate.

[0056] Second, the existing algorithm models have poor universality. On the one hand, most models are trained based on limited samples in specific cities and specific regions and are difficult to be directly applied and promoted to other regions. The topography, traffic rules, and human environment of different cities vary greatly, and fixed models cannot adapt to diverse changes. For example, in mountainous Chongqing, the roads have large undulations, many steep slopes and curves, and models trained based on data from plain cities almost fail here; on the other hand, the model updates are lagging, and it is difficult to cope with newly emerging abnormal cycling patterns. As the usage scenarios of shared e-bikes continue to expand, some emerging abnormal behaviors such as "group racing" and "using e-bikes to carry goods" have gradually emerged, and the existing models lack a self-learning and updating mechanism and cannot identify them in a timely manner.

[0057] Third, it is difficult to balance real-time performance and accuracy. Some early warning systems sacrifice judgment accuracy by simplifying the algorithm process in order to pursue rapid response; while some high-precision models are time-consuming in data processing due to complex calculations, and the accident may have already occurred when the early warning is issued, so they cannot really play a role in early prevention and are difficult to meet the actual traffic management needs.

[0058] Embodiment 1

[0059] As Figure 1 shown in the flowchart of a shared e-bike abnormal riding safety warning method based on population data, which includes the following steps:

[0060] S1: Collect data from the vehicle-mounted terminal.

[0061] In specific implementation, the startup of the shared bike system is the starting point of the entire warning process, which may be triggered by the user unlocking the shared e-bike by scanning the code. The specific data collection steps are as follows:

[0062] (1) The vehicle-mounted terminal is powered on. When the user unlocks the shared e-bike, the vehicle-mounted terminal is powered on, which is the prerequisite for data collection and processing.

[0063] (2) Collect data through devices such as the speed sensor, GPS / Beidou positioning, camera recognition, gyroscope sensor, acceleration sensor, sensor installed on the crank or pedal shaft, pressure sensor, wheel speed sensor, and electronic compass of the vehicle-mounted terminal.

[0064] The speed sensor is used to accurately measure the riding speed of the e-bike, providing basic data for subsequent judgment of whether the riding behavior is normal; for example, it can monitor the driving speed of the rider on the road in real time to ensure the accuracy of the data. The GPS / Beidou positioning device can obtain the accurate position information of the e-bike, which helps to track the riding trajectory and can judge whether the rider deviates from the normal route or enters a dangerous area; its positioning accuracy is relatively high and can accurately reflect the vehicle position. Camera recognition can capture the riding posture of the rider and the surrounding road conditions; for example, it can detect whether the rider rides with one hand, both hands off the handlebars, or carries passengers illegally, and can also identify traffic lights and road signs and other information. The speed change during the whole ride (record the speed value every 10 seconds).

[0065] The gyroscope sensor is installed on the vehicle frame and can measure the angular velocity of the vehicle in three axes. By integrating the angular velocity, the tilt angle can be obtained, and it can monitor the left-right tilt and front-back tilt of the vehicle in real time. For example, the MPU6050 six-axis sensor integrates a 3-axis gyroscope and a 3-axis accelerometer, which can measure the angular velocity and acceleration at the same time. The gyroscope sensor is used in combination with the electronic compass. The gyroscope sensor can measure the angular velocity of the vehicle and provide more accurate direction change information in a short time, especially when the electronic compass is interfered or has insufficient accuracy, playing a supplementary and corrective role.

[0066] The acceleration sensor is used in conjunction with the gyroscope sensor. When the vehicle tilts, the gravitational components of the acceleration sensor in different axes will change. By measuring and analyzing the acceleration, it is also possible to assist in determining the tilt angle of the vehicle. The acceleration sensor is installed near the frame or the wheel axle, etc., and can measure the acceleration changes of the vehicle in all directions during driving. By analyzing the acceleration data, features such as the frequency and amplitude of vibrations are extracted to understand the road conditions and the comfort of riding. Acceleration data (also calculated every 10 seconds).

[0067] The sensor installed on the crank or the pedal axle, as the crank makes a circular motion around the center axis of the bicycle, the acceleration sensor inside the sensor can measure the number of circles of the circular motion of the crank by detecting the rotation in the direction of gravitational acceleration, thereby calculating the pedaling frequency data of the ride.

[0068] The six-axis sensor installed on the cycling shoes. The six-axis sensor is installed on the cycling shoes. By combining the data of the gyroscope and the accelerometer and extracting the attitude information of the foot movement through a specific algorithm, the pedaling frequency is then calculated.

[0069] The pressure sensor is installed on the pedal or the crank. When the cyclist steps on the pedal, the pressure sensor will measure the pressure change generated by the stepping. By analyzing and processing the pressure data, the magnitude and change of the stepping force can be obtained.

[0070] The wheel speed sensor usually uses a magnetosensitive or optoelectronic sensor and is installed near the spokes or the hub of the wheel. By detecting the rotation speed of the wheel and combining parameters such as the circumference of the wheel, the speed of the ride is calculated, and then the speed change rate is obtained.

[0071] In addition to providing location information, the GPS module can also calculate the speed and the speed change rate through continuous location data.

[0072] The electronic compass, also called a magnetometer, is installed on the frame and can measure the direction of the earth's magnetic field, thereby determining the driving direction of the vehicle. By continuously monitoring and analyzing the driving direction data, the change rate of the driving direction is calculated.

[0073] S2: The communication management module aggregates the in-vehicle terminal data, and then obtains the corresponding cyclist data during the ride, such as the cyclist's age accurate to the specific number of years, gender, ride time accurate to minutes, and regional type data, such as the regional types to which the departure and destination belong (marked as commercial areas, residential areas, school areas, etc. through the geographic information system), traffic flow level data (low, medium, high, determined according to the road sensor data), and meteorological information (such as rain, wind, light intensity, visibility in fog, temperature and humidity, etc.), and packs them to form group data.

[0074] In specific implementation, the regional data of the regions to which the departure and destination belong can be obtained through the following several channels:

[0075] Online map platforms. Baidu Map: It has rich geographical data and accurate positioning functions. By searching for the departure and destination, the regional type information around them can be viewed. For example, commercial areas, residential areas, school areas, etc. can be directly marked on the map. Moreover, its intelligent recognition function can be used to automatically judge and display the type of the area where one is located. Amap: Similar to Baidu Map, it provides detailed geographical information and regional markings. In addition to the basic map display, through layer switching and other methods, the distribution of different types of regions can be viewed more clearly. Its traffic condition information and real-time navigation function also help to understand the regional situation more accurately.

[0076] Google Map: It has high-precision geographical data globally, and the markings of regional types are relatively detailed and accurate. It has advantages in obtaining geographical information in foreign regions. Its street view function can more intuitively observe the actual situation of the area.

[0077] Geographic Information System (GIS) software. ArcGIS: A powerful professional GIS software with a large amount of geographical data and rich analysis tools. Users can import custom geographical data, such as urban planning maps, land use maps, etc. Through spatial analysis functions, the regional types to which the departure and destination belong can be accurately determined. Data editing, mapping, etc. operations can also be carried out on the Open Source China Community. QGIS: An open-source GIS software with functions similar to ArcGIS. It can be used and obtained for free, supports multiple data formats. Users can obtain free geographical data from the Internet, such as OpenStreetMap data, etc. By loading and analyzing these data, regional type information can be obtained.

[0078] Government department and relevant institution websites. Natural resources department: Websites of the Ministry of Natural Resources of the country and local natural resources bureaus will provide geographical information data on land use status, urban planning, etc. These data can help determine the regional type and are usually published in the form of maps, reports, statistical data, etc. Some data can be downloaded for free. Statistical department: The National Bureau of Statistics and local statistical bureaus will release data such as population censuses and economic censuses, which contain information on population distribution, types of economic activities, etc. in different regions. Based on this, the regional type can be inferred. For example, areas with dense population may be residential areas or commercial areas, and areas with concentrated industrial enterprises may be industrial areas, etc. Urban planning department: Websites of the planning bureaus of each city will publish urban master plans, detailed plans, etc., which clearly divide different functional areas such as commercial areas, residential areas, school areas, etc. The required information can be obtained by querying specific planning maps and relevant documents.

[0079] Commercial data service providers. Gartner: It provides various geospatial data and analysis services. Its data covers a variety of regional type information globally and can provide customized data solutions according to customer needs, but usually requires paying for services. Analysys: It has rich experience and a professional data team in geospatial data analysis and industry research. It can provide multi-dimensional data such as business activity, population density, and consumption ability in different regions of the city, helping users understand regional types and characteristics more comprehensively. Some data reports are available for free, while detailed data and customized services require payment.

[0080] Crowdsourced geospatial data platforms. OpenStreetMap: An open-source mapping project created and maintained by global volunteers, containing rich geospatial information such as roads, buildings, and land use types. Users can obtain regional type data through this platform and edit and update it. Waze: A traffic navigation application. During use, users can provide real-time feedback on road conditions, road information, and regional changes. Its data has high real-time and accuracy. Although it mainly focuses on traffic-related information, some clues about regional types can also be obtained from it.

[0081] In specific implementation, when the in-vehicle terminal uploads data to the cloud server, through the communication management module, the data collected by the in-vehicle terminal is stably and efficiently transmitted to the data processing center of the cloud server, while ensuring the accuracy and integrity during the data transmission process.

[0082] Furthermore, the in-vehicle communication management module preferably adopts a 5G link management module to stabilize the transmission rate. During data transmission, it preferably selects the 5G network for data transmission. The 5G network has the characteristics of high speed and low latency, and its theoretical peak transmission rate is much higher than that of the 4G network. For example, the peak transmission rate of the 5G network can reach 1 - 10 Gbps, while the peak transmission rate of the 4G network is generally between 100 Mbps - 1 Gbps. This high-speed and stable transmission ensures that data can reach the data processing center in a timely and accurate manner, avoiding untimely warnings caused by data transmission delays.

[0083] Furthermore, the MQTT protocol is adopted during transmission to reduce data redundancy. In scenarios of large-scale data collection and transmission, it helps save network bandwidth and improve data transmission efficiency. For example, when multiple vehicle-mounted terminals transmit data to the data processing center simultaneously, the MQTT protocol can avoid the transmission of duplicate data. Meanwhile, through a concise message header and an efficient transmission mechanism, it reduces the amount of data transmitted, thereby reducing data storage costs. MQTT (Message Queuing Telemetry Transport) is a lightweight Internet of Things message transmission protocol based on the publish-subscribe model. It is designed for low-bandwidth, high-latency, or unreliable network environments and can efficiently transmit messages between resource-constrained devices (such as sensors, microcontrollers, etc.) and servers.

[0084] S3: Perform cleaning and preprocessing on the group data, such as verification, unified format, noise filtering, missing value processing, etc.

[0085] In specific implementation, the data processing center is responsible for comprehensively preprocessing the data collected and transmitted from vehicle-mounted terminals, including operations such as data verification, unified format, noise filtering, matching, feature extraction, and anomaly determination, providing accurate data support for subsequent early warning decisions. Specifically, it includes the following steps:

[0086] S31: When the data processing center receives the data transmitted from the vehicle-mounted terminal, perform CRC verification on the data. When the sender transmits the data, it will generate a verification code through a specific CRC algorithm according to the data content and attach it to the data for transmission together. After receiving the data, the receiver (data processing center) calculates the data (excluding the verification code) using the same CRC algorithm to obtain a local verification code. Then compare the local verification code with the received verification code. If the two verification codes are consistent, it indicates that the data has not been corrupted during transmission; if they are inconsistent, it means that the data has experienced a transmission error, and the data processing center can request retransmission of the data to ensure the accuracy of the data. For example, during the transmission of shared electric bicycle riding data, ensure the accuracy of data such as speed and location.

[0087] S32: Unify the format of multi-source group data. The data collected by vehicle-mounted terminals comes from various sources, including sensors or devices such as speed sensors, GPS / Beidou positioning, cameras, etc. These data formats are different. When the data processing center receives the data, further, it will first unify the format of these multi-source data. For example, convert the speed data from the original format of the sensor to the standard numerical format, and convert the location data from the specific coding format of GPS / Beidou to the common latitude and longitude format, etc., to ensure the compatibility of data from different sources in subsequent processing.

[0088] S33: Perform noise filtering on the group data. Noise data filtering: Due to the complexity of the acquisition environment, the collected data may contain noise data. For example, a speed sensor may generate instantaneous abnormal speed values due to factors such as slight vibrations of the vehicle and electromagnetic interference. The data processing center filters these noise data by setting reasonable thresholds and algorithms. Usually, based on statistical principles, the normal data range is determined according to historical data. When the received data exceeds the normal range to a certain extent and the duration is short, it is judged as noise data and filtered. For instance, the normal cycling speed is between 10 - 20 km / h. If the speed value suddenly becomes 50 km / h at a certain moment and only lasts for 0.1 second, then this data is very likely to be noise data and will be filtered out to prevent these abnormal data from interfering with subsequent analysis.

[0089] S34: Handle missing values in the group data. Use the ARIMA model to handle data missing. During the data acquisition process, data missing may occur. The ARIMA model analyzes historical data to identify autoregressive, differencing, and moving average characteristics in the data, and then predicts the missing data. In cycling data, if the cycling speed data for a certain period is missing, the data processing center can use the ARIMA model to predict the approximate speed value for that period based on the speed data before and after, ensuring the integrity and coherence of the data and enabling subsequent data analysis to proceed smoothly.

[0090] There are also the following steps: Remove data with obvious errors or missing key information from the collected data. For example, if the age value of a certain record is negative or the speed value exceeds a reasonable range (such as exceeding 100 km / h, a speed that an electric bicycle cannot reach), it will be deleted.

[0091] S4: Extract features, encode the group data after cleaning processing, and use it for training the MLP learning model.

[0092] S41: For the data after cleaning, the data processing center will normalize and extract features from numerical data, and encode categorical data to form a historical dataset. For example, for speed data, features such as average speed, maximum speed, and speed change rate may be extracted. For location data, features of the riding trajectory will be extracted, such as the riding route and whether it deviates from the normal route. Vehicle parking behavior features will be extracted from the data collected by the camera. Tilt angle features: By analyzing the left-right tilt and front-back tilt angles, it is determined whether the vehicle tilts to the left, right, front, or back. For example, there will be obvious left-right tilts when turning, and front-back tilts when going uphill or downhill. Tilt degree classification features: The tilt angle can be divided into different levels, such as slight tilt (0 - 10 degrees), moderate tilt (10 - 30 degrees), and severe tilt (above 30 degrees), which are used to evaluate the stability of the riding state. In competitive cycling, riders may have large tilt angles when turning at high speeds, while in ordinary riding scenarios, the tilt angles are usually small. Tilt change frequency features: Calculate the number of tilt angle changes per unit time, which reflects the frequency of the rider's maneuvering the vehicle to turn or ride on undulating roads. For example, in mountain biking, due to the complex terrain, the tilt change frequency of the vehicle will be much higher than on flat roads. Vibration amplitude features: The maximum amplitude and average amplitude of the vibration can be extracted to measure the bumpiness of the road surface. If the maximum vibration amplitude exceeds a certain threshold, it may indicate that the vehicle is passing over potholes or speed bumps and other obstacles. Vibration frequency features: Analyze the frequency of the vibration and divide it into low-frequency vibrations (possibly due to long-wave undulations of the road surface) and high-frequency vibrations (possibly because of small stones or unevenness on the road surface); for example, when riding on a cobblestone road, high-frequency vibrations will be generated, while when riding on an asphalt road with gentle slopes, low-frequency vibrations may occur. Vibration energy features: By comprehensively calculating the vibration amplitude and frequency, the vibration energy is obtained; a large vibration energy indicates poor riding comfort and greater impact on the vehicle and components, which is of great significance for evaluating the wear and service life of vehicle components. Pedaling frequency features are further divided into:

[0093] Average pedaling frequency: Calculate the average pedaling frequency during the entire ride to measure the overall riding intensity of the rider. For example, the average pedaling frequency may be between 60 - 80 revolutions per minute during casual riding, while in intense competition scenarios, the average pedaling frequency may exceed 100 revolutions per minute.

[0094] Pedaling frequency change trend: Observe whether the pedaling frequency gradually increases (such as during the acceleration process), gradually decreases (such as when fatigued), or remains stable, so as to analyze the rider's physical strength distribution and riding intention. For example, during the sprint stage, the rider will quickly increase the pedaling frequency.

[0095] Pedaling frequency distribution characteristics: Count the number of occurrences or the time proportion in different pedaling frequency intervals to understand at what pedaling frequencies cyclists tend to ride. For example, some cyclists may ride in the interval of 70 - 90 revolutions per minute for most of the time.

[0096] Pedaling force characteristics, which can be further divided into:

[0097] Average pedaling force: Calculate the average pedaling force during the ride. Combining with the pedaling frequency, the rider's output power can be roughly estimated. For example, a larger average pedaling force and a higher pedaling frequency mean a higher output power, which may indicate that the cyclist is riding fast or climbing a slope.

[0098] Range of pedaling force variation: Extract the difference between the maximum and minimum pedaling forces to measure the unevenness of the rider's exertion. During the starting and accelerating phases, the range of pedaling force variation is usually large, while it is relatively small during uniform riding.

[0099] Peak pedaling force characteristics: Record the time and frequency of the peak pedaling force. The peak may occur during acceleration, climbing, or sprinting phases. These peak characteristics are important for analyzing the rider's power output strategy.

[0100] Speed change rate characteristics, which can be further divided into:

[0101] Characteristics of acceleration and deceleration phases: Determine the time periods of acceleration and deceleration, and analyze their durations and magnitudes. For example, rapid acceleration may indicate that the cyclist is rushing for a green light or overtaking other vehicles, while sudden deceleration may be to avoid pedestrians.

[0102] Average acceleration and deceleration: Calculate the average acceleration and deceleration during the entire ride to evaluate the aggressiveness of the rider's cycling. In a racing scenario, a cyclist may have a higher average acceleration, while in an urban commuting scenario, the average acceleration is relatively low.

[0103] Acceleration change frequency: Count the number of acceleration changes per unit time. A high acceleration change frequency may mean a complex riding environment where the speed needs to be adjusted frequently, such as riding on a road with heavy traffic flow.

[0104] Travel direction change rate characteristics, which can be further divided into:

[0105] Turning angle characteristics: Extract the size of each turning angle to determine whether it is a small-angle turn (such as avoiding roadside obstacles) or a large-angle turn (such as turning at an intersection). In complex urban roads, small-angle turns are more common, while on suburban roads, large-angle turns may be more frequent.

[0106] Turning frequency feature: Calculate the number of turns within a unit of time, which is used to describe the tortuosity of the cycling route. For example, when cycling on campus or in a residential area, due to the complex road layout and numerous obstacles, the turning frequency will be higher than that on a straight road with fewer obstacles.

[0107] Turning speed feature: Combine speed and turning angle data to analyze the speed change during the turning process. If the speed remains at a relatively high level during turning, it may indicate that the cyclist has high cycling skills or is familiar with the road conditions.

[0108] Normalize the above-extracted features. Normalization is to transform the data according to certain rules so that the data falls into a specific interval, usually the [0,1] or [-1,1] interval. Its main purpose is to eliminate the influence of the dimension (unit) between data features, make different features comparable, and help some algorithms (such as gradient descent method) converge faster.

[0109] Meteorological data contains rich information that can affect cycling. Multiple features can be extracted, and these features are of great value in meteorological research on cycling behavior. The following explains the features that can be extracted from the perspective of different meteorological element data:

[0110] Temperature data

[0111] Basic statistical features: Include daily average, monthly average, annual average temperature, the highest and lowest temperatures and their occurrence times, etc. For example, by analyzing the average highest temperature in July over the years in a certain area, the general situation of the summer heat in this area can be understood.

[0112] Temperature change features: Such as the diurnal temperature range, which is the difference between the highest and lowest temperatures in a day, reflecting the temperature fluctuation amplitude within a day; the seasonal temperature difference, reflecting the temperature difference between different seasons. By observing the change trend of temperature over a period of time, such as the warming or cooling rate, it can be judged whether the climate has abnormal changes.

[0113] Precipitation data

[0114] Precipitation amount features: There are total amount indicators such as daily precipitation, monthly precipitation, and annual precipitation, as well as precipitation intensity, that is, the amount of precipitation per unit time (such as the distinction between heavy rain and light rain). By statistically calculating the proportion of heavy rain precipitation in the annual precipitation of a certain area, the contribution of heavy precipitation to the total precipitation in this area can be evaluated.

[0115] Precipitation frequency and persistence: Precipitation frequency refers to the number of precipitation occurrences within a certain period of time; precipitation persistence represents the duration of a precipitation process. For example, by studying the number of times of continuous precipitation exceeding 5 days in a certain area, the potential risk of flood disasters in this area can be understood.

[0116] Humidity data

[0117] Absolute humidity and relative humidity: Absolute humidity reflects the mass of water vapor contained in a unit volume of air; relative humidity refers to the percentage of the actual water vapor pressure in the air to the saturated water vapor pressure at the same temperature, which can better reflect the impact of the humidity of the air on human comfort and the moisture condition of objects. Analyzing the changes in relative humidity in different seasons can help understand the impact of the dry-wet condition of the climate on life and production.

[0118] Dew point temperature: Under the condition that the water vapor content in the air remains unchanged and the air pressure is kept constant, the temperature at which the air cools to saturation. The difference between the dew point temperature and the current air temperature can reflect the degree of the air's proximity to water vapor condensation, which is of great significance for predicting phenomena such as precipitation and dew.

[0119] Wind speed and wind direction data

[0120] Average wind speed and maximum wind speed: The average wind speed describes the average flow velocity of the wind over a period of time, and the maximum wind speed reflects the strongest wind force during that period, which is crucial for evaluating the impact of the wind on buildings, agriculture, etc. For example, when selecting a site for a wind farm, the average wind speed needs to be considered as a key factor to ensure the availability of wind energy resources.

[0121] Characteristics of the wind rose diagram: By drawing a wind rose diagram, the frequency of different wind directions and the average wind speed of each wind direction can be visually displayed. The dominant wind direction in the area can be seen from the diagram, which has guiding significance for the layout of industrial areas and residential areas in urban planning and the analysis of the diffusion of atmospheric pollutants.

[0122] Atmospheric pressure data

[0123] Sea-level atmospheric pressure: The atmospheric pressure values at different altitudes in various places are uniformly corrected to the sea-level altitude, which is convenient for global comparison and analysis. Analyzing the distribution of the sea-level atmospheric pressure field can help understand the atmospheric circulation situation, such as the position and intensity of high and low pressure centers, which is of great significance for weather forecasting and climate research.

[0124] Trend of atmospheric pressure change: The rising or falling trend of atmospheric pressure is often related to the movement and evolution of weather systems. For example, a continuous decrease in atmospheric pressure may indicate the approach of a cyclone and the weather will turn bad; while a gradual increase in atmospheric pressure may indicate the control of an anticyclone and the weather tends to be clear.

[0125] Cloud data

[0126] Cloud cover: It refers to the proportion of the sky obscured by clouds, divided into total cloud cover and low cloud cover. Total cloud cover reflects the overall degree of the sky covered by clouds, which has an important impact on the shielding of solar radiation and the return of the earth's long-wave radiation, and thus affects the ground air temperature. For example, in cloudy weather, solar radiation is weaker and the ground warms up less significantly.

[0127] Types of clouds: Different types of clouds (such as cirrus clouds, cumulus clouds, stratus clouds, etc.) are associated with different weather phenomena. For example, cumulonimbus clouds are usually accompanied by severe convective weather such as thunderstorms and heavy rains. By identifying the types of clouds, it can assist in weather forecasting.

[0128] Radiation data

[0129] Solar radiation: It includes total solar radiation (the total amount of solar radiation reaching the Earth's surface), direct solar radiation (solar radiation reaching the ground directly without passing through atmospheric scattering), and scattered solar radiation (solar radiation reaching the ground after passing through atmospheric scattering). Analyzing the seasonal changes in the amount of solar radiation can help understand the differences in solar energy received by the Earth's surface in different seasons, which is of great significance for fields such as solar energy utilization and agricultural production.

[0130] Longwave radiation: Infrared radiation emitted by the Earth's surface and the atmosphere. The balance of longwave radiation affects the Earth's energy balance and climate system. By studying the characteristics of longwave radiation, the change mechanism of the Earth's climate can be understood in depth.

[0131] Encode categorical data. For example, encode gender as 0 (male) and 1 (female), and encode regional type as 1 (commercial area), 2 (residential area), 3 (school area), etc. Traffic flow level data can also be encoded in this way for model processing, or other distinguishable coding modes can be adopted.

[0132] Through these features, the system can better understand the riding behavior patterns of cyclists in different regions and different groups, providing a richer basis for subsequent matching and anomaly determination.

[0133] S42: Build an MLP learning model and, before the system runs, train and validate it with a large amount of historical riding data in the historical dataset so that the MLP learning model can learn the riding characteristics of different groups.

[0134] When constructing a multi-layer perceptron (MLP) learning model, a suitable network structure is adopted, such as including an input layer, two hidden layers, and an output layer. The number of nodes in the input layer is determined according to the number of features after preprocessing. For example, the encoded features such as age, gender, and region mentioned above, as well as the continuous features such as the normalized speed and acceleration. For instance, if there are a total of 10 features, the number of nodes in the input layer is 10. The number of nodes in the hidden layers is determined through experimental debugging and can be set to 20 and 15 respectively. The number of nodes in the output layer corresponds to the category of group features to be learned. For example, classified by age group into young people (18 - 35 years old), middle-aged people (36 - 55 years old), and elderly people (56 years old and above), a total of 3 categories, 2 genders, time periods divided into morning rush hour on weekdays (7:00 - 9:00), evening rush hour (17:00 - 19:00), and off-peak period (other time periods), a total of 3 categories, and 3 regions, totaling 3×2×3×3 = 54 categories. The number of nodes in the output layer is 54. Other output node numbers can also be used according to actual needs.

[0135] Use the historical cycling data in the preprocessed historical dataset to train the MLP learning model. Adopt the stochastic gradient descent algorithm and set appropriate learning rates (such as 0.01) and the number of iterations (such as 1000 times) to minimize the cross-entropy loss function. During the training process, record the accuracy of the model on the validation set (the validation set divided from the historical dataset at an 8:2 ratio) every 100 iterations. When the accuracy no longer improves, stop the training.

[0136] S5: Obtain new cycling data and extract the corresponding data preprocessing, feature vector extraction, and classification encoding, and input them into the MLP learning model for group feature matching. Specifically, it includes the following steps:

[0137] S51: Collect, preprocess, extract feature vectors from new cycling data, and encode the classification data. When new cycling data enters the data processing center, collect various data of cyclists at the same frequency (every 10 seconds), including information such as age, gender, current time, location area, speed, acceleration, and traffic flow. Process the newly collected data according to the data preprocessing method in the training stage, extract feature vectors, and prepare to input them into the MLP model and the subsequent anomaly determination module.

[0138] S52: Input the feature vectors and classification encoding extracted from the new cycling data into the trained MLP learning model, and the model outputs the probability distribution of the group category to which the current cyclist belongs. For example, if the output result shows that the current cyclist has a 0.6 probability of belonging to the group of young people, business district, and morning rush hour on weekdays, a 0.3 probability of belonging to the group of middle-aged people, residential area, and off-peak period, and a 0.1 probability of belonging to other groups, then select the group category with the highest probability as the matching group for the current cyclist.

[0139] S6: Determine whether the current rider's behavior conforms to the normal riding pattern of their group. If not, it is an abnormal ride.

[0140] Use a dynamic threshold mechanism and the HMM model to identify and determine the rider's behavior.

[0141] HMM, i.e., Hidden Markov Model, is a statistical model used to describe a Markov process with hidden unknown parameters. It is mainly used to process time series data or sequence signals, etc., and has extensive applications in many fields such as speech recognition, natural language processing, and bioinformatics. Under HMM pattern recognition, riding data (such as speed, acceleration, etc.) has temporal characteristics, and HMM can handle this kind of time series data well. HMM models the hidden states of the data (such as the rider's intention, behavior pattern), and based on historical data and a pre-trained model, calculates the probabilities of different riding behaviors in the current state. By analyzing the speed change pattern of the rider over a period of time, it can accurately identify whether the rider has abnormal acceleration or deceleration behavior. If the calculated probability of abnormal behavior exceeds a certain threshold, it is determined that the riding behavior is abnormal, providing reliable data support for the entire warning system. The specific steps are as follows:

[0142] S61: Dynamically adjust the abnormal threshold according to the characteristics of the current rider, road conditions, and environment, etc.

[0143] Traditional static thresholds have limitations in judging whether a riding behavior is abnormal because the range of normal riding behaviors is different under different riding scenarios and group characteristics. The data processing center adopts a dynamic threshold mechanism and dynamically adjusts the threshold for judging whether a riding behavior is abnormal according to factors such as the characteristics of the current rider (such as age, gender), the current road conditions (such as flat road, uphill, downhill), etc. In the case of poor road conditions, the normal speed of the rider may be lower than that on a flat road, and the data processing center will dynamically adjust the speed threshold according to these situations to judge whether there is speeding. This dynamic adjustment can more accurately reflect the actual riding situation and reduce misjudgments. For example, the current rider is a 25-year-old male, riding in a commercial area, the road condition is a flat road, and the traffic flow is medium. Based on historical data statistical analysis, under similar conditions, the average riding speed of this group is 18 km / h, and the standard deviation is 2 km / h. Usually, the normal range is set as the average speed plus or minus 1.5 times the standard deviation, that is, the lower limit is 15 km / h and the upper limit is 21 km / h, which is used as the current speed threshold. If the speed of the current rider exceeds 21 km / h at a certain moment, it is initially determined that there may be an abnormal speeding behavior. For the abnormal acceleration determination, similarly based on historical data, in this group, road conditions, and environment, the normal acceleration range is between -0.5 and 0.5 m / s2 If the detected cyclist acceleration exceeds this range, such as a sudden hard brake resulting in an acceleration of -1 m / s 2 , it is also marked as a possible abnormal behavior.

[0144] Assign a weight to each factor, and determine the weight size according to the degree of association between each factor and abnormal cycling behavior. The greater the association, the higher the weight. Each factor is first standardized and transformed into the range of 0-1 to eliminate the influence of dimension.

[0145] The abnormal threshold calculation formula is as follows:

[0146]

[0147] Among them, y is the abnormal threshold, w i is the weight of the i-th factor; x i is the value after standardizing the i-th factor. For example:

[0148] Age factor: Assuming that young people (18-35 years old) generally ride faster, the age standardized value x 年龄 , 18 years old is 1, 35 years old is 0, and the older the age, the smaller the value. If it is considered that age is more important for judging cycling abnormalities, the weight w 年龄 = 0.15.

[0149] Road condition factor: The flat road x i is 0, the uphill x i is 0.5, and the downhill x i is 1. If the road condition has a greater impact, the weight w 路况 = 0.2.

[0150] Area type factor: There are many people and it is complex in the business district, x i is set to 1; the residential area is relatively regular, x i is set to 0.6; the school area is complex during school arrival and dismissal times, x i is set to 0.8. The weight w 区域类型 = 0.15.

[0151] Traffic flow level factor: Low flow x i is 0, medium flow x i is 0.5, and high flow x i is 1. The weight w 交通流量 = 0.2.

[0152] Meteorological factor: Calculate a value x between 0-1 by integrating rain, wind, light intensity, fog visibility, temperature and humidity, etc. 气象 . The weight w 气象 = 0.3.

[0153] Furthermore, x气象 The calculation formula is as follows:

[0154]

[0155] Among them, RainScore, WindScore, LightScore, FogScore, TempScore, and HumScore are the values obtained after scoring for rain, wind, light intensity, fog visibility, temperature, and humidity respectively, and the value range is between 0 and 1; W rain、 W wind、 W light W fog、 W temp、 W hum are the weight coefficients of each factor, and the values are determined according to specific scenarios and requirements. For example, in the travel scenario, rain may have a greater impact on travel, and the weight of W rain can be set to 0.3, while the impact of light intensity is relatively small, and W light is set to 0.1.

[0156] Examples of scoring for each factor:

[0157] Rain (RainScore):

[0158] No rain: 1

[0159] Light rain (rainfall < 10 mm / day): 0.8

[0160] Moderate rain (10 - 25 mm / day): 0.6

[0161] Heavy rain (25 - 50 mm / day): 0.4

[0162] Rainstorm (> 50 mm / day): 0.2

[0163] Wind (WindScore):

[0164] Calm (wind speed < 0.3 m / s): 1

[0165] Light air (0.3 - 1.5 m / s): 0.9

[0166] Gentle breeze (1.6 - 3.3 m / s): 0.8

[0167] Moderate breeze (3.4 - 5.4 m / s): 0.7

[0168] Fresh breeze (5.5 - 7.9 m / s): 0.6

[0169] Strong breeze (8.0 - 10.7 m / s): 0.5

[0170] Near gale (10.8 - 13.8 m / s): 0.4

[0171] Strong wind (13.9 - 17.1 m / s): 0.3

[0172] Gale (17.2 - 20.7 m / s): 0.2

[0173] Violent wind (20.8 - 24.4 m / s): 0.1

[0174] Storm (24.5 - 28.4 m / s): 0.05

[0175] Hurricane (> 28.4 m / s): 0

[0176] Light intensity (LightScore):

[0177] Very strong (direct sunlight at noon on a sunny day, > 100000 lux): 0.8

[0178] Stronger (slanting sunlight in the morning / afternoon on a sunny day, 10000 - 100000 lux): 0.9

[0179] Moderate (diffused light on a cloudy day, 1000 - 10000 lux): 1

[0180] Weaker (around sunrise / sunset, 10 - 1000 lux): 0.9

[0181] Very weak (at night, < 10 lux): 0.8

[0182] Fog visibility (FogScore):

[0183] Visibility > 1000 m: 1

[0184] 500 - 1000 m: 0.8

[0185] 200 - 500 m: 0.6

[0186] 50 - 200 m: 0.4

[0187] < 50 m: 0.2

[0188] Temperature (TempScore):

[0189] Very comfortable (18 - 25 °C): 1

[0190] Relatively comfortable (15 - 18 °C or 25 - 28 °C): 0.9

[0191] Relatively hot (28 - 32 °C): 0.8

[0192] Hot (32 - 38 °C): 0.6

[0193] Scorching (> 38 °C): 0.4

[0194] Relatively cold (10 - 15 °C): 0.8

[0195] Cold (5 - 10 °C): 0.6

[0196] Severely cold (<5 °C): 0.4

[0197] Humidity (HumScore):

[0198] Comfortable (40% - 60%): 1

[0199] Relatively comfortable (30% - 40% or 60% - 70%): 0.9

[0200] Dry (20% - 30%): 0.8

[0201] Very dry (<20%): 0.6

[0202] Humid (70% - 80%): 0.8

[0203] Very humid (>80%): 0.6.

[0204] S62: According to the anomaly threshold, use a pre-trained Hidden Markov Model (HMM) to determine whether the cycling behavior conforms to the normal cycling pattern of its group. If not, it is determined as a cycling anomaly. This model is trained with cycling data based on the above historical dataset, and models time series data such as speed and acceleration for different groups. For example, for the group of young people traveling during the early morning rush hour on weekdays in the business district, the Hidden Markov Model (HMM) learns their typical speed change pattern: starting from a speed of 0, gradually accelerating to about 20 km / h, maintaining a stable speed for a few minutes, then decelerating to about 10 km / h at intersections, and then accelerating again to pass through. When a new cycling data sequence is input into the Hidden Markov Model (HMM), the model calculates the probabilities of different cycling behaviors in the current state. Suppose the speed change of a cyclist in the past 5 minutes (30 data points, one every 10 seconds) is analyzed, and the probability of a sudden and unjustified continuous acceleration (the speed increases from 15 km / h to 30 km / h within 2 minutes, exceeding the normal acceleration pattern of this group) is calculated to be 0.8. If the anomaly threshold probability is set to 0.7, then this cycling behavior is determined to be abnormal.

[0205] S7: Determine the type and weight of the cycling anomaly, and select the corresponding warning level according to the corresponding weight to generate a warning decision instruction.

[0206] An early warning decision module is set up to receive data from the data processing center that has been cleaned, matched, and preliminarily judged as abnormal. Then, the final early warning decision is made according to the preset rules and algorithms, and the corresponding instructions are generated to start the subsequent early warning push process. Abnormal behavior weight determination: A set of weight determination rules are preset in the early warning decision module. These rules are set based on the evaluation of the impact of different abnormal riding behaviors on riding safety. For example, serious speeding (such as exceeding the average speed of the group by more than 2 standard deviations), driving against the flow in the motor vehicle lane, and illegal carrying of passengers are considered to be behaviors that pose a major threat to riding safety and will be given a higher weight, while relatively less serious behaviors such as short-term small-scale speeding are given a lower weight. For each abnormal behavior, the module calculates the weight according to the degree of deviation from the normal riding mode and the potential danger. For example, when the data shows that the rider is speeding, the module will calculate the weight of the speeding behavior based on factors such as the speeding amplitude (such as the proportion exceeding the normal speed range) and duration. At the same time, for situations where there are multiple abnormal behaviors concurrently, such as speeding and single-handed vehicle control, the module will comprehensively consider the weights of these behaviors for superposition or weighted calculation.

[0207] The specific steps are as follows:

[0208] S71: Determine the type of riding abnormality and calculate the corresponding weight.

[0209] 1. Classification and weights of serious threatening behaviors. Severe speeding: When the speed of a cyclist exceeds the average speed of the group by more than 2 standard deviations, it is judged as severe speeding. For example, for middle-aged people, residential areas, and off-peak travel groups, the average riding speed is 16 kilometers per hour and the standard deviation is 2 kilometers per hour. If the speed of the cyclist reaches 20 kilometers per hour or more, it is considered severe speeding. Such behaviors are given an initial weight of 0.8, indicating that they pose a very high threat to cycling safety. Going against the flow in the motor vehicle lane: Once it is detected that the riding trajectory is opposite to the normal driving direction of the motor vehicle and lasts for more than 30 seconds, it is judged as going against the flow in the motor vehicle lane. This behavior directly puts the cyclist in danger of a high-speed collision with a motor vehicle and is given a weight of 0.9. Illegal carrying of passengers: If the riding vehicle is designed for single-person riding but carries passengers, regardless of the length of the journey, it is considered as illegal carrying of passengers and is given a weight of 0.7, because this will affect the riding controllability and stability and increase the risk of accidents.

[0210] 2. Moderate Threat Behavior Classification and Weights. General speeding: The speed of the rider exceeds the average speed of the group by 1 standard deviation but does not reach 2 standard deviations, such as the middle-aged, residential, and off-peak groups mentioned above, and the speed is between 18-20 km / h. This is considered general speeding, and the initial weight is set to 0.4. Frequent sudden braking or acceleration: Within 5 minutes, sudden braking (acceleration less than -1m / s 2) or rapid acceleration (acceleration greater than 1 m / s 2 ) If the number of occurrences exceeds 3 times, it is determined as frequent hard braking or rapid acceleration, and a weight of 0.5 is assigned. This can easily cause the rider to lose balance or the following vehicle to fail to avoid in time.

[0211] 3. Classification and weights of mild threat behaviors. Short-term and small-scale speeding: The speed exceeds within 0.5 times the standard deviation of the average speed of the group and the duration does not exceed 1 minute. For example, for groups such as young people, commercial areas, and the morning rush hour on weekdays, the average speed is 18 km / h and the standard deviation is 2 km / h. If the speed is within 19 km / h and the speeding time is short, the weight is set to 0.2.

[0212] S72: According to the calculated weights of abnormal behaviors, the group warning decision module divides different warning levels and generates corresponding decision instructions. Usually, it can be divided into three levels: low, medium, and high. For example, 0 - 30 points is low risk, corresponding to a low-level warning; 31 - 60 points is medium risk, triggering a medium-level warning; 61 points and above is high risk, initiating a high-level warning. The generated decision instructions contain detailed warning information, including warning level, abnormal type (such as speeding, illegal operation, etc.), description of the abnormal degree, vehicle number, rider identification, and current location information, etc. For example, the instruction may be "High-risk warning: Rider [identification] of vehicle [number] is seriously speeding at [specific speed] at [location]. Please take immediate measures."

[0213] S8: Select the corresponding warning channels for warning according to the warning decision instructions.

[0214] (1) Vehicle central control voice reminder. The voice synthesis system will convert the text of abnormal information to be reminded, such as "You are speeding. Please slow down", etc., into voice signals through voice synthesis technology. Usually, pre-recorded voice segments are spliced, or a voice synthesis model based on deep learning is used to generate natural and fluent voices according to the text content.

[0215] (2) Operation platform reminder. When an abnormal situation occurs, the warning system encapsulates data such as abnormal information, vehicle information, and rider information into data packets in a specified format (the specified format can adopt existing formats), and transmits them to the operation platform through the network. After receiving the data, the operation platform stores it in the background database and displays it in different ways on the interface of the operation platform according to the type and level of the abnormality. For example, low-level abnormalities may be marked with a specific color in the data report, while high-level abnormalities pop up a warning window to remind the operation personnel to handle them in time. At the same time, the operation platform can also analyze and statistically process the abnormal data to provide a basis for operation decisions.

[0216] (3) Mobile App Message Push Mechanism. A long connection is established between the mobile App and the server, and a push notification service (such as APNs for Apple or FCM for Android) is adopted. When the early warning system detects abnormal cycling, it sends a push request to the server. The server encapsulates the abnormal information in the format supported by the mobile App and pushes the message to the user's mobile App through the push notification service. After receiving the push message, the mobile App on the user's mobile phone gives local reminders according to the settings, such as vibration, sound prompt, etc. At the same time, the abnormal information is displayed on the interface of the mobile App in the form of pop-up windows, message lists, etc. The user can click to view the detailed content, such as the specific location, type, and recommended handling method of the abnormality.

[0217] (4) User Mobile Phone SMS Reminder. The early warning system is docked with the SMS gateway, and the SMS gateway is connected to the SMS platforms of major telecom operators. When an SMS reminder needs to be sent to the user, the early warning system sends information such as the SMS content and the user's mobile phone number to the SMS gateway. The SMS gateway encodes and converts the SMS content according to the operator's protocol and then sends it to the corresponding operator's SMS platform. The operator's SMS platform is responsible for sending the SMS to the user's mobile phone. After receiving the SMS, the user's mobile phone can view the specific information containing the abnormal situation, such as "You have exceeded the speed limit during cycling. Please abide by traffic rules."

[0218] (5) Collaborative Processing and Law Enforcement Application by Traffic Management Departments. After receiving the data, the system of the traffic management department conducts legality verification and data parsing, and integrates the abnormal information with the traffic management business process. For example, an illegal record is automatically generated in the traffic violation processing system, and the real-time location and information of the abnormal vehicle are provided to the traffic police in the command and dispatch system so that the traffic police can go to the scene for handling in a timely manner, realizing the collaborative management and law enforcement of abnormal cycling behaviors of shared electric bicycles.

[0219] The beneficial effects of the solution of this application are as follows:

[0220] This application is a method for abnormal cycling safety early warning of shared electric bicycles based on group data.

[0221] In terms of cycling safety guarantee, by using technologies such as group data matching, dynamic threshold mechanism, and HMM pattern recognition, compared with existing technologies, it can identify cycling abnormal behaviors more accurately. Existing technologies may only rely on simple speed or position threshold judgments, which are prone to misjudgments. This solution fully considers the characteristics of different cyclist groups (such as age, gender, cycling time period, region, etc.) and cycling scenarios (such as road conditions, weather, etc.), reduces misjudgments and missed judgments, discovers and prevents cycling safety hazards in a timely manner, and guarantees the safety of cyclists. The system conducts early warning push through multiple channels such as vehicle central control voice reminder, App, SMS, traffic management department, etc. Existing technologies may only rely on a single reminder method, and it is easy for cyclists not to receive early warning information in a timely manner. This solution ensures that cyclists, operation platforms, and traffic management departments can all obtain abnormal information in a timely manner. Cyclists can immediately correct their behaviors, operation platforms can take measures such as remote locking, power off, and speed reduction in a timely manner, and traffic management departments can enforce the law in a timely manner, comprehensively guaranteeing cycling safety.

[0222] In terms of data processing and transmission, the 5G link management module is preferred for data transmission. Compared with the 4G or lower-speed networks that may be used in existing technologies, the 5G network has the characteristics of high speed and low latency, ensuring that data is quickly and stably transmitted from the in-vehicle terminal to the data processing center. And when the 5G signal is poor, the network can be switched to avoid untimely early warnings caused by data transmission delays and ensure the coherence of data transmission. The data processing center uses technologies such as CRC check, noise data filtering, and ARIMA model data missing processing, and reduces data redundancy through the MQTT protocol. Existing technologies may lack these data processing means, which are prone to inaccurate and incomplete data and low data processing efficiency. This solution guarantees the accuracy and integrity of data, improves data processing and storage efficiency, reduces data processing costs, and efficiently processes a large amount of cycling data.

[0223] In terms of operation management, the early warning decision module determines the weight of abnormal behaviors and generates decision instructions. Existing technologies may lack refined analysis and decision-making for abnormal behaviors. According to the instructions of this solution, the operation platform can conduct intelligent management of vehicles and users. For example, it can perform early maintenance on vehicles with frequent abnormalities, restrict the use of users with many violations, optimize operation strategies, and improve operation management efficiency and scientificity. Through accurate early warnings and multi-channel push, the operation platform can manage vehicles and users in a targeted manner, reducing unnecessary on-site inspections and manual interventions. At the same time, the reasonable strategy of the power module when the battery power is low (such as reducing the sampling frequency of non-critical sensors) prolongs the device's battery life, reduces the battery replacement frequency, and reduces operation and maintenance costs, having a significant advantage over existing technologies in terms of operation and maintenance costs.

[0224] In terms of urban traffic order, information on abnormal riding events with serious violations is promptly pushed to the traffic management department. Existing technologies may not be able to transmit shared e-bike violation information to the traffic management department in a timely manner, resulting in untimely traffic law enforcement. This solution helps traffic police promptly grasp violation situations and enforce the law, reduce shared e-bike violations, maintain urban traffic order, and ensure public traffic safety. A large amount of riding data (including normal and abnormal behavior data) accumulated by the system can provide valuable data support for urban traffic planning and management. Existing technologies may not be able to make full use of riding data. By analyzing this data, the present invention can understand riders' travel habits, popular routes, high-incidence areas of violations, etc., providing a reference basis for optimizing urban traffic facilities (such as setting dedicated bike lanes, traffic signs, etc.) and formulating traffic management policies.

[0225] Embodiment 2

[0226] A shared e-bike riding abnormal safety warning system based on group data, which is applied to the above-mentioned shared e-bike riding abnormal safety warning method based on group data, includes:

[0227] An acquisition module, which is used to acquire data from in-vehicle terminals;

[0228] A communication management module, which is used to summarize in-vehicle terminal data and obtain corresponding rider data, regional type data, traffic flow level data, and meteorological data during riding, and package them to form group data;

[0229] A data processing center, which is used to perform cleaning and preprocessing on group data such as verification, unified format, noise filtering, and missing value processing; perform feature extraction, encoding on the cleaned group data, and use it for MLP learning model training; obtain new riding data and extract corresponding data preprocessing, feature vector extraction, and classification encoding, and input them into the MLP learning model for group feature matching; determine whether the behavior of the current rider conforms to the normal riding mode of the group he belongs to, and if not, it is an abnormal riding.

[0230] An early warning decision-making module, which is used to judge the type and weight of riding abnormalities, and select the corresponding early warning level to generate an early warning decision-making instruction according to the corresponding weight; select the corresponding early warning channel for early warning according to the early warning decision-making instruction.

[0231] Embodiment 3

[0232] A computer-readable storage medium, the computer-readable storage medium includes a stored program, wherein, when the program runs, it controls the device where the computer-readable storage medium is located to execute the above-mentioned shared e-bike riding abnormal safety warning method based on group data.

[0233] Embodiment 4

[0234] A processor is used to run a program. When the program runs, it executes the above-mentioned method for sharing electric bike riding abnormal safety warning based on population data.

[0235] This application provides a method for sharing electric bike riding abnormal safety warning based on population data, including the following steps: S1: Collect data from in-vehicle terminals; S2: Summarize the in-vehicle terminal data and obtain the corresponding rider data, regional type data, traffic flow level data, and meteorological data during riding, and package them to form population data; S3: Perform cleaning and preprocessing on the population data, such as verification, unified format, noise filtering, and missing value processing; S4: Extract features, encode the cleaned population data, and use it for training the MLP learning model; S5: Obtain new riding data, extract the corresponding data preprocessing, feature vector extraction, and classification encoding, and input them into the MLP learning model for population feature matching; S6: Determine whether the behavior of the current rider conforms to the normal riding mode of the group he belongs to. If not, it is an abnormal riding. In terms of riding safety guarantee, by using technologies such as population data matching, dynamic threshold mechanism, and HMM pattern recognition, compared with the prior art, it can more accurately identify abnormal riding behaviors. The prior art may only rely on simple speed or position thresholds for judgment, which is prone to misjudgment. This solution fully considers the characteristics of different rider groups (such as age, gender, riding time period, region, etc.) and riding scenarios (such as road conditions, weather, etc.), reduces misjudgment and missed judgment, discovers and prevents riding safety hazards in a timely manner, and guarantees the safety of riders. The system pushes warnings through multiple channels such as vehicle central control voice reminders, Apps, text messages, and traffic management departments. The prior art may only rely on a single reminder method, and it is easy for riders not to receive warning information in a timely manner. This solution ensures that riders, operation platforms, and traffic management departments can all obtain abnormal information in a timely manner. Riders can immediately correct their behaviors, operation platforms can take measures such as remote locking, power-off, and speed reduction in a timely manner, and traffic management departments can enforce the law in a timely manner, comprehensively guaranteeing riding safety.

[0236] Those of ordinary skill in the art can realize that the units of each example described in combination with the embodiments disclosed herein can be implemented by electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components of each example have been generally described according to their functions in the above description. Whether these functions are executed in a hardware or software manner depends on the specific application and design constraints of the technical solution. Professional technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of the present invention.

[0237] In the embodiments provided by the present invention, it should be understood that the division of units is only a logical function division. In actual implementation, there may be other division methods. For example, multiple units can be combined into one unit, one unit can be split into multiple units, or some features can be ignored, etc.

[0238] In addition, each functional unit in various embodiments of the present invention can be integrated in a processing unit, or each unit can exist physically alone, or two or more units can be integrated in one unit. The above-mentioned integrated units can be implemented in the form of hardware or in the form of software functional units.

[0239] If the above-mentioned integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it 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 makes a contribution to the prior art, or all or 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 can 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, read-only memories (ROMs), random access memories (RAMs), mobile hard disks, magnetic disks, or optical discs that can store program codes.

[0240] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit 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 perform equivalent replacements on some or all of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of various embodiments of the present invention, and they should all be covered by the scope of the claims and the description of the present invention.

Claims

1. A method for early warning of abnormal safety of shared electric motorcycle riding based on group data, characterized in that: The following steps are involved: S1: collect data from the vehicle terminal; S2: Summarize the vehicle terminal data and obtain the corresponding rider data, regional type data, traffic flow level data and meteorological data during riding, and package them to form group data; S3: Perform cleaning preprocessing such as verification, unified format, noise filtering, and missing value processing on group data; S4: Extract and encode the cleaned group data and use it for MLP learning model training; S5: Obtain new riding data and extract corresponding data preprocessing, feature vector extraction and classification coding, and input them into the MLP learning model for group feature matching; S6: Determine whether the current rider's behavior conforms to the normal riding pattern of his group. If not, it is considered as riding abnormality.

2. The method for early warning of abnormal safety of shared electric motorcycle riding based on group data according to claim 1 is characterized in that: The step S4 includes the following steps: S41: for the cleaned data, normalize and extract features for the numerical data, and encode the categorical data to form a historical data set; S42: Construct an MLP learning model and train and verify it through historical riding data in the historical data set, so that the MLP learning model can learn the riding characteristics of different groups.

3. The method for early warning of abnormal safety of shared electric motorcycle riding based on group data according to claim 1 is characterized in that: The step S5 includes the following steps: S51: New riding data collection, preprocessing and feature vector extraction, encoding of categorical data; S52: input the feature vector and classification code extracted from the new riding data into the trained MLP learning model to calculate the probability distribution of the group category to which the current rider belongs; S53: Select the group category with the highest probability as the matching group for the current rider.

4. The method for early warning of abnormal safety of shared electric motorcycle riding based on group data according to claim 1 is characterized in that: Step S6 includes the following steps: S61: dynamically adjusting the abnormal threshold according to the current rider's characteristics, road conditions and environmental conditions; S62: Based on the abnormal threshold, a pre-trained Hidden Markov Model (HMM) is used to determine whether the riding behavior conforms to the normal riding pattern of the group to which the rider belongs. If not, the riding behavior is determined to be abnormal.

5. The method for early warning of abnormal safety of shared electric motorcycle riding based on group data according to claim 4 is characterized in that: The abnormal threshold calculation formula is as follows: y=∑ n i＝1 w i ×x i ; Among them, y is the abnormal threshold, w i is the weight of the i-th factor; x i is the standardized value of the i-th factor; n is the number of factors affecting cycling, including age, road conditions, area type, traffic flow level and meteorological factors.

6. The method for early warning of abnormal safety of shared electric motorcycle riding based on group data according to claim 1 is characterized in that: The following steps are also included: S7: Determine the type and weight of the riding abnormality, and select the corresponding warning level according to the corresponding weight to generate a warning decision instruction; S8: Select the corresponding warning channel for warning according to the warning decision instruction.

7. The method for early warning of abnormal safety of shared electric motorcycle riding based on group data according to claim 6 is characterized in that: The step S7 includes the following steps: S71: Determine the type of riding abnormality and calculate the corresponding weight; S72: According to the calculated abnormal behavior weights, different warning levels are divided and corresponding decision instructions are generated.

8. A shared electric motorcycle riding abnormal safety warning system based on group data, characterized in that: The method for early warning of abnormal safety of shared electric motorcycle riding based on group data as described in any one of claims 1 to 7 comprises: A collection module, which is used to collect data from the vehicle terminal; The communication management module is used to aggregate the vehicle terminal data and obtain the corresponding rider data, regional type data, traffic flow level data and meteorological data during riding, and package them to form group data; The data processing center is used to perform cleaning and preprocessing on group data, such as verification, unified format, noise filtering, and missing value processing; feature extraction and encoding of the cleaned group data, and use it for MLP learning model training; obtain new riding data and extract the corresponding data preprocessing, feature vector extraction and classification encoding, and input it into the MLP learning model for group feature matching; determine whether the current rider's behavior conforms to the normal riding mode of his group, if not, it is a riding abnormality.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium includes a stored program, wherein, when the program is running, the device where the computer-readable storage medium is located is controlled to execute the shared electric motorcycle riding abnormal safety warning method based on group data as described in any one of claims 1 to 7.

10. A processor, characterized in that: The processor is used to run a program, wherein when the program is running, the method for abnormal safety warning of shared electric motorcycle riding based on group data as described in any one of claims 1 to 7 is executed.

Images