Multi-mode vehicle charging and parking cooperative management method based on Internet of Things

Abstract

The invention provides a multi-mode vehicle charging and parking cooperative management method based on the Internet of Things, and relates to the technical field of intelligent parking. The application steps of the main control system are multi-modal data acquisition and dynamic parking and charging constraint initialization, multi-scale parallel parking and charging model construction and preliminary distribution, multi-scale error feedback and cross-node collaborative evaluation, and adaptive constraint adjustment and asynchronous distributed iterative optimization. Global parking and charging scheme integration, expandability verification, output optimization scheme and industrial application packaging are carried out, multi-mode data such as vehicle electric quantity, position and parking lot state are collected in real time through the Internet of Things, changes of charging and parking demands can be found in time, the problems of congestion and waiting caused by resource allocation lag are effectively avoided, and the service life of a vehicle is prolonged. The real-time performance and efficiency of management are improved, and dynamic parking and charging constraints are generated by adopting a deep reinforcement learning algorithm.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent parking, and specifically to a multi-mode vehicle charging and parking collaborative management method based on the Internet of Things. Background Art

[0002] With the popularization of electric vehicles and the acceleration of urbanization, vehicle charging and parking management has become an important challenge in modern transportation and energy systems. The rapid growth of electric vehicles has increased the demand for charging infrastructure and parking resources. Traditional charging and parking management methods are often inefficient and wasteful of resources, making it difficult to meet the large-scale demand during peak periods. In addition, problems such as uneven grid load, charging safety hazards, and insufficient user experience further exacerbate the management difficulty. Therefore, how to achieve efficient, safe, and personalized vehicle charging and parking collaborative management has become a research hotspot in the fields of intelligent transportation and energy.

[0003] Currently, the implementation of vehicle charging and parking management mainly relies on the following technical means: Manual management: Traditional charging and parking management relies on manual scheduling, where on-site staff allocate charging piles and parking spaces.

[0004] Fixed scheduling system: A fixed scheduling system based on preset rules attempts to optimize resource usage by pre-assigning charging piles and parking spaces.

[0005] Single sensor monitoring: A single sensor (such as a current sensor or an ultrasonic sensor) is used to monitor the status of charging piles and the occupancy of parking spaces to assist in management decisions.

[0006] Intelligent navigation technology: Based on GPS navigation and mobile applications, it provides users with information on the locations of parking spaces and charging piles.

[0007] Although existing technologies have played a certain role in vehicle charging and parking management, there are still many deficiencies: Problem 1: Traditional manual management and fixed scheduling systems are time-consuming and have limited coverage, unable to achieve real-time monitoring of charging and parking demands. This leads to low resource allocation efficiency during peak periods, and potential congestion and safety hazards are difficult to detect in a timely manner, affecting the overall operation.

[0008] Problem 2: Using technologies such as single sensors or intelligent navigation, although they can provide some support, the equipment deployment cost is high and the maintenance is complex, especially difficult to promote in large-scale parking lots or urban environments. In addition, a single data source is greatly affected by environmental interference (such as signal occlusion or sensor failure), and the data reliability is insufficient.

[0009] Problem 3: Most of the existing management means rely on periodic inspections or static planning, lacking real-time and continuous optimization capabilities. This non-real-time nature leads to the inability to dynamically adjust charging pile and parking space resources during peak demand or emergencies, missing the best allocation opportunity and causing excessive waiting time for users.

[0010] Problem 4: Although it is possible to predict demand by monitoring vehicle battery levels and parking lot status, traditional technologies mainly rely on simple threshold judgments and fail to fully utilize modern data analysis technologies (such as deep learning and distributed computing) to deeply mine multi-modal data. Simple threshold methods are prone to misjudgment and difficult to provide accurate resource allocation strategies, limiting the intelligence level of the management system.

[0011] Therefore, a multi-mode vehicle charging and parking collaborative management method based on the Internet of Things is needed to solve the above problems. Summary of the Invention

[0012] Technical Problems to be Solved In view of the deficiencies of the prior art, the present invention provides a multi-mode vehicle charging and parking collaborative management method based on the Internet of Things, which solves the problems mentioned in the above background technology.

[0013] Technical Solutions

[0014] To achieve the above objectives, the present invention is realized through the following technical solutions: A multi-mode vehicle charging and parking collaborative management method based on the Internet of Things, including a main control system and a parking lot. The application steps of the main control system are sequentially multi-modal data collection and dynamic parking and charging constraint initialization, multi-scale parallel parking and charging model construction and preliminary allocation, multi-scale error feedback and cross-node collaborative evaluation, adaptive constraint adjustment and asynchronous distributed iterative optimization, global parking and charging plan integration and scalability verification, and output of the optimized plan and industrial application encapsulation. The parking lot includes a barrier gate, a guardrail, a carriage, a charging pile, and a ground lock. At the entrance of the parking lot, there are a new energy license plate recognition camera, an entrance license plate recognition camera, a gate machine, and an exit license plate recognition camera. In the collaborative management method, each step is closely related. Multi-modal data collection and dynamic parking and charging constraint initialization is responsible for collecting vehicle and parking lot-related data and generating parking and charging constraint conditions, which is the basis for subsequent steps. Multi-scale parallel parking and charging model construction and preliminary allocation generate a preliminary plan. Then, multi-scale error feedback and cross-node collaborative evaluation compare the preliminary plan with the actual demand and calculate the error. Adaptive constraint adjustment and asynchronous distributed iterative optimization adjust and optimize the parking and charging constraints based on the error. The optimization results are integrated into a global parking and charging plan. After safety verification and scalability testing, finally, the output of the optimized plan and industrial application encapsulation link outputs the plan and applies it to the actual industrial scenario.

[0015] Preferably, when the master control system implements coordinated management of charging and parking, the specific steps are as follows: Sp1: Multi-modal data collection and initialization of dynamic charging and parking constraints: Collect three types of data, namely vehicle power, location, and parking lot status, through the Internet of Things. Among them, the vehicle power is transmitted to the Internet of Things gateway by the in-vehicle battery management system through the CAN bus at a frequency of once every 5 seconds, the location is uploaded to the cloud server by the in-vehicle GPS module through the 5G network at a frequency of once per second, and the parking lot status is transmitted to the edge computing node by the current sensor and the parking space ultrasonic sensor of the charging pile through the ZigBee protocol at a frequency of once per second; preprocess and slice the data, allocate it to multiple parallel computing nodes, and use the vehicle power demand, grid load, and real-time status of the parking lot to generate dynamic charging and parking constraints by using a deep reinforcement learning algorithm. Each node synchronously shares the constraints through the TCP / IP protocol; Sp2: Construction of multi-scale parallel charging and parking models and preliminary allocation: Based on the dynamic constraints, construct multi-scale charging and parking models on each computing node, including a coarse-grained global resource allocation model based on the grid load balance equation and a fine-grained local parking space matching model based on vehicle trajectory prediction. The nodes generate preliminary charging and parking plans in parallel and integrate the results through distributed memory sharing technology; Sp3: Multi-scale error feedback and cross-node collaborative evaluation: Compare the preliminary charging and parking plan with the actual demand data, calculate the global error and local error, adjust the error weights by using the cross-node distributed gradient aggregation technology, optimize the local allocation accuracy of each node, and the master node coordinates the direction of the global plan; Sp4: Adaptive constraint adjustment and asynchronous distributed iterative optimization: Adjust the charging and parking constraints based on multi-scale error feedback, use the asynchronous stochastic gradient descent algorithm and the alternating direction multiplier method to update the parameters in parallel, record the optimization process through the blockchain and verify the consistency until the error converges; Sp5: Integration of the global charging and parking plan and verification of scalability: Integrate the optimization results of each node to generate a global charging and parking plan, verify the safety in combination with the data of the thermal imaging sensor, and test the scalability by simulating scenarios with different vehicle scales; Sp6: Output of the optimized plan and encapsulation for industrial application: Output the final charging and parking management plan, provide a parallel interface to support dynamic configuration of computing resources, and generate a performance report including computing efficiency and safety indicators.

[0016] Preferably, the multi-modal data collection and dynamic parking charging constraint initialization includes performing denoising and standardization processing on vehicle power, position, and parking lot status data. Among them, wavelet transform algorithm is used for denoising, the signal is decomposed with Daubechies D4 wavelet basis, the decomposition level is 4 layers, the noise with frequency higher than 50Hz is removed, and the signal-to-noise ratio after processing is increased to more than 30dB; the Z-score method is used for standardization to normalize the data to the range of 0-1; according to the spatio-temporal distribution characteristics of the data, the K-means clustering algorithm is used to divide the data into N slices, and N is determined by dividing the total amount of data by the single-node processing capacity of 10GB; the DQN model in the deep reinforcement learning algorithm is used to generate the initial parking charging constraint, the state space includes the vehicle power percentage, position coordinates, and parking space occupancy rate, the action space includes the charging power range of 50-150kW and parking space selection, the penalty factor is that the vehicle power is lower than 20%, the reward factor is the grid load balance, the network structure is a 3-layer fully connected network, with 64 neurons in the input layer, 128 neurons in the hidden layer, and 32 neurons in the output layer, and the constraints are synchronized among nodes through the MPI protocol.

[0017] Preferably, the multi-scale parallel parking charging model construction and preliminary allocation includes constructing a coarse-grained global model based on the grid load balance equation on each computing node to calculate the allocation ratio of charging piles and parking spaces in the whole city; based on the vehicle trajectory prediction data, using the long short-term memory network to construct a fine-grained local model to predict the arrival time of a single vehicle and the best parking space; each node calculates the parking charging scheme of the coarse-grained and fine-grained models in parallel, and integrates the results through the RDD technology of ApacheSpark.

[0018] Preferably, the multi-scale error feedback and cross-node collaborative evaluation includes comparing the preliminary parking charging scheme with the actual demand data, where the actual demand data includes the charging waiting time and the parking space utilization rate, and calculating the global error and the local error; adjusting the error weights through the distributed gradient aggregation technology, calculating the weight values using the weighted least squares method, and the master node generates the optimization direction based on the global error and broadcasts it to each node through the UDP protocol.

[0019] Preferably, the adaptive constraint adjustment and asynchronous distributed iterative optimization includes adjusting the parking charging constraints according to the multi-scale error feedback, and using the support vector regression algorithm to predict the adjustment direction of the charging power range and the parking space allocation priority; updating the local parameters with a learning rate of 0.01 through the asynchronous stochastic gradient descent algorithm, and coordinating the consistency with a penalty factor of 0.1 using the alternating direction method of multipliers; using the blockchain PBFT algorithm to record each adjustment and verify the consistency of parameter updates.

[0020] Preferably, the method includes a parking charging navigation mechanism based on acoustic wave positioning. Acoustic wave transmitters are deployed in the parking lot with a frequency range of 20 kHz - 40 kHz. The vehicle receives signals through an on-vehicle microphone and uses a triangulation algorithm to calculate the positions of available charging parking spaces; each node processes the acoustic wave data in parallel, generates a navigation path, and synchronizes it to the vehicle display screen.

[0021] Preferably, the method includes a distributed optimization mechanism based on swarm intelligence. Each vehicle operates the ant colony algorithm as an Internet of Things node, exchanges parking charging information with nearby vehicles through Bluetooth communication, and calculates a globally optimal allocation scheme; each node performs local optimization in parallel, and the master node aggregates the results every 15 seconds to generate a dynamic parking charging queue.

[0022] Preferably, the integration and scalability verification of the global parking charging scheme includes storing the optimization results of each node using the Hadoop distributed file system. The master node synthesizes the global scheme through the weighted voting method, and the weights are determined by the reciprocals of the errors of each node; the scalability is verified through test cases simulating vehicle scales from 1000 vehicles to 1 million vehicles. The verification metrics include a matching rate greater than 95%, a calculation time less than 60 seconds, and a parallel efficiency greater than 90%.

[0023] Preferably, the method includes a super-large-scale optimization mechanism based on quantum computing. The parking charging data of the whole city is collected through the Internet of Things and uploaded to the quantum computing platform; the quantum annealing algorithm is used to optimize the charging power allocation and parking space matching in parallel to generate a global scheme; each node synchronously updates the results, and the verification accuracy is greater than 98%.

[0024] Preferably, the method includes a security warning mechanism based on thermal imaging. The temperatures of the charging piles and vehicles are monitored through Internet of Things thermal imaging sensors at a sampling frequency of once per second; a convolutional neural network is used to analyze the temperature distribution to detect abnormal heat sources with a temperature greater than 80 °C; the master node broadcasts a warning signal through the TCP protocol, triggering cooling measures such as reducing the charging power to 50% and starting the ventilation device. Each node updates the parking charging scheme in parallel to avoid the abnormal area. The method includes introducing a data security mechanism for federated learning. During the multi-modal data collection process, each data source performs model training locally and only uploads the model parameter updates, which are aggregated and updated on the central server through encryption. The vehicle side uses its own power and location data to train the model locally and sends the parameter updates such as the model gradients to the central server through an encrypted channel. After the central server aggregates the parameter updates of all participating parties, it distributes the new parameters to each data source for continued local training, thereby protecting data privacy and using multi-source data to improve the performance of the collaborative management model.

[0025] Beneficial effects

[0026] The present invention provides a multi-mode vehicle charging and parking collaborative management method based on the Internet of Things. It has the following beneficial effects: 1. The present invention collects multi-modal data such as vehicle power, location, and parking lot status in real time through the Internet of Things, can timely detect changes in charging and parking demands, effectively avoid congestion and waiting problems caused by lagging resource allocation, and improve the real-time performance and efficiency of management. By using a deep reinforcement learning algorithm to generate dynamic parking and charging constraints and combining in-depth analysis of multi-modal data, the system not only relies on simple threshold judgments, but also provides a more accurate resource allocation strategy through comprehensive evaluation of vehicle status and grid load, reducing the probability of allocation errors.

[0027] 2. The present invention utilizes distributed computing and multi-scale modeling techniques to construct coarse-grained and fine-grained parking and charging models, achieving global optimal allocation of charging pile and parking space resources. Long short-term memory networks predict vehicle arrival times and optimal parking spaces, significantly improving matching accuracy and reducing user waiting times. By using asynchronous stochastic gradient descent and the alternating direction method of multipliers to optimize parameters in parallel, the system can quickly respond to demand changes during peak periods, ensure grid load balance, and reduce power system risks caused by uneven load.

[0028] 3. The present invention introduces thermal imaging safety warning and federated learning mechanisms, enhancing the safety of the charging process and data privacy protection. The thermal imaging sensor combines a convolutional neural network to monitor temperature in real time, detect abnormal heat sources, and trigger power reduction or ventilation measures, effectively avoiding safety hazards caused by overheating. Federated learning protects the privacy of vehicle and user data through local training and encrypted gradient uploading, while using multi-source data to improve model performance, achieving a balance between safety and efficiency.

[0029] 4. This method improves the user experience and system intelligence level through acoustic wave positioning navigation and personalized recommendation mechanisms. The acoustic wave positioning technology accurately navigates to an idle parking space using the triangulation algorithm, reducing user search time. The collaborative filtering algorithm analyzes user historical habits, provides personalized parking and charging solutions, and recommends the optimal charging pile and parking space, significantly improving user satisfaction. Users can provide feedback through in-vehicle terminals or mobile phone APPs to further optimize the recommendation model, forming a continuous improvement loop for the user experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 is the overall framework diagram of the present invention; Figure 2 is the specific flowchart of the present invention; Figure 3 is the simulation diagram of the overall method of the present invention; Figure 4 is the schematic diagram of the actual parking lot of the present invention; Figure 5 is the control diagram of the parking lot system of the present invention.

[0031] Legend Explanation: 1. Parking lot; 2. Vehicle stopper; 3. Guardrail; 4. Carriage; 5. New energy license plate recognition camera; 6. Charging pile; 7. Ground lock; 8. Import license plate recognition camera; 9. Gate; 10. Export license plate recognition camera. Detailed implementation mode

[0032] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention. Specific embodiment 1: As Figures 1-5 shown, a multi-mode vehicle charging and parking collaborative management method based on the Internet of Things, the method includes the following steps: Sp1: Multi-modal data collection and dynamic charging and parking constraint initialization: Collect three types of data of vehicle power, position, and parking lot status through the Internet of Things. Among them, the vehicle power is transmitted to the Internet of Things gateway by the in-vehicle battery management system through the CAN bus at a frequency of once every 5 seconds, the position is uploaded to the cloud server by the in-vehicle GPS module through the 5G network at a frequency of once per second, and the parking lot status is transmitted to the edge computing node by the current sensor and the parking space ultrasonic sensor of the charging pile through the ZigBee protocol at a frequency of once per second; Preprocess and slice the data, allocate it to multiple parallel computing nodes, and use the vehicle power demand, grid load, and real-time status of the parking lot to generate dynamic charging and parking constraints using a deep reinforcement learning algorithm. Each node synchronously shares the constraints through the TCP / IP protocol; Sp2: Multi-scale parallel charging and parking model construction and preliminary allocation: Build a multi-scale charging and parking model based on dynamic constraints on each computing node, including a coarse-grained global resource allocation model based on the grid load balance equation and a fine-grained local parking space matching model based on vehicle trajectory prediction. The nodes generate preliminary charging and parking plans in parallel and integrate the results through distributed memory sharing technology; Sp3: Multi-scale error feedback and cross-node collaborative evaluation: Compare the preliminary charging and parking plan with the actual demand data, calculate the global error and local error, use the cross-node distributed gradient aggregation technology to adjust the error weight, each node optimizes the local allocation accuracy, and the master node coordinates the global plan direction; Sp4: Adaptive constraint adjustment and asynchronous distributed iterative optimization: Adjust the charging and parking constraints based on multi-scale error feedback, use the asynchronous stochastic gradient descent algorithm and the alternating direction multiplier method to update the parameters in parallel, record the optimization process through the blockchain and verify the consistency until the error converges; Sp5: Global Charging Scheme Integration and Scalability Verification: Integrate the optimization results of each node to generate a global charging scheme, verify security in combination with thermal imaging sensor data, and test scalability by simulating scenarios with different vehicle scales; Sp6: Output of the Optimization Scheme and Industrial Application Encapsulation: Output the final charging management scheme, provide parallel interfaces to support dynamic configuration of computing resources, and generate a performance report including computing efficiency and security metrics.

[0034] Multi-modal data collection and dynamic charging constraint initialization include denoising and standardizing the vehicle battery level, location, and parking lot status data. For denoising, the wavelet transform algorithm is used, with the signal decomposed using the Daubechies D4 wavelet basis for 4 layers, removing noise with a frequency higher than 50 Hz, and the signal-to-noise ratio is increased to over 30 dB after processing. For standardization, the Z-score method is used to normalize the data to the range of 0 - 1. According to the spatio-temporal distribution characteristics of the data, the K-means clustering algorithm is used to divide the data into N slices, where N is determined by dividing the total data volume by the single-node processing capacity of 10 GB. The DQN model in the deep reinforcement learning algorithm is used to generate initial charging constraints. The state space includes the vehicle battery percentage, location coordinates, and parking space occupancy rate. The action space includes the charging power range of 50 - 150 kW and parking space selection. The penalty factor is when the vehicle battery level is below 20%, and the reward factor is the grid load balance. The network structure is a 3-layer fully connected network with 64 neurons in the input layer, 128 neurons in the hidden layer, and 32 neurons in the output layer. The constraints are synchronized between nodes through the MPI protocol.

[0035] Multi-scale parallel charging model construction and preliminary allocation include constructing a coarse-grained global model based on the grid load balance equation on each computing node to calculate the allocation ratio of charging piles and parking spaces in the whole city; constructing a fine-grained local model using the long short-term memory network based on vehicle trajectory prediction data to predict the arrival time of a single vehicle and the best parking space; each node calculates the charging scheme of the coarse-grained and fine-grained models in parallel, and the results are integrated through the RDD technology of ApacheSpark.

[0036] Multi-scale error feedback and cross-node collaborative evaluation include comparing the preliminary charging scheme with the actual demand data, where the actual demand data includes charging waiting time and parking space utilization rate, and calculating the global error and local error; adjusting the error weights through distributed gradient aggregation technology, calculating the weight values using the weighted least squares method, and the master node generates the optimization direction based on the global error and broadcasts it to each node through the UDP protocol.

[0037] Adaptive constraint adjustment and asynchronous distributed iterative optimization include adjusting the charging constraints according to multi-scale error feedback, and using the support vector regression algorithm to predict the adjustment direction of the charging power range and parking space allocation priority; updating local parameters with a learning rate of 0.01 through the asynchronous stochastic gradient descent algorithm, and coordinating consistency with a penalty factor of 0.1 by the alternating direction method of multipliers; using the blockchain PBFT algorithm to record each adjustment and verify the consistency of parameter updates.

[0038] The method includes a charging navigation mechanism based on acoustic wave positioning. Acoustic wave transmitters are deployed in the parking lot with a frequency range of 20 kHz - 40 kHz. Vehicles receive signals through on-vehicle microphones and calculate the positions of available charging parking spaces using the triangulation algorithm; each node processes acoustic wave data in parallel, generates a navigation path and synchronizes it to the vehicle display screen.

[0039] The method includes a distributed optimization mechanism based on swarm intelligence. Each vehicle runs the ant colony algorithm as an Internet of Things node, exchanges charging information with nearby vehicles through Bluetooth communication, and calculates the globally optimal allocation scheme; each node executes local optimization in parallel, and the master node aggregates the results every 15 seconds to generate a dynamic charging queue.

[0040] Global charging scheme integration and scalability verification include using the Hadoop distributed file system to store the optimization results of each node. The master node synthesizes the global scheme through the weighted voting method, and the weights are determined by the reciprocals of the errors of each node; verifying scalability through test cases simulating vehicle scales from 1000 to 1 million vehicles. The verification metrics include a matching rate greater than 95%, a calculation time less than 60 seconds, and a parallel efficiency greater than 90%.

[0041] The method includes a super-large-scale optimization mechanism based on quantum computing. The charging data of the whole city is collected through the Internet of Things and uploaded to the quantum computing platform; the quantum annealing algorithm is used to optimize the charging power distribution and parking space matching in parallel to generate a global scheme; each node synchronously updates the results, and the verification accuracy is greater than 98%.

[0042] The method includes a thermal imaging-based safety warning mechanism that monitors the temperatures of charging piles and vehicles through Internet of Things thermal imaging sensors at a sampling frequency of once per second; uses a convolutional neural network to analyze the temperature distribution and detect abnormal heat sources with a temperature greater than 80°C; the master node broadcasts a warning signal via the TCP protocol, triggering cooling measures such as reducing the charging power to 50% and starting the ventilation device, and each node updates the parking and charging plan in parallel to avoid the abnormal area. The method includes introducing a data security mechanism for federated learning. During the multi-modal data collection process, each data source trains the model locally and only uploads the model parameter updates, which are aggregated and updated on the central server through encryption. The vehicle terminal uses its own power and location data to train the model locally and sends the parameter updates such as the model gradients to the central server through an encrypted channel. After the central server aggregates the parameter updates of all participating parties, it distributes the new parameters to each data source for continued local training, thereby protecting data privacy and using multi-source data to improve the performance of the collaborative management model. The method includes a user personalized service recommendation mechanism. When outputting the final parking and charging management plan, it analyzes the user's historical charging and parking habits (charging time preferences, frequently visited parking lots, etc.), and uses a recommendation algorithm to provide the user with a personalized parking and charging plan. For example, for users who often charge from 7 pm to 9 pm on weekdays, it preferentially recommends parking lots with available charging piles during this period and close to the user's workplace, and provides the estimated charging cost and waiting time. The user can provide feedback on the recommended plan on the in-vehicle terminal or mobile phone APP to optimize the recommendation model and enhance the user experience.

[0043] As Figure 4 and Figure 5 shown, Figure 4 Parking lot 1 in Figure 5 is a scenario. The present invention does not limit the usage scenarios. At the same time The C++ underlying program on the server translates the underlying data into JSON strings that can be understood by the WEB side and the applet side. The communication between the underlying C++ program and the WEB background uses the HTTP protocol. The underlying protocol conversion C++ program is responsible for handling the communication protocols of the ground lock and the charging pile. The WEB background realizes the charging function by configuring and controlling the ground lock and the charging pile through the interfaces provided by the underlying protocol conversion C++ program. The entrance and exit recognition cameras cooperate with the WEB background program and the WeChat applet through the MQTT protocol to realize the charging function of the parking lot. The WEB background end is responsible for providing the administrator with the display and configuration operations of various system information, and the WeChat applet end is responsible for providing the customer with an interactive interface for payment and charging, which is applicable to similar parking lot areas. Specific Embodiment 2: As Figures 1-5 shown, the following are the complete steps of the content described in Embodiment 1: This method collects multimodal data through the Internet of Things (IoT), and uses various technologies such as deep reinforcement learning, distributed computing, and blockchain verification to collaboratively manage vehicle charging and parking demands. From data collection to solution output, the method forms a closed loop, gradually optimizing resource allocation to ensure efficiency, security, and user experience. The following are the specific implementations, connections, and application scenarios of each step: The system first collects three types of data through the IoT: vehicle battery level, location, and parking lot status. The vehicle battery level is transmitted from the in-vehicle battery management system to the IoT gateway via the CAN bus at a frequency of once every 5 seconds. The gateway monitors the battery capacity and encrypts and sends it to the edge computing node via the MQTT protocol. The vehicle location is collected by the in-vehicle GPS module at a rate of once per second for longitude and latitude and uploaded to the cloud server via the 5G network with a latency of less than 50 milliseconds. The parking lot status is collected by the charging pile current sensor and the parking space ultrasonic sensor at a rate of once per second and transmitted to the edge computing node via the ZigBee protocol. The node filters out outliers, and the data is preprocessed at the edge node. Wavelet transform algorithm is used for denoising, and Z-score method is used for normalization to the range of 0 - 1. The data is divided into N shards using the K-means clustering algorithm and distributed to parallel computing nodes. Then, based on the vehicle battery demand, grid load, and parking lot 1 status, the system uses the DQN model to generate dynamic charging and parking constraints. The state space includes battery percentage, location coordinates, and parking space occupancy rate. The action space is the charging power of 50 - 150 kW and parking space selection. The penalty factor is when the battery level is below 20%, and the reward factor is grid load balancing. The network is a 3-layer fully connected network. After training for 1000 rounds, the constraints are generated and synchronized to each node every 10 seconds via the TCP / IP protocol. This step lays the foundation for subsequent modeling.

[0045] Each node constructs a multi-scale charging and parking model based on the constraints. The coarse-grained global model calculates the resource allocation ratio for the whole city based on grid load balancing, with the goal of minimizing the peak load. The fine-grained local model uses a long short-term memory network to predict the arrival time and the best parking space of a single vehicle. The input is 30 minutes of GPS and speed data. The network is 3 layers, and the training data is the historical trajectories in the cloud. The nodes perform parallel computing, and the results are integrated through Apache Spark RDD. This step uses Sp1 data to generate a preliminary solution.

[0046] The preliminary solution is compared with the actual demands (charging waiting time, parking space utilization rate), and the global error (difference from the mean by 300 seconds) and local error (difference from 180 seconds for a single vehicle) are calculated. The error adjusts the weights through distributed gradient aggregation, calculated using the weighted least squares method. The master node generates the optimization direction based on the global error and broadcasts it via the UDP protocol every second. This step evaluates the Sp2 solution and provides feedback for optimization.

[0047] Based on error feedback, the system adjusts the constraints. The support vector regression algorithm predicts the power range and priority, and the training data is a 1-hour error sequence. The nodes update the parameters through asynchronous stochastic gradient descent and the alternating direction method of multipliers. The blockchain PBFT records and iterates until the error < 0.05 or 500 times. This step optimizes the Sp3 feedback and forms a closed loop with the Sp1 constraint.

[0048] In the integration phase, Hadoop stores the results, and the master node synthesizes the scheme through weighted voting. The thermal imaging sensor monitors the temperature, and the convolutional neural network detects anomalies above 80°C, triggering a power reduction to 50% or ventilation (500 m³ / h). The scalability test is carried out for vehicle scales from 1000 to 1 million, verifying a matching rate > 95%, a calculation time < 60 seconds, and an efficiency > 90%. This step integrates the Sp4 results and verifies the safety.

[0049] The final output scheme provides a parallel interface. Federated learning protects privacy. The data source conducts local training, uploads differentially private encrypted gradients, and the central server aggregates them using FedAvg. Personalized recommendations utilize collaborative filtering, analyze user habits to generate schemes, and users provide feedback through in-vehicle terminals or mobile phone APPs. Additional mechanisms include acoustic positioning, ant colony algorithms, and quantum annealing. Specific Embodiment 3: As Figures 1-5 shown, the key algorithms mentioned in Embodiment 1 are analyzed in detail below, including their core mathematical formulas and explanations: Deep reinforcement learning algorithm (SP1):

[0051] Among them is the Q-value of choosing action in state , representing the long-term expected reward; is the current state, including the vehicle battery percentage (0 - 100%), location coordinates (latitude and longitude, range such as 30 - 40° north latitude, east longitude), and parking space occupancy rate (0 - 1); is the action, including the charging power (50 - 150 kW) and parking space selection (number 1 - N, where N is the total number of parking spaces); is the immediate reward, defined as the grid load balance (variance < 10 kW is +1) minus the penalty for the battery level below 20% (-1); is the learning rate, with a value of 0.001, controlling the Q-value update speed based on ensuring convergence; is the discount factor, with a value of 0.9, measuring the importance of future rewards based on the long-term planning of Poisson charging; is the next state, updated by vehicle movement and charging;

[0052] Generate dynamic charging constraints through this algorithm to solve the problems of low vehicle battery power and uneven grid load. The deep reinforcement learning algorithm approximates the Q value through a neural network (with 64 - 128 - 32 neurons in 3 layers), inputs the state, outputs the G value of each action, and selects the action with the largest Q value (such as allocating 100 kW power and parking space A5 to a vehicle with 15% battery power). After 1000 rounds of training, the constraints are stably applied in Sp1. The constraints ensure that vehicles with low battery power are charged first and the grid load is balanced.

[0053] Asynchronous Stochastic Gradient Descent Algorithm (ASGD) - Sp4:

[0054] where The parameters of the t-th iteration, such as the charging power range (50 - 150 kW) and the parking space priority (0 - 1); is the learning rate, with a value of 0.01, based on the stability of the charging parameter adjustment; ; The number of iterations, ranging from 0 to 500. This method parallelly optimizes the parameters of each node through this algorithm to solve the communication delay problem of synchronous calculation.

[0055] Alternating Direction Method of Multipliers (ADMM) - Sp4:

[0056] where is the local parameter of the i-th node, such as power distribution; The local objective function, such as the local error (the difference between the waiting time of a single vehicle and 180 seconds); is the global parameter to coordinate the consistency of each node; is the Lagrange multiplier, initially 0, to update the coordination deviation; is the penalty factor, with a value of 0.1, based on the charging consistency requirement; The number of nodes, ranging from 1 to 1024. This algorithm solves the problem of asynchronous update conflicts. After the nodes optimize the local error, the master node aggregates the global parameters and the multiplier adjusts the deviation.

[0057] Wavelet Transform Algorithm - Sp1:

[0058] where The wavelet coefficient of the signal x(t) at scale a and position b; is the original data, such as the battery power time series (0 - 100%); is the mother wavelet, and Daubechies D4 is selected, with 4th-order smoothing; The scale, taking 2 values of j (j = 1 - 4), decomposing 4 layers; Translation parameter, with a range of the data length (e.g., 300 seconds). This method removes high-frequency noise through this algorithm to solve the problem of poor data quality.

[0059] Triangulation algorithm - acoustic wave navigation:

[0060] The triangulation algorithm is derived from the geometric positioning principle, where is the distance from the vehicle to the i-th transmitter, calculated from the acoustic wave time difference; is the vehicle position, with a range of parking lot coordinates; is the transmitter position, which is known. In this method, the triangulation algorithm is used to locate free parking spaces for users to solve the problem of insufficient navigation accuracy.

[0061] Z-score normalization - Sp1:

[0062] where Normalized value, with a range of 0 - 1 (post-normalization); Original value, such as battery power (0 - 100%); Mean value, such as the average battery power of 50%; Standard deviation, such as a battery power fluctuation of 10%. This solves the problem of differences in multi-modal data.

[0063] K-means clustering algorithm - Sp1:

[0064] where is the objective function, minimizing the within-cluster distance; ; ; Assignment indicator, 1 means belonging to the k-th cluster; Total amount of data (GB); Number of clusters, determined by N / 10GB. Specific embodiment four: Such as Figures 1-5 shown, the following use cases are provided according to the above content: Use case 1: Charging and parking collaborative management in the urban commercial area during peak hours Scenario description: In a certain urban commercial area, from 3 pm to 6 pm on weekends is the peak hour, involving 500 electric vehicles, with 50 charging piles and 100 parking spaces in the area. The peak grid load needs to be controlled below 5000 kW, and users require fast charging and convenient parking.

[0066] The specific logic is as follows: Data Acquisition and Constraint Initialization: The system collects the power, location, and parking lot status of 500 vehicles through the Internet of Things. The edge node uses wavelet transform for denoising, the signal-to-noise ratio is increased to 32 dB, the data is normalized to the range of 0 - 1, and K-means clustering is divided into 5 shards, with a total data volume of 50 GB and 10 GB processed per node. DQN generates constraints to identify 100 vehicles with a power level below 20% and preferentially allocate 100 kW of power. The grid load balancing reward is triggered, and the variance is maintained below 10 kW.

[0067] Multi-scale Model Construction: 5 nodes perform parallel computing. The coarse-grained model allocates 70 parking spaces for charging, and the fine-grained model predicts the vehicle arrival time and parking spaces through LSTM. The preliminary plan is integrated through RDD technology within 4 seconds.

[0068] Error Feedback: The actual average waiting time is 350 seconds, the target is 300 seconds. The weighted least squares method is used to adjust the weights, and the master node broadcasts the optimization direction, increasing the power of 10 charging piles to 120 kW.

[0069] Constraint Optimization: SVR predicts the power adjustment direction, ASGD updates the parameters to 110 kW with a learning rate of 0.01, ADMM coordinates the consistency with a penalty factor of 0.1, and the blockchain verifies and updates through PBFT, reducing the error to 0.04.

[0070] Scheme Integration: The weighted voting synthesizes the scheme. 70 charging parking spaces match 430 vehicles. The thermal imaging detects that the temperature of the charging pile exceeds 82 °C, reduces the power to 50%, and starts the ventilation device with an air volume of 500 m³ / h. The scalability verification shows that the matching rate reaches 96%.

[0071] Output and Recommendation: The scheme is output through the API. Federated learning protects privacy with an accuracy of 90%. Collaborative filtering recommends users to parking spaces, with a cost of 40 yuan, a waiting time of 3 minutes, and a user rating of 4 points.

[0072] Additional Mechanism: Acoustic wave positioning navigates to the specified parking space with an accuracy of 0.1 meters. The ant colony algorithm optimizes the queue, and 10 vehicles cooperate for allocation.

[0073] Result: Among 500 vehicles, 480 vehicles are successfully charged and parked. The average waiting time is reduced to 280 seconds, the grid load is 4900 kW, and the user satisfaction is increased to 85%.

[0074] Use Case 2: Highway Service Area - Fast Charging for Long-distance Vehicles Scenario Description: At a service area on the G2 Beijing-Shanghai Expressway, there are 2000 long-distance electric vehicles, mainly logistics vehicles, during the peak holiday period. There are 30 charging piles, and the grid capacity is 3000 kW. Users need to charge quickly and continue their journey.

[0075] The specific steps are as follows: Collect data of 2000 vehicles, and slice the power and location information to 20 nodes. The DQN constraint preferentially allocates 150 kW power to 500 vehicles with a battery level below 20%.

[0076] The coarse-grained model allocates 25 charging piles for fast charging, and the LSTM predicts the vehicle arrival time.

[0077] The average waiting time is 400 seconds, the target is 300 seconds, and it is recommended to add 5 150 kW charging piles after weighted adjustment.

[0078] The SVR predicts the power range to 140 - 160 kW, and the ASGD and ADMM are used for optimization, reducing the error to 0.03.

[0079] The integrated solution matches 1800 vehicles. The temperature of a certain charging pile is detected to be too high by thermal imaging, and the power is reduced to 75 kW. The quantum optimization verifies the scenario of 100,000 vehicles, with an accuracy of 98%.

[0080] Output the solution, recommend users to the designated charging pile, the cost is 80 yuan, the waiting time is 8 minutes, and the user rating is 5 points.

[0081] Additional mechanism: Acoustic wave positioning for navigation to the designated charging pile, and quantum annealing for global allocation optimization.

[0082] Result: 1900 vehicles complete charging, the waiting time is reduced to 290 seconds, the grid load is 2950 kW, and the efficiency is increased to 95%.

[0083] Use Case 3: Residential Community - Optimization of Nighttime Residents' Charging Scenario description: In a certain residential community, 50 residents charge their vehicles from 7 pm to 9 pm. There are 20 charging piles, and the grid capacity during the low valley period is 1000 kW. Users need to charge at low cost.

[0084] The specific steps are as follows: Collect data of 50 vehicles, slice it to 2 nodes, and the DQN optimizes the allocation during the low valley period, preferentially using 20 kW slow charging.

[0085] The coarse-grained allocation is 15 slow charging piles, and the LSTM predicts the time when residents return home.

[0086] The average waiting time is 150 seconds, the target is 180 seconds, and the weighted adjustment optimizes the direction.

[0087] The SVR predicts the priority, and the ASGD adjusts the parameters, reducing the error to 0.02.

[0088] The integration matches 48 vehicles, and there is no abnormality detected by thermal imaging.

[0089] Recommend users to the parking spaces, the cost is 15 yuan, the waiting time is 0 minutes, and federated learning protects privacy.

[0090] Additional mechanism: Ant colony algorithm for collaborative allocation.

[0091] Result: 49 vehicles were charged, waiting time was 120 seconds, grid load was 800kW, and costs were reduced by 30%.

[0092] Use Case 4: Industrial Park - Efficient Logistics Fleet Management Scenario description: In an industrial park, 200 electric logistics vehicles are charged collectively at 8 a.m., with 50 charging piles and a grid capacity of 4,000 kW.

[0093] Application logic: Collect data, shard to 5 nodes, and DQN prioritizes 100kW fast charging.

[0094] 45 fast charging piles are allocated at a coarse-grained level, and LSTM is used to predict attendance time.

[0095] Waiting time 200 seconds, target 180 seconds, optimize power distribution.

[0096] SVR is adjusted to 110kW, ASGD and ADMM are optimized.

[0097] Integrated and matched 190 vehicles, thermal imaging ensures safety.

[0098] Recommend users to the designated parking space, the fee is 60 yuan, and the waiting time is 5 minutes.

[0099] Additional mechanism: Quantum optimization improves efficiency.

[0100] Results: 195 vehicles were charged, waiting time was 180 seconds, grid load was 3900kW, and efficiency was 92%. Specific embodiments

[0101] like Figures 1-5 As shown, the following experimental data are provided based on the above content: Experimental setup: Environment: Simulate cities (5-1 million vehicles), highways (2,000-100,000 vehicles), residential areas (50-500 vehicles), and campuses (200-1,000 vehicles), using a Spark cluster (16 nodes, 32 GB memory / node) and a quantum simulator.

[0102] Benchmark comparison: traditional static scheduling (no dynamic constraints, no distributed optimization, no personalized recommendations).

[0103] Indicators: matching rate (proportion of successful charging and parking), waiting time (minutes), grid load balance (variance kW), calculation time (seconds), parallel efficiency (%), safety (proportion of abnormal heat sources), user satisfaction (score 1-5).

[0104]

[0105] The following analysis results can be obtained from the above data: Efficiency: The matching rate is increased by 9% - 19%, the waiting time is reduced by 43% - 53%, and the calculation time is shortened by 62% - 67%. Thanks to the asynchronous optimization of DQN dynamic constraints, ASGD, and ADMM, as well as the distributed computing of Spark.

[0106] Grid load balancing: The variance is reduced by 60% - 67%, achieved through the reward mechanism of DQN and the optimization of the coarse-grained model.

[0107] Parallel efficiency: It is increased by 36% - 64%, attributed to K-means sharding and multi-scale parallel modeling.

[0108] Security: The proportion of abnormal heat sources is reduced to 0% - 0.8%. The CNN thermal imaging detection and warning mechanism significantly reduces risks.

[0109] Scalability: The matching rate remains > 95% for vehicles from 500 to 1 million. The quantum annealing algorithm supports ultra-large-scale optimization.

[0110] User experience: The satisfaction is increased by 16% - 20%. Personalized recommendations (collaborative filtering) and navigation (acoustic positioning) meet user needs.

[0111] Privacy protection: Federated learning ensures that data does not leave the local area, and gradient encryption enhances security.

[0112] The following are the specific steps to implement and verify the beneficial effects in the above table: Experimental steps: Data generation: Use the Poisson distribution to generate vehicle arrival times (λ = 10 vehicles / minute in the city, λ = 50 vehicles / minute on the highway, λ = 2 vehicles / minute in the community, λ = 5 vehicles / minute in the park).

[0113] Randomly generate battery levels (0 - 100%), locations (latitude and longitude ranges such as 30 - 40°N, 110 - 120°E), and parking lot statuses (charging pile current 0 - 200A, parking space occupancy rate 0 - 1).

[0114] Operation of this solution: Sp1: Collect data (battery level every 5 seconds, location and status every second), denoise using wavelet transform, perform K-means clustering (N = data volume / 10GB), generate constraints using DQN (reward grid balance, punish battery level < 20%). Sp2: Coarse-grained model for resource allocation, use LSTM to predict arrival time and parking spaces, and integrate preliminary solutions using RDD. Sp3: Calculate errors (global target 300 seconds, local 180 seconds), and adjust weights using weighted least squares method. Sp4: Use SVR to predict adjustment directions, and optimize using ASGD (learning rate 0.01) and ADMM (penalty factor 0.1). Sp5: Integrate using weighted voting, detect thermal anomalies using CNN (trigger power reduction when > 80°C), simulate 1000 - 1 million vehicles to verify scalability. Sp6: Protect privacy using federated learning, recommend using collaborative filtering, and record user ratings.

[0115] Traditional method operation: Fixed power (e.g., 50kW / station), first-come-first-served for parking space allocation, no dynamic optimization, no thermal imaging detection, no recommendation.

[0116] Data collection: Record matching rate (proportion of vehicles successfully charged and parked), waiting time (minutes), grid load variance (kW), calculation time (seconds), parallel efficiency (%), proportion of abnormal heat sources (%), user satisfaction (1 - 5 points).

[0117] Analysis and comparison: Calculate the difference in metrics and the percentage improvement between this solution and the traditional method.

[0118] The following table shows the experimental data for specific scenarios, demonstrating the sources and bases for the achievements in the above table. Each metric includes the measured values and improvement margins for this solution and the traditional method, and the data is based on simulation experiments.

[0119] Specific embodiment six: As Figures 1-5 shown, the simulation diagram in Figure 3 is specifically analyzed as follows: Sub Figure 1 is the matching rate (%), showing the proportion of vehicles successfully allocated charging and parking spaces. The X-axis represents the simulation scenarios, and the Y-axis represents the percentage (%). Through sub Figure 1 the performance of the solution in resource allocation efficiency can be reflected. The higher the matching rate, the higher the success rate of vehicle charging and parking.

[0121] Sub Figure 2 is the waiting time distribution (minutes), used to show the waiting time distribution of vehicles from arrival to the start of charging. The X-axis represents the waiting time (minutes), and the Y-axis represents the probability, showing the central tendency of the user waiting time. The more concentrated and shorter the waiting time distribution, the higher the service efficiency.

[0122] Sub Figure 3 is the grid load curve (kW), which shows the variation of the grid load over time during the simulation. The X-axis represents time (minutes), and the Y-axis represents the load (kW), used to reflect the dynamic balance of the grid load. A smooth load curve that does not exceed the capacity indicates effective energy management.

[0123] Sub Figure 4 is the user satisfaction distribution (points), used to show the satisfaction distribution of users with charging and parking services. The X-axis represents the satisfaction (points), and the Y-axis represents the probability. The entire Sub Figure 4 represents the quality of the user experience. The more concentrated and higher the satisfaction distribution, the better the personalized service and navigation effect.

[0124] It should be noted that in this article, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or device comprising a series of elements not only includes those elements, but also includes other elements not expressly listed, or elements inherent to such process, method, article or device. Without further limitation, an element defined by the statement "comprising a reference structure" does not exclude the existence of additional identical elements in the process, method, article or device comprising the element.

[0125] Although the embodiments of the present invention have been shown and described, those of ordinary skill in the art can understand that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A multi-mode vehicle charging and parking collaborative management method based on the Internet of Things, characterized in that It includes a master control system and a parking lot (1). The application steps of the master control system are, in sequence, multi-modal data collection and dynamic charging and parking constraint initialization, multi-scale parallel charging and parking model construction and preliminary allocation, multi-scale error feedback and cross-node collaborative evaluation, adaptive constraint adjustment and asynchronous distributed iterative optimization, global charging and parking solution integration and scalability verification, and output of optimized solutions and industrial application encapsulation. The parking lot (1) contains a barrier gate (2), a guardrail (3), a carriage (4), a charging pile (6), and a ground lock (7). At the entrance of the parking lot (1), there are a new energy vehicle license plate recognition camera (5), an entrance vehicle license plate recognition camera (8), a gate machine (9), and an exit vehicle license plate recognition camera (10). In the collaborative management method, each step is closely related. Multi-modal data collection and dynamic charging and parking constraint initialization are responsible for collecting vehicle and parking lot-related data and generating charging and parking constraint conditions, which are the basis for subsequent steps. Multi-scale parallel charging and parking model construction and preliminary allocation generate a preliminary solution. Then, multi-scale error feedback and cross-node collaborative evaluation compare the preliminary solution with the actual requirements and calculate the error. Adaptive constraint adjustment and asynchronous distributed iterative optimization adjust and optimize the charging and parking constraints based on the error. The optimization results are integrated into a global charging and parking solution. After safety verification and scalability testing, finally, the output of optimized solutions and industrial application encapsulation links output the solution and apply it to the actual industrial scenario.

2. The multi-mode vehicle charging and parking collaborative management method based on the Internet of Things according to claim 1, characterized in that: When the master control system implements charging and parking collaborative management, the specific steps are as follows: Sp1: Multi-modal data collection and dynamic charging and parking constraint initialization: Collect three types of data, namely vehicle power, location, and parking lot status, through the Internet of Things. Among them, vehicle power is transmitted from the in-vehicle battery management system to the Internet of Things gateway through the CAN bus, the location is uploaded from the in-vehicle GPS module to the cloud server through the 5G network, and the parking lot status is transmitted from the current sensor of the charging pile and the ultrasonic sensor of the parking space to the edge computing node through the ZigBee protocol; preprocess and slice the data, and use the vehicle power demand, grid load, and real-time status of the parking lot to generate dynamic charging and parking constraints using a deep reinforcement learning algorithm. Each node synchronously shares the constraints through the TCP / IP protocol. Among them, the new energy vehicle license plate recognition camera (5) is used to identify new energy vehicles entering the parking lot, so as to implement the rule of free parking fees for new energy vehicles during charging in the follow-up. At the same time, the entrance vehicle license plate recognition camera (8) and the exit vehicle license plate recognition camera (10) are used to identify other vehicles and charge according to the national parking fee rules. The ground lock (7) is connected to the 4G transparent transmission through the 485 interface, and then connected to the C++ program that converts the underlying protocol to the network protocol on the server. When the new energy vehicle license plate recognition camera (5) recognizes a new energy vehicle, the ground lock (7) automatically descends to prevent unrelated vehicles from occupying the charging position, and the software background can remotely configure and forcibly control the ground lock (7) through this C++ program; Sp2: Multi-scale Parallel Charging Model Construction and Preliminary Allocation: On each computing node, construct a multi-scale charging model based on dynamic constraints, including a coarse-grained global resource allocation model based on the grid load balance equation and a fine-grained local parking space matching model based on vehicle trajectory prediction. The nodes generate preliminary charging plans in parallel and integrate the results through distributed memory sharing technology; Sp3: Multi-scale Error Feedback and Cross-node Collaborative Evaluation: Compare the preliminary charging plan with the actual demand data, calculate the global error and local error, adjust the error weights using cross-node distributed gradient aggregation technology, optimize the local allocation accuracy of each node, and the master node coordinates the global plan direction; Sp4: Adaptive Constraint Adjustment and Asynchronous Distributed Iterative Optimization: Adjust the charging constraints based on multi-scale error feedback, use asynchronous stochastic gradient descent algorithm and alternating direction multiplier method to update parameters in parallel, record the optimization process through blockchain and verify the consistency until the error converges; Sp5: Global Charging Plan Integration and Scalability Verification: Integrate the optimization results of each node to generate a global charging plan, verify the safety in combination with thermal imaging sensor data, and test the scalability by simulating scenarios with different vehicle scales; Sp6: Output Optimization Plan and Industrial Application Encapsulation: Output the final charging management plan, provide a parallel interface to support dynamic configuration of computing resources, and generate a performance report including computing efficiency and safety indicators.

3. The multi-mode vehicle charging and parking collaborative management method based on the Internet of Things according to claim 2, wherein: In the multi-modal data collection and dynamic charging constraint initialization step: Denoise and standardize the vehicle battery level, location, and parking lot status data. For denoising, use the wavelet transform algorithm, decompose the signal with the Daubechies D4 wavelet basis, and remove the noise with a frequency higher than 50Hz. For standardization, use the Z-score method to normalize the data. According to the spatio-temporal distribution characteristics of the data, use the K-means clustering algorithm to divide the data into N slices, where N is determined by dividing the total data volume by the single-node processing capacity of 10GB. Use the DQN model in the deep reinforcement learning algorithm to generate initial charging constraints. The state space covers the vehicle battery percentage, location coordinates, and parking space occupancy rate, and the action space includes the charging power range of 50 - 150kW and parking space selection.

4. The multi-mode vehicle charging and parking collaborative management method based on the Internet of Things according to claim 2, characterized in that: The multi-scale parallel charging model construction and preliminary allocation step includes: On each computing node, construct a coarse-grained global model based on the grid load balance equation to calculate the allocation ratio of charging piles and parking spaces in the whole city; Based on the vehicle trajectory prediction data, construct a fine-grained local model with the help of long short-term memory network to predict the arrival time of a single vehicle and the best parking space; Each node calculates the charging plans of the coarse-grained and fine-grained models in parallel.

5. The multi-mode vehicle charging and parking collaborative management method based on the Internet of Things according to claim 2, characterized in that: The multi-scale error feedback and cross-node collaborative evaluation step is as follows: Compare the preliminary charging plan with the actual demand data, where the actual demand data includes the charging waiting time and parking space utilization rate, and then calculate the global error and local error; Adjust the error weights through distributed gradient aggregation technology, calculate the weight value using the weighted least squares method, and the master node generates an optimization direction based on the global error and broadcasts it to each node through the UDP protocol.

6. The multi-mode vehicle charging and parking collaborative management method based on the Internet of Things according to claim 1, characterized in that: The adaptive constraint adjustment and asynchronous distributed iterative optimization steps include: adjusting the charging constraints according to the multi-scale error feedback, and using the support vector regression algorithm to predict the adjustment direction of the charging power range and the parking space allocation priority; updating the local parameters with a learning rate of 0.01 through the asynchronous stochastic gradient descent algorithm, and using the blockchain PBFT algorithm to record each adjustment.

7. The multi-mode vehicle charging and parking collaborative management method based on the Internet of Things according to claim 2, characterized in that: The method includes a charging navigation mechanism based on acoustic positioning. Acoustic wave transmitters are deployed in the parking lot, and vehicles receive signals through on-vehicle microphones, and use the triangulation algorithm to calculate the positions of idle charging parking spaces; each node processes the acoustic wave data in parallel, generates a navigation path and synchronizes it to the vehicle display screen.

8. The multi-mode vehicle charging and parking collaborative management method based on the Internet of Things according to claim 2, characterized in that: The method includes a distributed optimization mechanism based on swarm intelligence. Each vehicle runs the ant colony algorithm as an Internet of Things node, exchanges charging information with nearby vehicles through Bluetooth communication, and calculates the globally optimal allocation scheme; each node executes local optimization in parallel, and the master node aggregates the results every 15 seconds to generate a dynamic charging queue.

9. The multi-mode vehicle charging and parking collaborative management method based on the Internet of Things according to claim 1, wherein: The global charging scheme integration and scalability verification include using the Hadoop distributed file system to store the optimization results of each node, and the master node synthesizes the global scheme through the weighted voting method, and the weights are determined by the reciprocals of the errors of each node.

10. The multi-mode vehicle charging and parking collaborative management method based on the Internet of Things according to claim 1, characterized in that: The method includes a super-large-scale optimization mechanism based on quantum computing. The charging data of the whole city is collected through the Internet of Things and uploaded to the quantum computing platform; The quantum annealing algorithm is used to optimize the charging power allocation and parking space matching in parallel to generate a global scheme; each node synchronously updates the results. At the same time, the method includes a safety warning mechanism based on thermal imaging. The temperatures of the charging piles and vehicles are monitored through Internet of Things thermal imaging sensors, and the convolutional neural network is used to analyze the temperature distribution to detect abnormal heat sources with a temperature greater than 80°C; the master node broadcasts a warning signal through the TCP protocol to trigger cooling measures such as reducing the charging power to 50% and starting the ventilation device, and each node updates the charging scheme in parallel to avoid the abnormal area.

Images