Electric bicycle charging method and system based on mutual recognition mechanism

Abstract

The present invention provides an electric bicycle charging method and system based on a mutual recognition mechanism, which relates to the field of battery management technology, including: detecting battery status through a multi-sensor fusion algorithm, using deep reinforcement learning to predict cruising range and charging requirements, combining trajectory analysis to predict destination and arrival time, optimizing charging piles based on a multi-constraint path planning model and an improved multi-objective genetic algorithm, using double encrypted identity authentication for the target charging pile, using an adaptive charging control algorithm based on a fuzzy neural network to optimize the charging process, and uploading data to a charging scheduling server.

Description

Technical Field

[0001] The present invention relates to the technical field of battery management, and in particular to an electric bicycle charging method and system based on a mutual recognition mechanism. Background Art

[0002] Existing technologies lack accurate assessment of battery health and range. Most charging systems rely solely on simple voltage detection, which cannot accurately reflect the battery's actual performance and remaining range. This prevents users from properly planning charging times and often exposes them to the risk of sudden power outages, severely impacting their travel experience.

[0003] Traditional charging pile selection methods lack intelligence and personalization. Users usually need to manually search for nearby charging facilities and are unable to comprehensively consider multiple factors such as route planning, charging efficiency, and waiting time. This not only increases users' time costs, but may also lead to unreasonable allocation of charging resources, causing some charging piles to be idle for a long time while others are overcrowded.

[0004] Existing charging control algorithms lack adaptability and security. Most charging systems use fixed charging curves and cannot dynamically adjust based on the battery's real-time status and environmental conditions. This not only affects charging efficiency but can also lead to problems such as overcharging or undercharging, shortening battery life. Furthermore, the lack of effective identity authentication mechanisms poses security risks and fails to prevent unauthorized use and potential power theft.

[0005] Therefore, a solution is urgently needed to solve the problems existing in the prior art. Summary of the Invention

[0006] The embodiments of the present invention provide an electric bicycle charging method and system based on a mutual recognition mechanism, which can at least solve some of the problems existing in the prior art.

[0007] A first aspect of an embodiment of the present invention provides an electric bicycle charging method based on a mutual recognition mechanism, comprising:

[0008] The current location coordinates and driving trajectory data of the electric bicycle are collected, the battery status is detected through a multi-sensor fusion algorithm, and a battery health status assessment model is established. Based on the battery health status assessment model, a cruising range prediction model is constructed using a deep reinforcement learning algorithm. The model uses historical riding data, ambient temperature, road condition information, and riding habits as input features, outputs remaining cruising range and charging demand prediction results, and sends the charging demand prediction results to a charging scheduling server. The trajectory features are extracted through a hierarchical time memory structure and a spatial attention mechanism. The driving trajectory data is deeply analyzed in combination with road network topology constraints and historical trajectory similarity calculation to generate destination prediction results and arrival time prediction results;

[0009] A multi-constrained path planning model is constructed based on the current location coordinates and the destination prediction result. An improved Dijkstra algorithm is used to introduce an adaptive penalty factor to dynamically avoid congested sections. Path weights are updated online in combination with real-time traffic information to search for candidate paths. Charging piles on the candidate paths are optimized using an improved multi-objective genetic algorithm. Charging power matching, path deviation, and waiting time are used as optimization objectives. A multi-objective solution is performed using an adaptive crossover operator and a mutation operator. A Pareto solution set is generated and each solution in the Pareto solution set is scored. The charging pile with the highest score is selected as the target charging pile.

[0010] The target charging pile adopts an identity authentication mechanism based on double encryption, and authenticates the electric bicycle through a dynamic key generation algorithm and an identity feature extraction algorithm. During the charging process, an adaptive charging control algorithm based on a fuzzy neural network is used to online learn the battery charging characteristics, and a charging condition evaluation model is established in combination with temperature compensation and voltage compensation. A fuzzy inference engine is used to optimize the charging current in real time, and the charging curve is adaptively adjusted according to the battery health status. After charging is completed, the charging process data is uploaded to the charging scheduling server.

[0011] In an optional embodiment,

[0012] The current location coordinates and driving trajectory data of the electric bicycle are collected, and the battery status is detected through a multi-sensor fusion algorithm. A battery health status assessment model is established. Based on the battery health status assessment model, a deep reinforcement learning algorithm is used to build a range prediction model. The historical riding data, ambient temperature, road condition information, and riding habits are used as input features. The output of the remaining range and charging demand prediction results include:

[0013] The current position coordinates of the electric bicycle are collected according to a positioning module pre-installed in the electric bicycle, and the acceleration and angular velocity are collected as riding status data according to a pre-installed inertial measurement unit. A controller in the electric bicycle reads the data uploaded by the positioning module and the inertial measurement unit through a bus and generates corresponding driving trajectory data of the electric bicycle in combination with an electronic map;

[0014] The battery voltage, current, and temperature are collected separately according to a pre-set sensor array, wherein the sensor array includes a voltage sensor, a current sensor, and a temperature sensor. The collected signals are filtered, normalized, and feature extracted using a multi-sensor fusion algorithm to obtain a multidimensional feature vector representing the battery state. The multidimensional feature vector is input into a pre-built battery health status assessment model, wherein the battery health status assessment model is trained using a support vector machine algorithm, and the health status corresponding to the multidimensional feature vector is divided into different levels, and a corresponding battery discharge rate is assigned to each level;

[0015] Acquire historical riding data of the electric bicycle, including riding time, riding mileage, average speed and average power; and simultaneously acquire ambient temperature data collected by an ambient temperature sensor, road condition data collected by a road condition sensor, and user riding habit data, wherein the road condition data includes road type, road slope, and road curvature; and the user riding habit data includes riding time period, riding frequency, and pedaling force. Use the historical riding data, ambient temperature, road condition information, and riding habits as input features, and the battery discharge rate, remaining cruising range, and charging demand as output targets to construct a cruising range prediction model. Train the cruising range prediction model according to a deep Q learning algorithm, use a convolutional neural network as an approximate function of the Q function, perform environmental interaction, learn a mileage prediction strategy, and obtain remaining cruising range and charging demand prediction results.

[0016] In an optional embodiment,

[0017] The charging demand prediction result is sent to the charging scheduling server. Trajectory features are extracted through a hierarchical time memory structure and spatial attention mechanism. Combined with the road network topology constraints and historical trajectory similarity calculation, the driving trajectory data is deeply analyzed to generate destination prediction results and arrival time prediction results, including:

[0018] The controller provided in the electric bicycle transmits the charging demand prediction result output by the range prediction model to a pre-set charging scheduling server. The charging scheduling server parses the demand prediction result and obtains the position coordinate sequence and timestamp sequence of the electric bicycle. The server performs feature learning on the driving trajectory of the electric bicycle in combination with a hierarchical time memory structure. The hierarchical time memory structure includes multiple levels of neuron nodes. Each neuron node learns the local spatiotemporal features of the driving trajectory to obtain a hierarchical abstract representation corresponding to the driving trajectory.

[0019] Based on the hierarchical abstract representation, the local spatiotemporal features are weightedly aggregated in combination with the attention mechanism to obtain the context-related representation corresponding to the driving trajectory. The context-related representation is added to three parallel fully connected networks to predict the destination position, arrival time and remaining power of the electric bicycle after arrival. During the prediction process, the charging scheduling server obtains the road network topology information from the pre-acquired electronic map and maps and constrains the trajectory points. At the same time, the similarity between the current trajectory and the historical trajectory is calculated based on the Hamming distance, and the historical trajectory with the highest similarity is selected as the candidate trajectory. Each candidate trajectory is weighted and fused into the destination position to obtain the destination prediction result and arrival time prediction result.

[0020] In an optional embodiment,

[0021] A multi-constraint path planning model is constructed based on the current location coordinates and the destination prediction result. An improved Dijkstra algorithm is used to introduce an adaptive penalty factor to dynamically avoid congested sections. The path weight is updated online in combination with real-time traffic information to search for candidate paths. Charging piles on the candidate paths are optimized using an improved multi-objective genetic algorithm. Charging power matching, path deviation, and waiting time are used as optimization objectives. An adaptive crossover operator and a mutation operator are combined to perform a multi-objective solution. A Pareto solution set is generated and each solution in the Pareto solution set is scored. The charging pile with the highest score is selected as the target charging pile, which includes:

[0022] After obtaining the current location coordinates and predicted destination location of the electric bicycle, the charging scheduling server constructs a multi-constraint path planning model. The multi-constraint path planning model abstracts the road network into a weighted directed graph. The nodes in the graph represent intersections, the edges represent road sections, and the weights of the edges represent the cost of passing through the road section corresponding to the current edge. The cost is determined based on the length of the road section, the travel time, and the energy consumption. The optimization goal is to minimize the total path cost, and the constraints are set in combination with the mileage limit and the available information of the charging piles.

[0023] Solving the electric bicycle's travel path based on the multi-constraint path planning model and the improved Dijkstra algorithm, obtaining the congestion score of the road section in real time, classifying the traffic status of each road section according to a preset congestion threshold, and if the congestion score is higher than the congestion threshold, considering the current road section as a congested road section, increasing the weight corresponding to the congested road section according to the congestion score and performing dynamic avoidance, repeatedly updating the weight corresponding to each road section and solving the problem to obtain a candidate path;

[0024] Based on the candidate path, the charging piles on the candidate path are selected according to an improved genetic algorithm, and the charging piles on the candidate path are encoded as individuals through real number coding to generate an initial population. For each individual in the initial population, the charging power matching, path deviation and waiting time are used as the optimization objectives of the genetic algorithm, the corresponding fitness value is calculated and non-dominated sorting is performed to determine the non-dominated individuals. For each non-dominated individual, descending sorting is performed based on the fitness value, and the individuals in the top 10% are selected for binary tournament selection, adaptive crossover and mutation operations to generate a child population and repeat the iteration until the preset maximum number of iterations is reached. The non-dominated individual combination generated by the last iteration is selected to obtain a Pareto solution set. For each solution in the Pareto solution set, the corresponding score is calculated based on the optimization objective of the genetic algorithm, and the charging pile corresponding to the solution with the highest score is selected as the target charging pile.

[0025] In an optional embodiment,

[0026] The weight corresponding to each road segment is updated according to the improved Dijkstra algorithm as shown in the following formula:

[0027]

[0028] Among them, W n Represents the edge weight after update, W0 represents the edge weight before update, α represents the congestion penalty weight, I(SC>TH) represents the indicator function, which is used to indicate whether the congestion score of the current road section is higher than the preset congestion threshold. If it is higher, the indicator function takes the value of 1, otherwise it takes 0. SC represents the congestion score corresponding to the current road section, TH represents the preset congestion threshold, CP represents the congestion penalty factor, β represents the dynamic adjustment weight, and DT represents the dynamic adjustment factor.

[0029] In an optional embodiment,

[0030] The target charging pile adopts an identity authentication mechanism based on double encryption, and authenticates the electric bicycle through a dynamic key generation algorithm and an identity feature extraction algorithm, including:

[0031] The target charging pile determines the precise timestamp of receiving the electric bicycle's identity authentication request through an identity authentication mechanism based on double encryption and a pre-selected key generation algorithm, converts the precise timestamp into a text string in a fixed format and records it as a timestamp text string, generates a random number, converts the random number into a random number text string and concatenates it with the timestamp text string, combines a permutation operation and an obfuscation operation to generate an intermediate key and adds the intermediate key to a pre-set one-way mapping function to generate the dynamic key, sends the dynamic key to the electric bicycle through a secure channel, and after receiving the dynamic key, the electric bicycle encrypts the dynamic key through a key encryption algorithm to obtain an encrypted authentication key;

[0032] The electric bicycle sends the encrypted authentication key and identity feature information to the target charging pile, wherein the identity feature information includes the electric bicycle's unique identification code, battery type and rated voltage. After receiving the information sent by the electric bicycle, the target charging pile decrypts the authentication key through a key encryption algorithm and compares it with the dynamic key before sending. If they are consistent, the electric bicycle is considered to have passed the key verification, extracts the unique identification code from the identity feature information and matches it with a pre-set legal identity feature library. If the match is successful, the electric bicycle is considered to have passed the identity authentication and is allowed to access charging. Otherwise, the charging request is rejected.

[0033] In an optional embodiment,

[0034] During the charging process, an adaptive charging control algorithm based on a fuzzy neural network is used to learn the battery charging characteristics online. A charging condition evaluation model is established by combining temperature compensation and voltage compensation. A fuzzy inference engine is used to optimize the charging current in real time, and the charging curve is adaptively adjusted according to the battery health status. After charging is completed, the charging process data is uploaded to the charging scheduling server, including:

[0035] If the electric bicycle passes the identity authentication, the target charging pile starts charging. During the charging process, the target charging pile obtains voltage parameters, current parameters, and temperature parameters during the charging process through an adaptive charging control algorithm based on a fuzzy neural network and pre-set sensors, and uses these parameters as battery charging characteristics. The battery charging characteristics are added to a pre-set fuzzy neural network model, and the battery charging characteristics are mapped to a fuzzy set through a membership function. A decision is made through fuzzy inference rules to obtain a defuzzified control output.

[0036] Based on the defuzzified control output, combined with the temperature parameters and voltage parameters during charging, piecewise linear fitting and polynomial fitting are performed according to the theoretical model and experimental data of the current electric bicycle to generate voltage compensation rules and temperature compensation rules. Based on the voltage compensation rules and the temperature compensation rules, a voltage correction coefficient and a temperature correction coefficient are generated and sent to the fuzzy inference engine for voltage optimization and temperature optimization, and real-time optimization of the charging current is achieved according to the optimization results.

[0037] Obtaining battery usage time and charge and discharge records of the electric bicycle, evaluating the battery health status of the electric bicycle based on a Kalman filter algorithm to obtain a battery health, and adaptively adjusting the charging curve based on the battery health in combination with a control method corresponding to the charging curve. If the battery health is greater than a preset health threshold, increasing the constant current charging current and constant voltage charging voltage; otherwise, reducing the constant current charging current and constant voltage charging voltage until charging is completed;

[0038] After charging is completed, the charging start time, charging end time, charging quantity and average charging power are sent to the charging scheduling server as charging process data.

[0039] A second aspect of an embodiment of the present invention provides an electric bicycle charging system based on a mutual recognition mechanism, comprising:

[0040] The first unit is used to collect the current position coordinates and driving trajectory data of the electric bicycle, detect the battery status through a multi-sensor fusion algorithm, establish a battery health status assessment model, and construct a cruising range prediction model based on the battery health status assessment model using a deep reinforcement learning algorithm. The model uses historical riding data, ambient temperature, road condition information, and riding habits as input features, outputs remaining cruising range and charging demand prediction results, and sends the charging demand prediction results to the charging scheduling server. The model extracts trajectory features through a hierarchical time memory structure and a spatial attention mechanism, combines road network topology constraints and historical trajectory similarity calculation, and performs in-depth analysis on the driving trajectory data to generate destination prediction results and arrival time prediction results.

[0041] The second unit is used to build a multi-constraint path planning model based on the current location coordinates and the destination prediction result, adopt an improved Dijkstra algorithm to introduce an adaptive penalty factor to dynamically avoid congested sections, and update the path weight online in combination with real-time traffic information to search for candidate paths. The improved multi-objective genetic algorithm is used to optimize the charging piles on the candidate paths, with charging power matching, path deviation and waiting time as optimization objectives. The adaptive crossover operator and mutation operator are combined to perform multi-objective solution, generate a Pareto solution set, score each solution in the Pareto solution set, and select the charging pile with the highest score as the target charging pile;

[0042] The third unit is used for the target charging pile to adopt an identity authentication mechanism based on double encryption, and authenticates the electric bicycle through a dynamic key generation algorithm and an identity feature extraction algorithm. During the charging process, an adaptive charging control algorithm based on a fuzzy neural network is used to online learn the battery charging characteristics, and a charging condition evaluation model is established in combination with temperature compensation and voltage compensation. A fuzzy inference engine is used to optimize the charging current in real time, and the charging curve is adaptively adjusted according to the battery health status. After charging is completed, the charging process data is uploaded to the charging scheduling server.

[0043] According to a third aspect of the embodiments of the present invention,

[0044] An electronic device is provided, comprising:

[0045] processor;

[0046] a memory for storing processor-executable instructions;

[0047] The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0048] According to a fourth aspect of the embodiments of the present invention,

[0049] A computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the method described above is implemented.

[0050] In the present invention, through multi-sensor fusion and deep reinforcement learning algorithm, the battery status and cruising range are accurately predicted, and combined with destination prediction and multi-constraint path planning, the optimal charging pile is recommended to the user, which not only improves the charging efficiency, but also greatly improves the user experience, making the charging process more convenient and intelligent. The identity authentication mechanism based on double encryption is adopted to effectively prevent illegal charging and information leakage. The adaptive charging control algorithm based on fuzzy neural network can dynamically adjust the charging strategy according to the battery characteristics and health status, improve charging safety and battery life, and upload the charging demand prediction results and charging process data to the charging scheduling server to realize the intelligent scheduling and optimal allocation of charging resources. In summary, the present invention not only improves the operating efficiency of the entire charging network, but also provides strong support for the planning and management of electric bicycle charging infrastructure in the future. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 This is a flow chart of an electric bicycle charging method based on a mutual recognition mechanism according to an embodiment of the present invention;

[0052] Figure 2This is a structural diagram of an electric bicycle charging system based on a mutual recognition mechanism according to an embodiment of the present invention. DETAILED DESCRIPTION

[0053] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.

[0054] The following specific embodiments are used to describe the technical solution of the present invention in detail. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described in detail in some embodiments.

[0055] Figure 1 This is a flow chart of an electric bicycle charging method based on a mutual recognition mechanism according to an embodiment of the present invention. Figure 1 As shown, the method includes:

[0056] S1. Collect the current location coordinates and driving trajectory data of the electric bicycle, detect the battery status through a multi-sensor fusion algorithm, establish a battery health assessment model, and construct a range prediction model based on the battery health assessment model using a deep reinforcement learning algorithm. The model uses historical riding data, ambient temperature, road conditions, and riding habits as input features, outputs remaining range and charging demand prediction results, and sends the charging demand prediction results to the charging scheduling server. The model extracts trajectory features through a hierarchical temporal memory structure and spatial attention mechanism, combines road network topology constraints and historical trajectory similarity calculation, and conducts in-depth analysis of the driving trajectory data to generate destination prediction results and arrival time prediction results.

[0057] The hierarchical time memory structure is an architecture for capturing long-term and short-term dependencies in time series data, and is capable of storing and processing time information at different levels. The road network topology constraint refers to considering the geometric relationships between road connections, nodes, and paths when analyzing the traffic network to ensure that the model conforms to the actual traffic flow patterns.

[0058] In an optional embodiment,

[0059] The current location coordinates and driving trajectory data of the electric bicycle are collected, and the battery status is detected through a multi-sensor fusion algorithm. A battery health status assessment model is established. Based on the battery health status assessment model, a deep reinforcement learning algorithm is used to build a range prediction model. The historical riding data, ambient temperature, road condition information, and riding habits are used as input features. The output of the remaining range and charging demand prediction results include:

[0060] The current position coordinates of the electric bicycle are collected according to a positioning module pre-installed in the electric bicycle, and the acceleration and angular velocity are collected as riding status data according to a pre-installed inertial measurement unit. A controller in the electric bicycle reads the data uploaded by the positioning module and the inertial measurement unit through a bus and generates corresponding driving trajectory data of the electric bicycle in combination with an electronic map;

[0061] The battery voltage, current, and temperature are collected separately according to a pre-set sensor array, wherein the sensor array includes a voltage sensor, a current sensor, and a temperature sensor. The collected signals are filtered, normalized, and feature extracted using a multi-sensor fusion algorithm to obtain a multidimensional feature vector representing the battery state. The multidimensional feature vector is input into a pre-built battery health status assessment model, wherein the battery health status assessment model is trained using a support vector machine algorithm, and the health status corresponding to the multidimensional feature vector is divided into different levels, and a corresponding battery discharge rate is assigned to each level;

[0062] Acquire historical riding data of the electric bicycle, including riding time, riding mileage, average speed and average power; and simultaneously acquire ambient temperature data collected by an ambient temperature sensor, road condition data collected by a road condition sensor, and user riding habit data, wherein the road condition data includes road type, road slope, and road curvature; and the user riding habit data includes riding time period, riding frequency, and pedaling force. Use the historical riding data, ambient temperature, road condition information, and riding habits as input features, and the battery discharge rate, remaining cruising range, and charging demand as output targets to construct a cruising range prediction model. Train the cruising range prediction model according to a deep Q learning algorithm, use a convolutional neural network as an approximate function of the Q function, perform environmental interaction, learn a mileage prediction strategy, and obtain remaining cruising range and charging demand prediction results.

[0063] The support vector machine algorithm is a supervised learning model that achieves data classification by finding the optimal hyperplane and is suitable for high-dimensional data. The deep Q learning algorithm is a method that combines deep learning and reinforcement learning. It uses a deep neural network to approximate the Q function to optimize the decision-making process. The Q function represents the expected reward of taking an action in a specific state and is a key concept in reinforcement learning.

[0064] A positioning module and an inertial measurement unit are installed on the electric bicycle. The positioning module uses the GPS or Beidou satellite positioning system to obtain the real-time latitude and longitude coordinates of the electric bicycle. The inertial measurement unit includes a three-axis accelerometer and a three-axis gyroscope for measuring the acceleration and angular velocity of the electric bicycle. The controller is connected to the positioning module and the inertial measurement unit through the CAN bus to read the position and motion status data in real time.

[0065] The controller matches the collected position coordinates with a pre-stored electronic map to generate the driving trajectory of the electric bicycle. Specifically, the controller samples the position coordinates every 1 second and connects consecutive coordinate points to form a trajectory line. At the same time, the controller calculates the speed and direction of the electric bicycle based on the acceleration and angular velocity data to supplement the trajectory information when the GPS signal is weak or interrupted. For example, if the acceleration measured at a certain moment is 2m / s^2 and the duration is 5 seconds, the speed change can be calculated to be 10m / s.

[0066] The battery management system uses a sensor array, including voltage, current, and temperature sensors. The voltage sensor uses a high-precision ADC chip with a sampling rate of 1kHz and a measurement range of 0-100V. The current sensor uses a Hall effect sensor with a sampling rate of 1kHz and a measurement range of -100A to 100A. The temperature sensor uses an NTC thermistor with a sampling rate of 1Hz and a measurement range of -20°C to 80°C.

[0067] The collected raw signals are processed through the Kalman filter algorithm to remove noise, and then normalized. Statistical features such as mean, variance, peak value, etc. are extracted for the voltage, current and temperature data respectively to form a 30-dimensional feature vector. The feature vector is input into the pre-trained support vector machine model to divide the battery health status into five levels: excellent, good, fair, poor and very poor, with corresponding health measurement values ​​of 1.0, 0.8, 0.6, 0.4 and 0.2 respectively.

[0068] The last 30 days of historical riding data are extracted from the e-bike's controller, including daily riding time, mileage, average speed, and average power. Ambient temperature data is collected by a temperature sensor mounted on the outside of the vehicle, with a sampling interval of 10 minutes. Road condition data is obtained from pre-stored map information, including road type (flat, uphill, downhill), slope angle, and road curvature radius. User riding habits data is obtained by analyzing historical riding records, such as common riding time periods, weekly riding frequency, and average pedaling power.

[0069] Taking the above data as input features and the battery discharge rate, remaining range and charging requirement as output targets, a deep reinforcement learning model was constructed, using a convolutional neural network as an approximation of the Q function. The input layer contains 64 neurons, the hidden layer contains 3 convolutional layers and 2 fully connected layers, and the output layer contains 3 neurons corresponding to the remaining range, charging requirement and reward value respectively.

[0070] The model training process is as follows: the Q network parameters are initialized. Then, in each training round, an action is selected based on the current state (predicting range and charging requirements), and the system interacts with the environment to obtain rewards and the next state. The transition samples (state, action, reward, and next state) are stored in an experience replay pool. A batch of samples is randomly sampled from the experience replay pool. The mean squared error between the target Q value and the current Q value is calculated as the loss function. The network parameters are updated using stochastic gradient descent, and the iterations are repeated until convergence.

[0071] By inputting features such as the current battery status, environmental conditions, and user habits, the model can output a predicted remaining range and charging requirements. For example, if the battery health is 0.8, the ambient temperature is 20°C, the road is flat, and the user has a moderate riding habit, the model may predict a remaining range of 45km and recommend finding a charging station within 10km.

[0072] In this embodiment, through multi-sensor fusion and deep learning algorithms, accurate prediction of the range of electric bicycles is achieved, which improves the user's riding experience and battery utilization efficiency. Multi-dimensional information such as battery discharge rate, environmental factors and user habits are taken into consideration, making the prediction results more personalized and accurate, and effectively avoiding the range anxiety problem. In summary, this embodiment realizes the accurate perception of the operating status and battery health status of electric bicycles, and through multi-sensor fusion and deep reinforcement learning, an accurate and reliable range prediction system is established, providing users with scientific charging decision support.

[0073] In an optional embodiment,

[0074] The charging demand prediction result is sent to the charging scheduling server. Trajectory features are extracted through a hierarchical time memory structure and spatial attention mechanism. Combined with the road network topology constraints and historical trajectory similarity calculation, the driving trajectory data is deeply analyzed to generate destination prediction results and arrival time prediction results, including:

[0075] The controller provided in the electric bicycle transmits the charging demand prediction result output by the range prediction model to a pre-set charging scheduling server. The charging scheduling server parses the demand prediction result and obtains the position coordinate sequence and timestamp sequence of the electric bicycle. The server performs feature learning on the driving trajectory of the electric bicycle in combination with a hierarchical time memory structure. The hierarchical time memory structure includes multiple levels of neuron nodes. Each neuron node learns the local spatiotemporal features of the driving trajectory to obtain a hierarchical abstract representation corresponding to the driving trajectory.

[0076] Based on the hierarchical abstract representation, the local spatiotemporal features are weightedly aggregated in combination with the attention mechanism to obtain the context-related representation corresponding to the driving trajectory. The context-related representation is added to three parallel fully connected networks to predict the destination position, arrival time and remaining power of the electric bicycle after arrival. During the prediction process, the charging scheduling server obtains the road network topology information from the pre-acquired electronic map and maps and constrains the trajectory points. At the same time, the similarity between the current trajectory and the historical trajectory is calculated based on the Hamming distance, and the historical trajectory with the highest similarity is selected as the candidate trajectory. Each candidate trajectory is weighted and fused into the destination position to obtain the destination prediction result and arrival time prediction result.

[0077] The neuron node is the basic computing unit in the neural network, responsible for receiving input, performing weighted calculations and activation, and thus generating output. The local spatiotemporal features refer to features extracted in both time and space dimensions, which can capture dynamic changes and local relationships. The hierarchical abstract representation is to convert data into higher-level conceptual representations through feature extraction at different levels to facilitate the processing of complex tasks. The Hamming distance is an indicator used to measure the difference between two strings or binary numbers, indicating how many bits need to be different to transform one string into another.

[0078] A controller with GPS positioning function is installed on the electric bicycle to collect the real-time location information and power information of the electric bicycle. The controller is pre-loaded with a range prediction model. The model is trained based on historical driving data and can predict the remaining mileage based on factors such as current power, weather, and road conditions.

[0079] When the controller detects that the battery level of the electric bicycle is lower than a preset threshold (such as 20%), it will trigger a charging demand prediction. The controller will call the cruising range prediction model, input the current battery level, weather, road conditions and other information, and obtain the predicted remaining mileage. For example, if the current battery level is 15%, the weather is clear, and the road is flat, the predicted remaining mileage is 5 kilometers.

[0080] The controller sends the prediction results together with the unique identification ID of the electric bicycle via the wireless network to a pre-set charging scheduling server. After receiving the data, the charging scheduling server parses the electric bicycle ID and the predicted remaining mileage, and queries the GPS trajectory data of the vehicle in the last hour based on the ID to obtain the position coordinate sequence and the corresponding timestamp sequence.

[0081] The charging scheduling server uses a hierarchical time memory structure to learn features of trajectory data. This structure consists of three layers, each of which contains multiple neuron nodes. The first layer of neurons learns the spatiotemporal features between two adjacent trajectory points, such as speed, acceleration, and direction changes. The second layer of neurons learns the local road section features composed of multiple consecutive trajectory points, such as turning, acceleration, and deceleration behavior patterns. The third layer of neurons learns trajectory features with longer time spans, such as recurring path patterns. Through layer-by-layer abstraction, a hierarchical representation of the trajectory is finally obtained.

[0082] The charging scheduling server uses the spatial attention mechanism to perform weighted aggregation of features at different levels. Specifically, the similarity between the current trajectory point and all historical trajectory points is calculated as the attention weight. The higher the similarity, the greater the feature weight corresponding to the historical trajectory point. Through weighted summation, the feature representation related to the context of the current trajectory point is obtained.

[0083] The context-related feature representation is input into three parallel fully connected neural networks to predict the destination location, arrival time, and remaining battery life after arrival. During the prediction process, the charging scheduling server also considers the following constraints:

[0084] Road network topology constraints: Extract road network information from pre-acquired electronic maps and map trajectory points to the nearest roads to ensure that the prediction results are consistent with the actual road conditions.

[0085] Historical trajectory similarity: Calculate the similarity between the current trajectory and the historical trajectory. Use a similarity calculation method based on Hamming distance to discretize the trajectory into a grid sequence and calculate the degree of overlap between the two sequences. Select the top three historical trajectories with the highest similarity as candidate trajectories.

[0086] Candidate trajectory fusion: The destinations of the three candidate trajectories are fused by weighted averaging to obtain the final destination prediction result, with the weight proportional to the similarity.

[0087] The charging scheduling server outputs the destination prediction result (latitude and longitude coordinates), estimated arrival time and remaining power after arrival. For example, the predicted destination is (116.3, 39.9), and it is expected to arrive in 20 minutes with a remaining power of 5% upon arrival. Based on the prediction results and the distribution of charging piles around the destination, the charging scheduling server recommends the optimal charging route and charging station for the user.

[0088] In this embodiment, through the hierarchical time memory structure and spatial attention mechanism, the multi-scale spatiotemporal features of the electric bicycle trajectory are effectively extracted, which improves the accuracy of destination prediction. Combined with the road network topology constraints and historical trajectory similarity, the prediction results are more in line with the actual road conditions and user habits, thereby improving the reliability of the prediction. The end-to-end prediction method based on deep learning does not require manual feature design, has strong generalization ability and adaptability, and can be applied to different types of electric bicycles and complex urban environments. In summary, this embodiment realizes the precise perception of the operating status and battery health status of electric bicycles. Through multi-sensor fusion and deep reinforcement learning, an accurate and reliable cruising range prediction system is established, providing users with scientific charging decision support.

[0089] S2. Construct a multi-constrained path planning model based on the current location coordinates and the predicted destination. Use an improved Dijkstra algorithm to introduce an adaptive penalty factor to dynamically avoid congested sections. Path weights are updated online based on real-time traffic information to search for candidate paths. Then, use an improved multi-objective genetic algorithm to optimize charging stations along these candidate paths. Using charging power matching, path deviation, and waiting time as optimization objectives, an adaptive crossover operator and mutation operator are combined to perform a multi-objective solution. A Pareto solution set is generated, and each solution in the Pareto solution set is scored. The charging station with the highest score is selected as the target charging station.

[0090] The multi-constrained path planning model is used to find the optimal path to meet specific requirements while considering multiple constraints (such as time, cost, capacity, etc.). The improved Dijkstra algorithm is an optimization based on the classic Dijkstra algorithm, and usually introduces additional strategies to improve path search efficiency and adaptability. The adaptive penalty factor is a parameter dynamically adjusted according to the current state, which is used to penalize paths that violate constraints, thereby guiding the search for more compliant paths. The path deviation measures the difference between the actual path and the expected path and is used to evaluate the rationality of path selection.

[0091] In an optional embodiment,

[0092] A multi-constraint path planning model is constructed based on the current location coordinates and the destination prediction result. An improved Dijkstra algorithm is used to introduce an adaptive penalty factor to dynamically avoid congested sections. The path weight is updated online in combination with real-time traffic information to search for candidate paths. Charging piles on the candidate paths are optimized using an improved multi-objective genetic algorithm. Charging power matching, path deviation, and waiting time are used as optimization objectives. An adaptive crossover operator and a mutation operator are combined to perform a multi-objective solution. A Pareto solution set is generated and each solution in the Pareto solution set is scored. The charging pile with the highest score is selected as the target charging pile, which includes:

[0093] After obtaining the current location coordinates and predicted destination location of the electric bicycle, the charging scheduling server constructs a multi-constraint path planning model. The multi-constraint path planning model abstracts the road network into a weighted directed graph. The nodes in the graph represent intersections, the edges represent road sections, and the weights of the edges represent the cost of passing through the road section corresponding to the current edge. The cost is determined based on the length of the road section, the travel time, and the energy consumption. The optimization goal is to minimize the total path cost, and the constraints are set in combination with the mileage limit and the available information of the charging piles.

[0094] Solving the electric bicycle's travel path based on the multi-constraint path planning model and the improved Dijkstra algorithm, obtaining the congestion score of the road section in real time, classifying the traffic status of each road section according to a preset congestion threshold, and if the congestion score is higher than the congestion threshold, considering the current road section as a congested road section, increasing the weight corresponding to the congested road section according to the congestion score and performing dynamic avoidance, repeatedly updating the weight corresponding to each road section and solving the problem to obtain a candidate path;

[0095] Based on the candidate path, the charging piles on the candidate path are selected according to an improved genetic algorithm, and the charging piles on the candidate path are encoded as individuals through real number coding to generate an initial population. For each individual in the initial population, the charging power matching, path deviation and waiting time are used as the optimization objectives of the genetic algorithm, the corresponding fitness value is calculated and non-dominated sorting is performed to determine the non-dominated individuals. For each non-dominated individual, descending sorting is performed based on the fitness value, and the individuals in the top 10% are selected for binary tournament selection, adaptive crossover and mutation operations to generate a child population and repeat the iteration until the preset maximum number of iterations is reached. The non-dominated individual combination generated by the last iteration is selected to obtain a Pareto solution set. For each solution in the Pareto solution set, the corresponding score is calculated based on the optimization objective of the genetic algorithm, and the charging pile corresponding to the solution with the highest score is selected as the target charging pile.

[0096] The weighted directed graph represents the structure of nodes and edges in a network. The weight of an edge generally represents the cost or distance of passing through the edge. The passing cost refers to the expense or resource consumption that needs to be considered in path planning. The congestion score reflects the congestion of the path and helps to optimize path selection. The binary tournament selection is a selection mechanism that randomly selects two individuals for comparison and selects the individual with higher fitness to improve the overall quality of the population.

[0097] The road network is abstracted into a weighted directed graph, where nodes represent intersections and edges represent road sections. The weight of each edge is calculated, which is the cost of passing through the road section. The cost is determined based on the length of the road section, travel time, and energy consumption. The optimization goal is to minimize the total path cost, and constraints are set based on mileage restrictions and the availability of charging piles.

[0098] Based on a multi-constrained path planning model and an improved Dijkstra algorithm, the solution obtains the congestion score of each road section in real time and classifies the traffic status of the road section according to a preset congestion threshold. If the congestion score of a road section is higher than the threshold, the road section is considered congested. The corresponding weight of the road section is increased according to the congestion score, and dynamic avoidance is performed. The weight of each road section is repeatedly updated and solved to obtain candidate paths.

[0099] Based on the candidate path, an improved genetic algorithm is used to select the charging piles on the path. The charging piles on the candidate path are encoded as individuals through real number coding to generate an initial population. For each individual in the initial population, the charging power matching, path deviation and waiting time are used as the genetic algorithm optimization objectives. The corresponding fitness value is calculated and non-dominated sorting is performed to determine the non-dominated individuals. For each non-dominated individual, descending sorting is performed based on the fitness value. The individuals in the top 10% are selected for binary tournament selection, adaptive crossover and mutation operations to generate a child population. The iteration is repeated until the preset maximum number of iterations is reached. The non-dominated individual combination generated by the last iteration is selected to obtain the Pareto solution set. For each solution in the Pareto solution set, the corresponding score is calculated based on the genetic algorithm optimization objective, and the charging pile corresponding to the solution with the highest score is selected as the target charging pile.

[0100] For example, suppose a shared electric bicycle needs to travel from location A to location B, but the battery is insufficient to support the entire journey. Based on the current location coordinates of the electric bicycle and the predicted destination location, the system constructs a weighted directed graph with 50 nodes and 80 edges as a multi-constraint path planning model. The weight of each edge is calculated based on the length, travel time, and energy consumption of the corresponding road segment. The constraints are set as a maximum range of 20 kilometers for the electric bicycle and an available charging station along the route.

[0101] Using the improved Dijkstra algorithm, we obtain three candidate paths, P1, P2, and P3. We then obtain a real-time congestion score for each road segment. If the score is higher than 0.6, the segment is considered congested. For example, if the congestion score for the third road segment on P1 is 0.8, its weight is increased by 50% and the solution is repeated, ultimately yielding three candidate paths after dynamic avoidance.

[0102] The charging piles on the candidate paths are encoded to generate an initial population of 50 individuals. The fitness values ​​of the individuals are calculated based on the charging power matching, path deviation, and waiting time. After non-dominated sorting, the top 10% of individuals are selected for genetic manipulation to generate a new offspring population. After 50 iterations, a Pareto solution set is obtained, which contains 8 non-dominated solutions. The score of each solution is calculated and it is found that the charging pile corresponding to the solution number 15 has the highest score, the shortest distance, the best power matching, and the shortest waiting time. Therefore, this charging pile is selected as the target charging pile.

[0103] The system plans an optimal route for the electric bicycle from location A to location B via the target charging station, and sends the navigation information to the electric bicycle.

[0104] In this embodiment, a multi-constraint path planning model is used to comprehensively consider factors such as road section length, travel time and energy consumption, and plan a driving path with the lowest total cost, so that electric bicycles can reach their destination by the optimal route, reduce power consumption, and improve cruising range. The dynamic update of congestion conditions and real-time adjustment of the path ensure that the driving path is always optimal. Even if sudden congestion occurs, it can be avoided in time to improve traffic efficiency. The genetic algorithm comprehensively considers factors such as the power matching degree of the charging pile, the distance from the vehicle location, and the vehicle waiting time, and selects the best charging pile from the candidate charging piles, so that electric bicycles can receive charging services in the shortest time, reduce charging waiting time, and optimize the user's charging experience. The reasonable allocation of charging resources can shorten the waiting time of each vehicle, while improving the utilization rate of charging stations and realizing efficient utilization of charging resources. In summary, this embodiment can plan the optimal driving path and the best charging plan for electric bicycles, comprehensively improve the user experience and operational efficiency of electric bicycles, and has significant technical effects and application value.

[0105] In an optional embodiment,

[0106] The weight corresponding to each road segment is updated according to the improved Dijkstra algorithm as shown in the following formula:

[0107]

[0108] Among them, W nRepresents the edge weight after update, W0 represents the edge weight before update, α represents the congestion penalty weight, I(SC>TH) represents the indicator function, which is used to indicate whether the congestion score of the current road section is higher than the preset congestion threshold. If it is higher, the indicator function takes the value of 1, otherwise it takes 0. SC represents the congestion score corresponding to the current road section, TH represents the preset congestion threshold, CP represents the congestion penalty factor, β represents the dynamic adjustment weight, and DT represents the dynamic adjustment factor.

[0109] In this embodiment, by introducing congestion penalties and dynamic adjustment mechanisms, the algorithm can better reflect real-time traffic conditions, avoid congested sections, provide users with better driving routes, effectively reduce driving time and fuel consumption, dynamically adjust parameters according to different time periods, weather conditions and other factors, adapt to complex and changeable traffic environments, improve the accuracy and practicality of path planning, and continuously optimize parameter settings through historical data analysis and prediction, improve algorithm performance, and provide strong support for the development of intelligent transportation systems. In summary, this embodiment realizes the accurate quantification and dynamic response of the congestion status of road sections, provides a more accurate weight reference for path planning, and improves the practicality of the planning results.

[0110] S3. The target charging pile adopts an identity authentication mechanism based on double encryption, and authenticates the electric bicycle through a dynamic key generation algorithm and an identity feature extraction algorithm. During the charging process, an adaptive charging control algorithm based on a fuzzy neural network is used to online learn the battery charging characteristics, and a charging condition evaluation model is established in combination with temperature compensation and voltage compensation. A fuzzy inference engine is used to optimize the charging current in real time, and the charging curve is adaptively adjusted according to the battery health status. After charging is completed, the charging process data is uploaded to the charging scheduling server.

[0111] The dual-encryption-based identity authentication mechanism improves the security of identity authentication by using two encryption algorithms simultaneously, ensuring the authenticity of user identity and the confidentiality of data. The dynamic key generation algorithm generates a new encryption key in each communication to enhance security and prevent the key from being reused or leaked. The identity feature extraction algorithm is used to extract key features from the user's biometrics or behavior to support the identity authentication process. The fuzzy neural network-based adaptive charging control algorithm combines fuzzy logic and neural network technology to dynamically adjust the charging strategy according to the battery status and environmental conditions, optimizing the battery's charging efficiency and life. The battery charging characteristics describe the behavior and response of the battery during the charging process, including factors such as charging rate, efficiency, and temperature.

[0112] In an optional embodiment,

[0113] The target charging pile adopts an identity authentication mechanism based on double encryption, and authenticates the electric bicycle through a dynamic key generation algorithm and an identity feature extraction algorithm, including:

[0114] The target charging pile determines the precise timestamp of receiving the electric bicycle's identity authentication request through an identity authentication mechanism based on double encryption and a pre-selected key generation algorithm, converts the precise timestamp into a text string in a fixed format and records it as a timestamp text string, generates a random number, converts the random number into a random number text string and concatenates it with the timestamp text string, combines a permutation operation and an obfuscation operation to generate an intermediate key and adds the intermediate key to a pre-set one-way mapping function to generate the dynamic key, sends the dynamic key to the electric bicycle through a secure channel, and after receiving the dynamic key, the electric bicycle encrypts the dynamic key through a key encryption algorithm to obtain an encrypted authentication key;

[0115] The electric bicycle sends the encrypted authentication key and identity feature information to the target charging pile, wherein the identity feature information includes the electric bicycle's unique identification code, battery type and rated voltage. After receiving the information sent by the electric bicycle, the target charging pile decrypts the authentication key through a key encryption algorithm and compares it with the dynamic key before sending. If they are consistent, the electric bicycle is considered to have passed the key verification, extracts the unique identification code from the identity feature information and matches it with a pre-set legal identity feature library. If the match is successful, the electric bicycle is considered to have passed the identity authentication and is allowed to access charging. Otherwise, the charging request is rejected.

[0116] The text string refers to a sequence of characters used to represent text information. The obfuscation operation transforms the data to make it difficult to identify in order to protect information security. The substitution operation adjusts the position of elements in the data to increase complexity. The one-way mapping function is a mathematical function whose input value can be calculated but the original input cannot be reversed. It is widely used in cryptography.

[0117] The target charging pile receives the authentication request sent by the electric bicycle. The charging pile uses a precise time synchronization mechanism to record the exact timestamp of the received request, accurate to the millisecond level. For example, the recorded timestamp is "2023-05-20, 14:30:25.789".

[0118] The charging pile converts the timestamp into a fixed-format text string, such as "20230520143025789", uses a secure random number generator to generate a 128-bit random number, such as "9a7b3c1d5e8f2g6h", converts the random number into a text string, and concatenates it with the timestamp text string to obtain "20230520143025789-9a7b3c1d5e8f2g6h".

[0119] The charging pile performs permutation and obfuscation operations on the concatenated character strings. The permutation operation rearranges the characters in the string according to predetermined rules, such as swapping the positions of every three characters. The obfuscation operation replaces the characters, such as replacing numbers with corresponding letters, to generate the intermediate key "2c5b0a3d1e4f7g9h-9a7B3C1D5E8F2G6H".

[0120] The charging pile substitutes the intermediate key into a preset one-way mapping function, such as the SHA-256 hash function, to generate a final 256-bit dynamic key, and sends the dynamic key to the electric bicycle through a secure encrypted channel. After receiving the dynamic key, the electric bicycle encrypts it using the AES-256 encryption algorithm to obtain the encrypted authentication key. The electric bicycle prepares identity feature information, including a unique 16-bit identification code "EV20230520001234", battery type "lithium-ion battery" and rated voltage "48V".

[0121] The electric bicycle packages the encrypted authentication key and identity feature information and sends it to the target charging pile through a secure channel. After receiving the information, the charging pile uses the same AES-256 algorithm to decrypt the authentication key and compare it with the previously generated dynamic key. If they are exactly the same, it means that the electric bicycle has passed the key verification.

[0122] The e-bike's unique identification code, "EV20230520001234," is extracted from the identity feature information. The charging station maintains a pre-set database of legal identification features, which contains the unique identification codes of all e-bikes allowed to charge. The charging station matches the extracted identification code against this database. If a match is found, the e-bike has passed identity authentication.

[0123] Only when the electric bicycle passes both key verification and identity authentication will the charging pile allow it to connect to the charging station. Otherwise, the charging pile will reject the charging request and send a charging refusal message to the electric bicycle.

[0124] In this embodiment, double encryption and dynamic key mechanisms are adopted, which greatly increases the difficulty of forgery or cracking, effectively preventing unauthorized devices from accessing charging, and avoiding the risk of static keys being easily stolen by dynamically generating keys. At the same time, the authentication mechanism can be updated frequently as needed. The authentication process is highly automated and does not require manual intervention. Identity authentication can be completed quickly, thereby improving the utilization efficiency of charging piles. In summary, this embodiment implements a double authentication mechanism based on dynamic keys and identity features, provides high-security charging access control through multiple encryption and identity matching, and effectively ensures the safety and reliability of the charging process.

[0125] In an optional embodiment,

[0126] During the charging process, an adaptive charging control algorithm based on a fuzzy neural network is used to learn the battery charging characteristics online. A charging condition evaluation model is established by combining temperature compensation and voltage compensation. A fuzzy inference engine is used to optimize the charging current in real time, and the charging curve is adaptively adjusted according to the battery health status. After charging is completed, the charging process data is uploaded to the charging scheduling server, including:

[0127] If the electric bicycle passes the identity authentication, the target charging pile starts charging. During the charging process, the target charging pile obtains voltage parameters, current parameters, and temperature parameters during the charging process through an adaptive charging control algorithm based on a fuzzy neural network and pre-set sensors, and uses these parameters as battery charging characteristics. The battery charging characteristics are added to a pre-set fuzzy neural network model, and the battery charging characteristics are mapped to a fuzzy set through a membership function. A decision is made through fuzzy inference rules to obtain a defuzzified control output.

[0128] Based on the defuzzified control output, combined with the temperature parameters and voltage parameters during charging, piecewise linear fitting and polynomial fitting are performed according to the theoretical model and experimental data of the current electric bicycle to generate voltage compensation rules and temperature compensation rules. Based on the voltage compensation rules and the temperature compensation rules, a voltage correction coefficient and a temperature correction coefficient are generated and sent to the fuzzy inference engine for voltage optimization and temperature optimization, and real-time optimization of the charging current is achieved according to the optimization results.

[0129] Obtaining battery usage time and charge and discharge records of the electric bicycle, evaluating the battery health status of the electric bicycle based on a Kalman filter algorithm to obtain a battery health, and adaptively adjusting the charging curve based on the battery health in combination with a control method corresponding to the charging curve. If the battery health is greater than a preset health threshold, increasing the constant current charging current and constant voltage charging voltage; otherwise, reducing the constant current charging current and constant voltage charging voltage until charging is completed;

[0130] After charging is completed, the charging start time, charging end time, charging quantity and average charging power are sent to the charging scheduling server as charging process data.

[0131] The defuzzified control output refers to the output result of the fuzzy control system, which is converted into clear control instructions for practical application. The membership function is used to define the degree of membership of elements in a fuzzy set, characterizing the degree to which they belong to a certain fuzzy concept. The piecewise linear fitting is a data fitting method that divides the data into multiple intervals and fits them with linear functions to improve the fitting accuracy. The polynomial fitting uses polynomial functions to model the data to capture the trends and characteristics of the data.

[0132] After passing the certification, the charging pile starts the adaptive charging control algorithm based on fuzzy neural network, using pre-installed voltage, current and temperature sensors to collect relevant parameters of the charging process in real time. These parameters are input into the preset fuzzy neural network model as battery charging characteristics.

[0133] The fuzzy neural network model uses membership functions to map collected parameters such as voltage, current, and temperature into fuzzy sets. For example, voltage can be mapped into three fuzzy sets: "low," "medium," and "high." The specific mapping interval can be set according to the battery specifications. Decision analysis is performed using preset fuzzy inference rules. The fuzzy inference rules can be set in the form of "If the voltage is low and the temperature is normal, then increase the charging current." After defuzzification, the specific control output value, such as the charging current adjustment amount, is obtained.

[0134] The charging pile combines the temperature and voltage parameters collected in real time with the theoretical model and historical experimental data of this model of electric bicycle, and generates voltage compensation rules and temperature compensation rules through piecewise linear fitting and polynomial fitting. For example, the voltage interval is divided into multiple sub-intervals, and linear fitting is used in each sub-interval. Quadratic polynomial fitting can be used for temperature effects. Based on the compensation rules, the system calculates the voltage correction coefficient and temperature correction coefficient and sends them to the fuzzy inference engine.

[0135] The fuzzy inference engine comprehensively considers the control output value, voltage correction factor and temperature correction factor to optimize and adjust the charging current in real time. For example, it appropriately increases the charging current when the battery voltage is low, and reduces the charging current when the temperature is high, so as to balance charging efficiency and safety.

[0136] The charging pile obtains the battery usage time and historical charge and discharge records of the electric bicycle, and uses the Kalman filter algorithm to combine these historical data with the current charging status to conduct a real-time evaluation of the battery health status, and obtain a battery health value between 0 and 100. The system adaptively adjusts the standard charging curve based on the health level. Specifically, if the health level is higher than the preset threshold (such as 80), the current value of the constant current charging stage (such as from 1C to 1.2C) and the voltage value of the constant voltage charging stage (such as from 4.2V to 4.25V) are appropriately increased. Conversely, if the health level is lower than the threshold, the charging parameters are reduced accordingly to extend the battery life.

[0137] The charging process continues until the termination condition is reached. After charging is completed, the charging pile will send the recorded charging start time, end time, charging capacity (such as 2kWh) and average charging power (such as 500W) data to the charging scheduling server for subsequent analysis and optimization.

[0138] In this embodiment, by real-time monitoring and adjustment of charging parameters, the charging speed can be maximized and the charging time can be shortened while ensuring safety. The charging strategy can be dynamically adjusted according to the battery health status, which can effectively slow down battery degradation and extend the service life. Uploading charging data to the server helps to analyze charging patterns and predict charging needs, thereby optimizing the resource allocation and scheduling of the entire charging network. In summary, this embodiment realizes safe and efficient adaptive charging management, which not only ensures charging performance but also extends battery life and provides comprehensive charging process optimization.

[0139] Figure 2 FIG. 1 is a structural diagram of an electric bicycle charging system based on a mutual recognition mechanism according to an embodiment of the present invention. Figure 2 As shown, the system includes:

[0140] The first unit is used to collect the current position coordinates and driving trajectory data of the electric bicycle, detect the battery status through a multi-sensor fusion algorithm, establish a battery health status assessment model, and construct a cruising range prediction model based on the battery health status assessment model using a deep reinforcement learning algorithm. The model uses historical riding data, ambient temperature, road condition information, and riding habits as input features, outputs remaining cruising range and charging demand prediction results, and sends the charging demand prediction results to the charging scheduling server. The model extracts trajectory features through a hierarchical time memory structure and a spatial attention mechanism, combines road network topology constraints and historical trajectory similarity calculation, and performs in-depth analysis on the driving trajectory data to generate destination prediction results and arrival time prediction results.

[0141] The second unit is used to build a multi-constraint path planning model based on the current location coordinates and the destination prediction result, adopt an improved Dijkstra algorithm to introduce an adaptive penalty factor to dynamically avoid congested sections, and update the path weight online in combination with real-time traffic information to search for candidate paths. The improved multi-objective genetic algorithm is used to optimize the charging piles on the candidate paths, with charging power matching, path deviation and waiting time as optimization objectives. The adaptive crossover operator and mutation operator are combined to perform multi-objective solution, generate a Pareto solution set, score each solution in the Pareto solution set, and select the charging pile with the highest score as the target charging pile;

[0142] The third unit is used for the target charging pile to adopt an identity authentication mechanism based on double encryption, and authenticates the electric bicycle through a dynamic key generation algorithm and an identity feature extraction algorithm. During the charging process, an adaptive charging control algorithm based on a fuzzy neural network is used to online learn the battery charging characteristics, and a charging condition evaluation model is established in combination with temperature compensation and voltage compensation. A fuzzy inference engine is used to optimize the charging current in real time, and the charging curve is adaptively adjusted according to the battery health status. After charging is completed, the charging process data is uploaded to the charging scheduling server.

[0143] According to a third aspect of the embodiments of the present invention,

[0144] An electronic device is provided, comprising:

[0145] processor;

[0146] a memory for storing processor-executable instructions;

[0147] The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0148] According to a fourth aspect of the embodiments of the present invention,

[0149] A computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the method described above is implemented.

[0150] The present invention may be a method, an apparatus, a system and / or a computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for executing various aspects of the present invention.

[0151] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the above embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. The electric bicycle charging method based on the mutual recognition mechanism is characterized in that: include: The current location coordinates and driving trajectory data of the electric bicycle are collected, the battery status is detected through a multi-sensor fusion algorithm, and a battery health status assessment model is established. Based on the battery health status assessment model, a deep reinforcement learning algorithm is used to build a cruising range prediction model. The historical riding data, ambient temperature, road condition information, and riding habits are used as input features, and the remaining cruising range and charging demand prediction results are output. The charging demand prediction results are sent to the charging scheduling server. The trajectory features are extracted through a hierarchical time memory structure and a spatial attention mechanism. In combination with the road network topology constraints and the historical trajectory similarity calculation, the driving trajectory data is deeply analyzed to generate the destination prediction result and the arrival time prediction result, including: The controller provided in the electric bicycle transmits the charging demand prediction result output by the range prediction model to a pre-set charging scheduling server. The charging scheduling server parses the demand prediction result and obtains the position coordinate sequence and timestamp sequence of the electric bicycle. The server performs feature learning on the driving trajectory of the electric bicycle in combination with a hierarchical time memory structure. The hierarchical time memory structure includes multiple levels of neuron nodes. Each neuron node learns the local spatiotemporal features of the driving trajectory to obtain a hierarchical abstract representation corresponding to the driving trajectory. Based on the hierarchical abstract representation, the local spatiotemporal features are weightedly aggregated in combination with the attention mechanism to obtain a context-related representation corresponding to the driving trajectory. The context-related representation is added to three parallel fully connected networks to respectively predict the destination location, arrival time, and remaining battery power of the electric bicycle. During the prediction process, the charging scheduling server obtains road network topology information from a pre-acquired electronic map and maps and constrains the trajectory points. At the same time, the similarity between the current trajectory and the historical trajectory is calculated based on the Hamming distance. The historical trajectory with the highest similarity is selected as the candidate trajectory. Each candidate trajectory is weighted and fused into the destination location to obtain the destination prediction result and arrival time prediction result. A multi-constrained path planning model is constructed based on the current location coordinates and the destination prediction result. An improved Dijkstra algorithm is used to introduce an adaptive penalty factor to dynamically avoid congested sections. Path weights are updated online in combination with real-time traffic information to search for candidate paths. Charging piles on the candidate paths are optimized using an improved multi-objective genetic algorithm. Charging power matching, path deviation, and waiting time are used as optimization objectives. A multi-objective solution is performed using an adaptive crossover operator and a mutation operator. A Pareto solution set is generated and each solution in the Pareto solution set is scored. The charging pile with the highest score is selected as the target charging pile. The target charging pile adopts an identity authentication mechanism based on double encryption, and authenticates the electric bicycle through a dynamic key generation algorithm and an identity feature extraction algorithm. During the charging process, an adaptive charging control algorithm based on a fuzzy neural network is used to online learn the battery charging characteristics, and a charging condition evaluation model is established in combination with temperature compensation and voltage compensation. A fuzzy inference engine is used to optimize the charging current in real time, and the charging curve is adaptively adjusted according to the battery health status. After charging is completed, the charging process data is uploaded to the charging scheduling server.

2. The method according to claim 1, characterized in that The current location coordinates and driving trajectory data of the electric bicycle are collected, and the battery status is detected through a multi-sensor fusion algorithm. A battery health status assessment model is established. Based on the battery health status assessment model, a deep reinforcement learning algorithm is used to build a range prediction model. The historical riding data, ambient temperature, road condition information, and riding habits are used as input features. The output of the remaining range and charging demand prediction results include: The current position coordinates of the electric bicycle are collected according to a positioning module pre-installed in the electric bicycle, and the acceleration and angular velocity are collected as riding status data according to a pre-installed inertial measurement unit. A controller in the electric bicycle reads the data uploaded by the positioning module and the inertial measurement unit through a bus and generates corresponding driving trajectory data of the electric bicycle in combination with an electronic map; The battery voltage, current, and temperature are collected separately according to a pre-set sensor array, wherein the sensor array includes a voltage sensor, a current sensor, and a temperature sensor. The collected signals are filtered, normalized, and feature extracted using a multi-sensor fusion algorithm to obtain a multidimensional feature vector representing the battery state. The multidimensional feature vector is input into a pre-built battery health status assessment model, wherein the battery health status assessment model is trained using a support vector machine algorithm, and the health status corresponding to the multidimensional feature vector is divided into different levels, and a corresponding battery discharge rate is assigned to each level; Acquire historical riding data of the electric bicycle, including riding time, riding mileage, average speed and average power; and simultaneously acquire ambient temperature data collected by an ambient temperature sensor, road condition data collected by a road condition sensor, and user riding habit data, wherein the road condition data includes road type, road slope, and road curvature; and the user riding habit data includes riding time period, riding frequency, and pedaling force. Use the historical riding data, ambient temperature, road condition information, and riding habits as input features, and the battery discharge rate, remaining cruising range, and charging demand as output targets to construct a cruising range prediction model. Train the cruising range prediction model according to a deep Q learning algorithm, use a convolutional neural network as an approximate function of the Q function, perform environmental interaction, learn a mileage prediction strategy, and obtain remaining cruising range and charging demand prediction results.

3. The method according to claim 1, characterized in that A multi-constraint path planning model is constructed based on the current location coordinates and the destination prediction result. An improved Dijkstra algorithm is used to introduce an adaptive penalty factor to dynamically avoid congested sections. The path weight is updated online in combination with real-time traffic information to search for candidate paths. Charging piles on the candidate paths are optimized using an improved multi-objective genetic algorithm. Charging power matching, path deviation, and waiting time are used as optimization objectives. An adaptive crossover operator and a mutation operator are combined to perform a multi-objective solution. A Pareto solution set is generated and each solution in the Pareto solution set is scored. The charging pile with the highest score is selected as the target charging pile, which includes: After obtaining the current location coordinates and predicted destination location of the electric bicycle, the charging scheduling server constructs a multi-constraint path planning model. The multi-constraint path planning model abstracts the road network into a weighted directed graph. The nodes in the graph represent intersections, the edges represent road sections, and the weights of the edges represent the cost of passing through the road section corresponding to the current edge. The cost is determined based on the length of the road section, the travel time, and the energy consumption. The optimization goal is to minimize the total path cost, and the constraints are set in combination with the mileage limit and the available information of the charging piles. Solving the electric bicycle's travel path based on the multi-constraint path planning model and the improved Dijkstra algorithm, obtaining the congestion score of the road section in real time, classifying the traffic status of each road section according to a preset congestion threshold, and if the congestion score is higher than the congestion threshold, considering the current road section as a congested road section, increasing the weight corresponding to the congested road section according to the congestion score and performing dynamic avoidance, repeatedly updating the weight corresponding to each road section and solving the problem to obtain a candidate path; Based on the candidate path, the charging piles on the candidate path are selected according to an improved genetic algorithm, and the charging piles on the candidate path are encoded as individuals through real number coding to generate an initial population. For each individual in the initial population, the charging power matching degree, path deviation degree and waiting time are used as the optimization objectives of the genetic algorithm, the corresponding fitness value is calculated and non-dominated sorting is performed to determine the non-dominated individuals. For each non-dominated individual, descending sorting is performed based on the fitness value, and the individuals in the top 10% are selected for binary tournament selection, adaptive crossover and mutation operations to generate a child population and repeat the iteration until the preset maximum number of iterations is reached. The non-dominated individual combination generated by the last iteration is selected to obtain a Pareto solution set. For each solution in the Pareto solution set, the corresponding score is calculated based on the optimization objective of the genetic algorithm, and the charging pile corresponding to the solution with the highest score is selected as the target charging pile.

4. The method according to claim 1, wherein The weight corresponding to each road segment is updated according to the improved Dijkstra algorithm as shown in the following formula: ； in, W n represents the updated edge weight, W 0 represents the edge weight before update, α represents the congestion penalty weight, I ( SC > TH ) represents an indicator function, which is used to indicate whether the current road congestion score is higher than the preset congestion threshold. If it is higher, the indicator function takes the value of 1, otherwise it takes the value of 0. SC Indicates the congestion level score corresponding to the current road section. TH Indicates the preset congestion threshold. CP represents the congestion penalty factor, β Indicates dynamic adjustment of weights, DT Represents the dynamic adjustment factor.

5. The method according to claim 1, wherein The target charging pile adopts an identity authentication mechanism based on double encryption, and authenticates the electric bicycle through a dynamic key generation algorithm and an identity feature extraction algorithm, including: The target charging pile determines the precise timestamp of receiving the electric bicycle's identity authentication request through an identity authentication mechanism based on double encryption and a pre-selected key generation algorithm, converts the precise timestamp into a text string in a fixed format and records it as a timestamp text string, generates a random number, converts the random number into a random number text string and concatenates it with the timestamp text string, combines a permutation operation and an obfuscation operation to generate an intermediate key and adds the intermediate key to a pre-set one-way mapping function to generate the dynamic key, sends the dynamic key to the electric bicycle through a secure channel, and after receiving the dynamic key, the electric bicycle encrypts the dynamic key through a key encryption algorithm to obtain an encrypted authentication key; The electric bicycle sends the encrypted authentication key and identity feature information to the target charging pile, wherein the identity feature information includes the electric bicycle's unique identification code, battery type and rated voltage. After receiving the information sent by the electric bicycle, the target charging pile decrypts the authentication key through a key encryption algorithm and compares it with the dynamic key before sending. If they are consistent, the electric bicycle is considered to have passed the key verification, extracts the unique identification code from the identity feature information and matches it with a pre-set legal identity feature library. If the match is successful, the electric bicycle is considered to have passed the identity authentication and is allowed to access charging. Otherwise, the charging request is rejected.

6. The method according to claim 1, characterized in that During the charging process, an adaptive charging control algorithm based on a fuzzy neural network is used to learn the battery charging characteristics online. A charging condition evaluation model is established by combining temperature compensation and voltage compensation. A fuzzy inference engine is used to optimize the charging current in real time, and the charging curve is adaptively adjusted according to the battery health status. After charging is completed, the charging process data is uploaded to the charging scheduling server, including: If the electric bicycle passes the identity authentication, the target charging pile starts charging. During the charging process, the target charging pile obtains voltage parameters, current parameters, and temperature parameters during the charging process through an adaptive charging control algorithm based on a fuzzy neural network and pre-set sensors, and uses these parameters as battery charging characteristics. The battery charging characteristics are added to a pre-set fuzzy neural network model, and the battery charging characteristics are mapped to a fuzzy set through a membership function. A decision is made through fuzzy inference rules to obtain a defuzzified control output. Based on the defuzzified control output, combined with the temperature parameters and voltage parameters during charging, piecewise linear fitting and polynomial fitting are performed according to the theoretical model and experimental data of the current electric bicycle to generate voltage compensation rules and temperature compensation rules. Based on the voltage compensation rules and the temperature compensation rules, a voltage correction coefficient and a temperature correction coefficient are generated and sent to the fuzzy inference engine for voltage optimization and temperature optimization, and real-time optimization of the charging current is achieved according to the optimization results. Obtaining battery usage time and charge and discharge records of the electric bicycle, evaluating the battery health status of the electric bicycle based on a Kalman filter algorithm to obtain a battery health, and adaptively adjusting the charging curve based on the battery health in combination with a control method corresponding to the charging curve. If the battery health is greater than a preset health threshold, increasing the constant current charging current and constant voltage charging voltage; otherwise, reducing the constant current charging current and constant voltage charging voltage until charging is completed; After charging is completed, the charging start time, charging end time, charging quantity and average charging power are sent to the charging scheduling server as charging process data.

7. An electric bicycle charging system based on a mutual recognition mechanism, used to implement the method according to any one of claims 1 to 6, characterized in that: include: The first unit is used to collect the current position coordinates and driving trajectory data of the electric bicycle, detect the battery status through a multi-sensor fusion algorithm, establish a battery health status assessment model, and construct a cruising range prediction model based on the battery health status assessment model using a deep reinforcement learning algorithm. The model uses historical riding data, ambient temperature, road condition information, and riding habits as input features, outputs remaining cruising range and charging demand prediction results, and sends the charging demand prediction results to the charging scheduling server. The model extracts trajectory features through a hierarchical time memory structure and a spatial attention mechanism, combines road network topology constraints and historical trajectory similarity calculation, and performs in-depth analysis on the driving trajectory data to generate destination prediction results and arrival time prediction results. The second unit is used to build a multi-constraint path planning model based on the current location coordinates and the destination prediction result, adopt an improved Dijkstra algorithm to introduce an adaptive penalty factor to dynamically avoid congested sections, and update the path weight online in combination with real-time traffic information to search for candidate paths. The improved multi-objective genetic algorithm is used to optimize the charging piles on the candidate paths, with charging power matching, path deviation and waiting time as optimization objectives. The adaptive crossover operator and mutation operator are combined to perform multi-objective solution, generate a Pareto solution set, score each solution in the Pareto solution set, and select the charging pile with the highest score as the target charging pile; The third unit is used for the target charging pile to adopt an identity authentication mechanism based on double encryption, and authenticates the electric bicycle through a dynamic key generation algorithm and an identity feature extraction algorithm. During the charging process, an adaptive charging control algorithm based on a fuzzy neural network is used to online learn the battery charging characteristics, and a charging condition evaluation model is established in combination with temperature compensation and voltage compensation. A fuzzy inference engine is used to optimize the charging current in real time, and the charging curve is adaptively adjusted according to the battery health status. After charging is completed, the charging process data is uploaded to the charging scheduling server.

8. An electronic device, characterized in that: include: processor; a memory for storing processor-executable instructions; The processor is configured to call the instructions stored in the memory to execute the method according to any one of claims 1 to 6.

9. A computer-readable storage medium having computer program instructions stored thereon, characterized in that: When the computer program instructions are executed by a processor, the method according to any one of claims 1 to 6 is implemented.

Images