Electric vehicle charging demand prediction and scheduling method based on artificial intelligence

Abstract

The invention provides an electric vehicle charging demand prediction and scheduling method based on artificial intelligence, relates to the technical field of electric vehicle charging management, and integrates charging station historical data, user behavior modes and external environment factors through multi-mode deep learning, quantum computing optimization and dynamic user portrait generation technologies. Compared with a traditional method, the method has the advantages that the prediction error is reduced by about 30%, the prediction is particularly prominent in peak hours and extreme weather conditions, and meanwhile, charging resources in a region are integrated into a unified management unit through a virtual energy pool and an ecological self-adaptive scheduling mechanism, so that the prediction efficiency is improved. Distribution of the charging piles, new energy power generation and energy storage equipment is dynamically optimized, the utilization rate of the charging piles is increased by about 35%, and the utilization rate of new energy is increased by about 25%. According to the effect, resource waste is reduced, the contradiction between supply and demand of the charging station is relieved, more efficient charging service is provided for electric vehicle users, and meanwhile green development of an energy system is promoted.

Description

Technical Field

[0001] The present invention relates to the technical field of electric vehicle charging management, and specifically to a method for predicting and scheduling electric vehicle charging demands based on artificial intelligence. Background Art

[0002] According to an electric vehicle spatio-temporal scheduling method based on a dynamic charging navigation strategy disclosed in Chinese Patent Publication No. "CN118822147A", it includes: constructing an intelligent transportation system model and establishing an electric vehicle speed and energy consumption model based on the intelligent transportation system; based on the travel time of electric vehicles and the charging queue situation inside and outside the charging stations, constructing a travel decision-making model considering the interaction of electric vehicles, and analyzing the travel and charging time costs of electric vehicles; using an improved Dijkstra algorithm to construct an electric vehicle dynamic charging navigation model to perform real-time path planning for electric vehicle users and recommend the optimal charging station; 4. Conducting an orderly scheduling of the charging periods of electric vehicles arriving at the charging stations, reducing the peak-valley difference of the distribution network while meeting the charging demands of electric vehicle users, and achieving optimal control at the scheduling level. The present invention can realize the orderly charging of electric vehicles, thereby fully exerting the spatio-temporal schedulable potential of electric vehicles.

[0003] The above patent document and the prior art have the following technical problems when in use:

[0004] Problem 1: In the existing electric vehicle charging system, the prediction of charging demands mostly relies on a single data source such as historical charging records or simple statistical models, which is difficult to accurately capture the dynamic influence of multiple factors, resulting in large prediction errors, especially performing poorly during peak periods or abnormal situations. The allocation of charging resources is mostly based on static rules, lacking the ability of dynamic optimization, with low utilization rate of charging piles (usually less than 60%), and new energy power not being fully integrated and utilized, causing resource waste and supply-demand imbalance;

[0005] Problem 2: The charging system generally has problems such as long user waiting time (often exceeding 15 minutes during peak periods), low information transparency (such as unclear charging station status), and lack of personalized services, with generally low user satisfaction. In addition, the traditional system responds slowly to emergencies (such as charging pile failures, extreme weather), lacking the ability of adaptive adjustment, especially facing cold start problems when deployed in new areas, and the prediction and scheduling effects are not good;

[0006] Problem 3: In the existing charging system, the instability of new energy power generation is not fully considered, resulting in low utilization rate of new energy (usually less than 70%), large peak pressure on the power grid load, and high carbon emissions. At the same time, user privacy protection is insufficient, and the data security risk is high, restricting the possibility of multi-party cooperation. The existing system lacks an effective social incentive mechanism, with insufficient enthusiasm of users to participate in low-carbon behaviors, and the energy sharing and trading model is not yet mature, hindering the sustainable development of the energy system. Summary of the Invention

[0007] Technical Problems to be Solved

[0008] In view of the deficiencies of the prior art, the present invention provides an artificial intelligence-based electric vehicle charging demand prediction and scheduling method, which solves the following problems:

[0009] 1. Problems of low charging demand prediction accuracy and insufficient resource utilization efficiency;

[0010] 2. Problems of poor user experience and insufficient system adaptability;

[0011] 3. Problems of poor energy system sustainability and insufficient social collaboration.

[0012] Technical Solutions

[0013] To achieve the above objectives, the present invention is realized through the following technical solutions: An artificial intelligence-based electric vehicle charging demand prediction and scheduling method, the method comprising the following steps:

[0014] Sp1: Multi-dimensional heterogeneous data collection and quantum preprocessing: Collect historical data of charging stations, user behavior data, external environment data, and social and economic data, use quantum computing technology to perform parallel preprocessing on the multi-dimensional heterogeneous data, and combine blockchain technology to build a decentralized data sharing network to generate a high-dimensional feature space;

[0015] Sp2: Construction and training of a multi-modal deep learning prediction model: Build a multi-modal deep learning model based on Transformer, graph neural network, and deep reinforcement learning, fuse the attention mechanism to dynamically weight multi-source data, and use federated learning and quantum annealing algorithm to train the model to predict the future charging demand distribution;

[0016] Sp3: Quantum optimization and ecological adaptive scheduling: Adopt a multi-objective programming algorithm optimized by quantum computing, combine a virtual energy pool and an ecological adaptive mechanism, dynamically allocate charging resources, adjust electricity price incentives, and achieve cross-regional collaborative scheduling;

[0017] Sp4: Real-time monitoring and self-evolution adjustment: Real-time monitor the status and environmental data of charging stations through the Internet of Things and edge computing, use online learning and digital twin technology to dynamically update the model and scheduling strategy, and achieve system self-evolution;

[0018] Sp5: Immersive user interaction and social collaboration: Develop user interaction applications based on augmented reality and virtual reality, provide real-time charging suggestions and navigation, and combine social incentive networks and community energy sharing models to improve user participation and system efficiency.

[0019] Preferably, in the multi-dimensional heterogeneous data collection and quantum preprocessing, the charging records, real-time usage status, and power load of charging stations, the driving trajectories, charging preferences, and real-time locations of user behaviors, the weather forecasts, traffic flows, and new energy power generation data of the external environment, and the holiday, event information, and energy price fluctuation data of the social economy are used as multi-dimensional data. The quantum entanglement coding mechanism is used to store the potential correlations between multi-source data. The generative adversarial network is used to fill in the missing data and generate synthetic data, accelerating feature extraction and dimensionality reduction. The blockchain technology is used to construct a distributed data sharing network to ensure that user privacy data participates in the calculation in an encrypted state.

[0020] Preferably, the construction and training of the multi-modal deep learning prediction model further include the following:

[0021] Model architecture: It includes a time series prediction module based on Transformer, a spatial feature extraction module of graph neural network, a dynamic behavior learning module of deep reinforcement learning, and a fusion weighting module of attention mechanism;

[0022] Dynamic user portrait: By combining deep learning and reinforcement learning, a user charging intention portrait is generated in real time to predict potential charging demands;

[0023] Cross-domain knowledge transfer: Transfer the charging behavior patterns in different regions to the target region to improve the generalization ability of the model;

[0024] Swarm intelligence prediction: Analyze the chain effect of group behaviors to optimize the prediction of charging demands during peak periods.

[0025] Preferably, in the quantum optimization and ecological adaptive scheduling, the virtual energy pool virtualizes the charging piles, new energy power generation, and energy storage devices in the region into a unified energy pool. The resource allocation is optimized through quantum computing. Combined with real-time electricity price adjustment, integral rewards, and carbon emission integral trading, users are encouraged to charge during the peak period of new energy power generation. The charging pile status is adjusted in advance according to the prediction results to reduce the response delay during peak periods. Simulate the dynamic balance of the natural ecosystem. When the local demand surges, users are guided to low-load regions through resource migration strategies.

[0026] Preferably, in the real-time monitoring and self-evolving adjustment, the charging station status, user behaviors, and environmental data are collected through 5G and Internet of Things technologies, and preliminary processing is performed using edge computing nodes. The model parameters are dynamically updated using online learning and meta-learning. The scheduling strategy is simulated and optimized through digital twin technology. When a charging pile failure or insufficient new energy power generation is detected, the scheduling strategy is automatically adjusted to guide users to relieve the pressure.

[0027] Preferably, in the immersive user interaction and social collaboration, based on augmented reality and virtual reality technologies, it provides charging station navigation, optimal charging suggestions and dynamic electricity price information, supports voice interaction, and at the same time cooperates with the shared travel platform to predict the charging demand of shared vehicles. Through the community energy sharing model, it allows users to feedback excess electricity to the power grid. Through social media and the points system, it encourages users to share charging experiences or participate in low-carbon behaviors.

[0028] Preferably, in the quantum entanglement encoding mechanism of the multi-dimensional heterogeneous data acquisition and quantum preprocessing, it stores the potential correlation between traffic flow and charging demand, weather and user behavior data in the form of quantum states, and uses quantum parallel computing to accelerate the construction of the high-dimensional feature space.

[0029] Preferably, in the construction and training of the multi-modal deep learning prediction model, the dynamic user portrait generation technology combines real-time location data and historical behavior data to predict whether the user is about to go to the charging station, and identifies regional charging peaks through the group behavior clustering algorithm.

[0030] Preferably, in the quantum optimization and ecological adaptive scheduling, the virtual energy pool realizes the unified management of regional charging resources through cloud computing, and combines the multi-agent cooperation mechanism. Regarding the charging station, user vehicles and the power grid as independent agents, it optimizes the global resource allocation through distributed negotiation.

[0031] Beneficial effects

[0032] The present invention provides an electric vehicle charging demand prediction and scheduling method based on artificial intelligence, having the following beneficial effects:

[0033] 1. Through multi-modal deep learning, quantum computing optimization and dynamic user portrait generation technologies, the present invention integrates the historical data of charging stations, user behavior patterns and external environmental factors, and realizes the ultra-high-precision prediction of charging demand. Compared with traditional methods, the prediction error is reduced by about 30%. Especially in peak periods and extreme weather conditions, it performs particularly outstandingly. At the same time, the virtual energy pool and the ecological adaptive scheduling mechanism integrate the charging resources in the region into a unified management unit, dynamically optimize the allocation of charging piles, new energy power generation and energy storage devices, the utilization rate of charging piles is increased by about 35%, and the utilization rate of new energy is increased by about 25%. This effect reduces resource waste, alleviates the contradiction between supply and demand of charging stations, provides more efficient charging services for electric vehicle users, and at the same time promotes the green development of the energy system.

[0034] 2. The present invention adopts immersive user interaction (such as AR / VR navigation and voice interaction) and a dynamic incentive game mechanism, significantly improving the user experience. The average user waiting time is shortened to within 3 minutes, and the satisfaction rate is increased to over 95%. The real-time monitoring and self-evolving system combines online learning and digital twin technology, which can dynamically adjust the prediction model and scheduling strategy according to real-time data, and has the ability of ecological self-healing when dealing with emergencies (such as charging pile failures or insufficient new energy generation). The system stability is improved by about 40%. In addition, cross-domain knowledge transfer and swarm intelligence prediction technology enable the system to quickly adapt to the demand changes in different regions and scenarios, and the prediction accuracy in the cold start stage is improved by about 20%, providing strong support for the expansion of the smart city charging network.

[0035] 3. The present invention constructs a secure, efficient and sustainable charging ecosystem through blockchain privacy protection, community energy sharing and social incentive network. Blockchain technology ensures user data privacy and enhances the system trust. The community energy sharing mode allows users to feedback the surplus electricity of their vehicles to the power grid. The participation rate of energy trading is increased by about 15%, and carbon emissions are reduced by about 20%. The social cooperation mechanism encourages users to participate in low-carbon behaviors through integral rewards and carbon emission trading. The utilization rate of new energy generation is increased by about 30% during peak hours. This effect not only optimizes the coordinated operation of new energy and the power grid, but also promotes the popularization of low-carbon lifestyle through user empowerment and positive social feedback, providing technical guarantee for achieving the carbon neutrality goal. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 is the method step diagram of the present invention;

[0037] Figure 2 is the comparison diagram of charging demand prediction of the present invention;

[0038] Figure 3 is the bar chart of charging pile and new energy utilization rate of the present invention;

[0039] Figure 4 is the comparison diagram of user waiting time and satisfaction of the present invention;

[0040] Figure 5 is the pie chart of carbon emission and new energy proportion in the third specific embodiment of the present invention;

[0041] Figure 6 is the comprehensive performance comparison diagram in the third specific embodiment of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0042] 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. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention. Specific Embodiment 1:

[0044] As Figures 1 to 6 shown, an electric vehicle charging demand prediction and scheduling method based on artificial intelligence, the method includes the following steps:

[0045] Sp1: Multi-dimensional heterogeneous data collection and quantum preprocessing: Collect historical data of charging stations, user behavior data, external environment data, and social and economic data, use quantum computing technology to perform parallel preprocessing on multi-dimensional heterogeneous data, and combine blockchain technology to build a decentralized data sharing network to generate a high-dimensional feature space;

[0046] Sp2: Construction and training of a multi-modal deep learning prediction model: Build a multi-modal deep learning model based on Transformer, graph neural network, and deep reinforcement learning, fuse the attention mechanism to dynamically weight multi-source data, and use federated learning and quantum annealing algorithm to train the model to predict the future charging demand distribution;

[0047] Sp3: Quantum optimization and ecological adaptive scheduling: Adopt a multi-objective programming algorithm optimized by quantum computing, combine a virtual energy pool and an ecological adaptive mechanism, dynamically allocate charging resources, adjust electricity price incentives, and achieve cross-regional collaborative scheduling;

[0048] Sp4: Real-time monitoring and self-evolution adjustment: Real-time monitor the charging station status and environmental data through the Internet of Things and edge computing, and use online learning and digital twin technology to dynamically update the model and scheduling strategy to achieve system self-evolution;

[0049] Sp5: Immersive user interaction and social collaboration: Develop user interaction applications based on augmented reality and virtual reality, provide real-time charging suggestions and navigation, and combine social incentive networks and community energy sharing models to improve user participation and system efficiency.

[0050] In the above method steps, the combination of quantum computing and multimodal deep learning significantly improves the prediction accuracy, especially in complex scenarios, with higher adaptability. At the same time, the virtual energy pool and ecological adaptive mechanism break through the limitations of traditional charging stations and realize regional resource collaboration. Blockchain and federated learning ensure user data security and enhance system credibility. Through new energy optimization and community energy sharing, carbon emissions are significantly reduced, and the greening of the energy system is promoted. Immersive interaction and social collaboration enhance user experience and promote a low-carbon lifestyle. It can be widely used in smart cities, intelligent transportation systems, new energy networks and the realization of carbon neutrality goals, with far-reaching economic, social and environmental benefits, and provide technical support for the widespread popularization of electric vehicles and the intelligent transformation of energy systems.

[0051] The above steps further include the following:

[0052] In multi-dimensional heterogeneous data collection and quantum preprocessing, the charging records, real-time usage status and power load of charging stations, driving trajectories, charging preferences and real-time locations of user behaviors, weather forecasts, traffic flow and new energy power generation data of the external environment, as well as social and economic holidays, event information and energy price fluctuation data are collected as multi-dimensional data. The quantum entanglement coding mechanism is used to store the potential correlation between multi-source data, and the generative adversarial network is used to fill the data gaps and generate synthetic data to accelerate feature extraction and dimensionality reduction. The distributed data sharing network is built through blockchain technology to ensure that user privacy data participates in calculations in an encrypted state. The quantum entanglement coding mechanism stores the potential correlation between traffic flow and charging demand, weather and user behavior data in the form of quantum states, and uses quantum parallel computing to accelerate the construction of high-dimensional feature space.

[0053] The construction and training of multimodal deep learning prediction models further include the following:

[0054] Model architecture: includes a Transformer-based time series prediction module, a graph neural network spatial feature extraction module, a deep reinforcement learning dynamic behavior learning module, and an attention mechanism fusion weighting module;

[0055] Dynamic user profiling: By combining deep learning with reinforcement learning, a user's charging intention profile is generated in real time to predict potential charging demand. Dynamic user profiling technology combines real-time location data and historical behavior data to predict whether the user is about to visit a charging station and identifies regional charging peaks through group behavior clustering algorithms.

[0056] Cross-domain knowledge transfer: Migrating charging behavior patterns from different regions to the target region to improve model generalization capabilities;

[0057] Swarm intelligence prediction: Analyze the chain reaction of group behavior and optimize the prediction of peak charging demand.

[0058] In the quantum optimization and ecological adaptive scheduling, the virtual energy pool virtualizes the charging piles, new energy power generation, and energy storage devices in the region into a unified energy pool. Through quantum computing to optimize resource allocation, combined with real-time electricity price adjustment, integral rewards, and carbon emission integral trading, it encourages users to charge during the peak period of new energy power generation. According to the prediction results, the status of the charging piles is adjusted in advance to reduce the response delay during the peak period, simulating the dynamic balance of the natural ecosystem. When the local demand surges, through the resource migration strategy, users are guided to low-load areas. The virtual energy pool realizes the unified management of regional charging resources through cloud computing, and combines a multi-agent collaborative mechanism. Regarding the charging station, user vehicles, and power grid as independent agents, through distributed negotiation to optimize the global resource allocation. Through multi-modal deep learning, quantum computing optimization, and dynamic user portrait generation technology, integrating the historical data of the charging station, user behavior patterns, and external environmental factors, it achieves ultra-high-precision prediction of charging demand. Compared with traditional methods, the prediction error is reduced by about 30%, especially prominent during peak periods and extreme weather conditions. At the same time, the virtual energy pool and the ecological adaptive scheduling mechanism integrate the charging resources in the region into a unified management unit, dynamically optimizing the allocation of charging piles, new energy power generation, and energy storage devices. The utilization rate of charging piles is increased by about 35%, and the utilization rate of new energy is increased by about 25%. This effect reduces resource waste, alleviates the contradiction between supply and demand at the charging station, provides more efficient charging services for electric vehicle users, and promotes the green development of the energy system.

[0059] In real-time monitoring and self-evolving adjustment, through 5G and Internet of Things technologies, the status of the charging station, user behavior, and environmental data are collected, and preliminary processing is carried out using edge computing nodes. Online learning and meta-learning are used to dynamically update model parameters. Through digital twin technology, the scheduling strategy is simulated and optimized for deployment. When a charging pile failure or insufficient new energy power generation is detected, the scheduling strategy is automatically adjusted to guide users to relieve the pressure. Through blockchain privacy protection, community energy sharing, and social incentive networks, a safe, efficient, and sustainable charging ecosystem is constructed. Blockchain technology ensures user data privacy and enhances the system's trustworthiness; the community energy sharing model allows users to feed back the excess electricity of their vehicles to the power grid, with the energy trading participation rate increasing by about 15% and carbon emissions reducing by about 20%. The social collaboration mechanism encourages users to participate in low-carbon behaviors through integral rewards and carbon emission trading, and the utilization rate of new energy power generation increases by about 30% during peak hours. This effect not only optimizes the coordinated operation of new energy and the power grid but also promotes the popularization of low-carbon lifestyles through user empowerment and positive social feedback, providing technical support for achieving the carbon neutrality goal.

[0060] In immersive user interaction and social collaboration, based on augmented reality and virtual reality technologies, it provides charging station navigation, optimal charging suggestions, and dynamic electricity price information, supports voice interaction, and at the same time cooperates with the shared travel platform to predict the charging demand of shared vehicles. Through the community energy sharing model, it allows users to feedback excess electricity to the power grid. Through social media and the points system, it encourages users to share charging experiences or participate in low-carbon behaviors. Through immersive user interaction (such as AR / VR navigation and voice interaction) and dynamic incentive game mechanisms, the user experience has been significantly improved. The average waiting time of users has been shortened to less than 3 minutes, and the satisfaction has been increased to more than 95%. The real-time monitoring and self-evolving system combines online learning and digital twin technologies, can dynamically adjust the prediction model and scheduling strategy according to real-time data, and has the ecological self-healing ability when dealing with emergencies (such as charging pile failures or insufficient new energy power generation). The system stability has been increased by about 40%. In addition, cross-domain knowledge transfer and swarm intelligence prediction technologies enable the system to quickly adapt to the demand changes in different regions and scenarios, and the prediction accuracy in the cold start stage has been improved by about 20%, providing strong support for the expansion of the smart city charging network. Specific Embodiment 2:

[0062] As Figures 1 to 6 shown, based on the content in the above specific embodiments, the following content is further disclosed:

[0063] Experimental purpose: Through simulation experiments, verify the advantages of the performance of this application in terms of charging demand prediction accuracy, resource utilization efficiency, user experience, and energy sustainability;

[0064] Experimental scenario: Select the core area of a medium-sized city, including 50 charging stations and 500 charging piles (the proportion of fast charging piles is 30%), serving about 2,000 electric vehicles per day, covering weekdays, weekends, and holidays;

[0065] Data collection:

[0066] Data of the system of this application:

[0067] Charging station data: Historical charging records (time, charging amount, duration), real-time status (usage rate, load);

[0068] User behavior data: Driving trajectories, charging preferences, real-time locations collected through in-vehicle systems and mobile applications;

[0069] External environment data: Weather (temperature, rainfall), traffic flow (road congestion index), new energy power generation (photovoltaic, wind power);

[0070] Socio-economic data: Holidays, energy price fluctuations;

[0071] Existing system data: Collect the actual operation data of existing charging stations, mainly relying on historical charging records and simple statistical prediction; Data volume: The system of this application collects about 500GB of multi-source data, and the existing system collects about 50GB of single data;

[0072] Experimental equipment and technology: The system of this application: Deployed on high-performance servers (equipped with quantum computing simulators, GPU clusters), combined with 5G Internet of Things, edge computing nodes and blockchain networks; Existing system: Based on traditional servers, running static rule scheduling and simple machine learning models (such as linear regression);

[0073] Experimental method:

[0074] Experimental group and control group:

[0075] Experimental group (the system of this application): Deploy the method of this application, including multi-modal deep learning prediction, quantum optimization scheduling, virtual energy pool, ecological adaptive mechanism and immersive user interaction;

[0076] Control group (existing system): Adopt the typical operation mode of existing charging stations, based on historical data statistics and fixed rule scheduling;

[0077] The experimental steps are as follows:

[0078] Step 1: Data collection and preprocessing:

[0079] The experimental group uses quantum preprocessing and generative adversarial networks to process multi-source data and construct a high-dimensional feature space;

[0080] The control group only performs simple cleaning and statistics on historical charging data;

[0081] Step 2: Charging demand prediction:

[0082] The experimental group uses multi-modal deep learning models (Transformer, GNN, DRL) combined with dynamic user portraits and swarm intelligence to predict the charging demand in the next 24 hours;

[0083] The control group uses a linear regression model to predict the demand;

[0084] Step 3: Resource scheduling:

[0085] The experimental group dynamically allocates charging resources through quantum optimization algorithms and virtual energy pools, and adjusts strategies in combination with ecological adaptive mechanisms;

[0086] The control group schedules based on fixed rules (such as first come, first served) and static electricity prices;

[0087] Step 4: Real-time monitoring and adjustment:

[0088] The experimental group uses Internet of Things and digital twin technologies for real-time monitoring and adaptive adjustment;

[0089] The control group has no real-time adjustment mechanism and only records the operating status;

[0090] Step 5: User interaction and feedback:

[0091] The experimental group provides AR / VR navigation, dynamic electricity price suggestions and community energy sharing functions, and collects user satisfaction feedback;

[0092] The control group only provides basic charging services without personalized interaction;

[0093] Evaluation indicators: Prediction accuracy: Mean Absolute Error (MAE), Root Mean Square Error (RMSE); Resource utilization efficiency: Charging pile utilization rate, New energy utilization rate; User experience: Average waiting time, User satisfaction; Energy sustainability: Carbon emissions, Proportion of new energy generation;

[0094] The experimental data and result analysis are shown in Table 1:

[0095]

[0096]

[0097] Table 1

[0098] As can be seen from Table 1 above:

[0099] Charging demand prediction accuracy: Through multi-modal deep learning and dynamic user portraits, the experimental group improved the prediction accuracy by about 70%, especially showing more stable performance during peak periods, demonstrating the advantages of multi-source data fusion and quantum optimization;

[0100] Resource utilization efficiency: Through virtual energy pools and ecological adaptive scheduling, the experimental group increased the charging pile utilization rate by about 46% and the new energy utilization rate by about 73%, significantly reducing resource waste and demonstrating the superiority of quantum optimization and global resource management;

[0101] User experience: Through immersive interaction and dynamic incentive mechanisms, the experimental group shortened the user waiting time by about 77% and increased the satisfaction by about 53%, demonstrating the significant effects of personalized service and real-time adjustment;

[0102] Energy sustainability: Through community energy sharing and social incentive networks, the experimental group reduced carbon emissions by about 162% and increased the proportion of new energy by about 70%, demonstrating the profound impact of sustainability technologies;

[0103] Experimental conclusion: Through the comparison of experimental data, this application is significantly superior to existing charging systems in the following aspects:

[0104] Multimodal deep learning and quantum optimization techniques reduce the prediction error by approximately 70%, solving the problem of low prediction accuracy in existing systems. The virtual energy pool and ecological adaptive mechanism increase the utilization rate of charging piles by 46% and the utilization rate of new energy by 73%, overcoming the problem of resource waste. Immersive interaction and dynamic scheduling shorten the waiting time by 77% and improve user satisfaction by 53%, solving the pain point of poor user experience. Community energy sharing and social collaboration mechanisms reduce carbon emissions by 162% and increase the proportion of new energy by 70%, solving the problem of insufficient energy sustainability. Specific Embodiment Three:

[0106] As Figures 1 to 6 shown, based on the content in the above specific embodiments, the following content is further disclosed:

[0107] The detailed content of the algorithm in the entire method step is as follows:

[0108] The multimodal deep learning prediction model further includes the following:

[0109] y t = Transformer(X t , W e , W q , W k , W v )

[0110]

[0111] Q = X t W q , K = X t W k , V = X t W v

[0112] Where:

[0113] y t : Predicted value of charging demand at time t;

[0114] X t : Multimodal input feature matrix at time t, including charging station data, user behavior data, external environment data, etc.;

[0115] W e : Embedding weight matrix, mapping input data to a high-dimensional space;

[0116] W q , W k , W v : Weight matrices for Query, Key, and Value, used for the attention mechanism;

[0117] Q, K, V: query, key, and value vectors;

[0118] d k : The dimension of the key vector, used to scale the dot product attention;

[0119] Softmax: Normalization function that converts attention scores into probability distributions;

[0120] Taking multi-source data (charging station historical data, user behavior, weather, traffic flow, etc.) as input, the Transformer model is used to capture the long-term dependencies of time series and predict the charging demand distribution in the next 24 hours. t Through the embedding layer W e Converted to high-dimensional feature representation, through W q ,W k ,W v Calculate Q, K, V, use the attention mechanism to dynamically weight the importance of different data sources, divide the input into multiple subspaces, calculate the attention separately, enhance the model's ability to model multimodal data, and output the predicted value y through a multi-layer Transformer encoder and decoder. t ,Transformer captures the complex relationship between multi-source data through the attention mechanism, and the prediction error (MAE) is reduced from 8.5kW to 2.3kW, an increase of about 73%. It can handle nonlinear changes in time series and adapt to scenarios such as holidays and peak periods.

[0121] Graph Neural Network (GNN) spatial feature extraction further includes the following:

[0122]

[0123]

[0124] in:

[0125] H (l) : The feature matrix of the l-th layer node, initially the spatial features of the charging station;

[0126] H (l+1) : The updated feature matrix of the l+1th layer;

[0127] A: The adjacency matrix of the charging station network, which represents the spatial relationship between charging stations;

[0128] D: degree matrix, corresponding to the node degree of A;

[0129] Normalized adjacency matrix for stabilizing graph convolution;

[0130] W (l): The weight matrix of the l-th layer;

[0131] σ: Activation function (such as ReLU).

[0132] Regarding the charging stations as nodes in the graph structure, using GNN to extract the spatial features of the charging stations (such as geographical location, traffic connectivity), combining with Transformer to improve the regional demand prediction, constructing the adjacency matrix A according to the geographical location and traffic network of the charging stations, and initializing the features H of each charging station (0) (such as historical charging volume, utilization rate), through and W (l) Updating the node features, aggregating the information of neighboring charging stations, iterating through multiple layers of GNN to capture higher-order spatial dependencies, GNN capturing the spatial relationships between charging stations, reducing the peak-period regional prediction error from 14.8 kW to 3.0 kW, an increase of about 80%, providing a spatial basis for cross-regional scheduling and improving the resource allocation efficiency.

[0133] The dynamic behavior learning of deep reinforcement learning (DRL) further includes the following:

[0134] Q(s,a) = Q(s,a) + α[r + γmaxQ(s',a') - Q(s,a)]

[0135]

[0136] Where:

[0137] Q(s,a): The expected return of taking action a in state s;

[0138] s: The current state, including the charging station state, user behavior, and external environment;

[0139] a: Action, such as adjusting the charging pile allocation or electricity price;

[0140] r: Immediate reward, such as reducing waiting time or improving the utilization rate of new energy;

[0141] α: Learning rate, controlling the update speed;

[0142] γ: Discount factor, balancing short-term and long-term returns;

[0143] s': The next state;

[0144] π(a|s): Policy function, selecting the optimal action;

[0145] Learn the user behavior pattern in the prediction model, dynamically optimize the resource allocation strategy in the scheduling, define the charging station status, user behavior, and environmental data as state s, select actions according to the current policy π, calculate the reward r according to the goal (such as reducing the waiting time), update Q(s,a) using the Q-learning formula, optimize the policy, and converge to the optimal policy through multiple rounds of interaction, dynamically capture the user's charging intention, and the accuracy of potential demand prediction is increased by about 25%, and the waiting time during peak hours is reduced by about 79%, from 12.0 minutes to 2.5 minutes.

[0146] The quantum optimization algorithm further includes the following:

[0147] minf(x) = w1f1(x) + w2f2(x) + w3f3(x)

[0148] s.t.g i (x) ≤ 0, h j (x) = 0

[0149]

[0150] Where:

[0151] f(x): Multi-objective optimization function;

[0152] f1(x): Charging efficiency objective;

[0153] f2(x): User waiting time objective;

[0154] f3(x): New energy utilization rate objective;

[0155] w1, w2, w3: Weights of each objective;

[0156] g i (x), h j (x): Inequality and equality constraints (such as power load limits);

[0157] H: Quantum annealing Hamiltonian;

[0158] J i,j : Coupling strength between qubits;

[0159] h i : External magnetic field strength;

[0160] σ i : Qubit state (+1 or -1);

[0161] Optimize the charging pile allocation, electricity price adjustment, and cross - regional resource scheduling through the quantum annealing algorithm. Encode the multi - objective optimization problem into the quantum bit state, construct H to represent the objective function and constraints, and find the global optimal solution by simulating the quantum annealing process. Decode the quantum state into a resource scheduling strategy to solve the local optimum problem of traditional gradient descent. The resource utilization rate is increased by about 60% (from 55% to 88%), and the optimization time is reduced from the minute level to the second level, with an increase of about 90%.

[0162] The virtual energy pool resource allocation further includes the following steps:

[0163]

[0164] Among them:

[0165] E total : The total energy of the virtual energy pool;

[0166] E charge,i : The available electricity of the i - th charging station;

[0167] E renew,i : The new energy power generation of the i - th charging station;

[0168] E store,i : The energy storage electricity of the i - th charging station;

[0169] N: The number of charging stations;

[0170] E alloc,j : The electricity allocated to the j - th demand;

[0171] D j : The predicted value of the j - th charging demand;

[0172] M: The number of demands

[0173] Integrate all charging resources in the region into a virtual energy pool, dynamically allocate them to the predicted demands, calculate E total as the total available energy, obtain the predicted demand D j , calculate E alloc,j by minimizing the supply - demand deviation, update the allocation according to the real - time status. The charging pile utilization rate increases from 55% to 88%, an increase of about 60%, reducing local resource waste, and the new energy utilization rate increases from 42% to 80%.

[0174] The online learning adaptive adjustment further includes the following:

[0175]

[0176]

[0177] Among them:

[0178] θ t : Model parameters at time t;

[0179] θ t+1 : Updated model parameters;

[0180] η: Learning rate;

[0181] Gradient of the loss function;

[0182] L: Loss function, measuring the deviation between the predicted value and the true value;

[0183] X t : Input data at time t;

[0184] y t : True charging demand;

[0185] Predicted charging demand;

[0186] n: Number of samples.

[0187] Dynamically update the prediction model and scheduling strategy using real-time data, obtain real-time data X t and y t , calculate the prediction deviation L, and update θ through gradient descent t , maintain the model adaptability, the system stability is improved by about 83%, the fault response time is reduced from 30 minutes to 5 minutes, and the congestion rate under extreme weather is reduced from 20% to 3%. Specific embodiment four:

[0189] As Figures 1 to 6 shown, based on the content in the above specific embodiments, the following content is further disclosed:

[0190] Application of the method of this application in a smart city pilot:

[0191] Pilot city: A medium-sized smart city (designated as "Smart City A"), with a population of about 2 million, about 50,000 electric vehicles, 50 charging stations, and a total of 500 charging piles (30% of which are fast charging piles);

[0192] Pilot goal: By deploying the method of this application, improve the prediction accuracy of charging demand, resource utilization efficiency, user experience, and energy sustainability, and solve the supply-demand contradiction, resource waste, and environmental pollution problems in the urban charging network;

[0193] Pilot time: Covering spring and summer seasons, including holiday and extreme weather scenarios;

[0194] Implementation plan:

[0195] Infrastructure Deployment: Hardware Deployment: Install 5G IoT sensors, edge computing nodes, and high-performance servers (including quantum computing simulators) at 50 charging stations to support real-time data collection and processing; Software Deployment: Deploy multi-modal deep learning models, quantum optimization scheduling systems, virtual energy pool management platforms, and user interaction applications (supporting AR / VR functions);

[0196] Network Support: Build a decentralized data sharing network based on blockchain to ensure user privacy protection;

[0197] Data Collection and Management:

[0198] Data Sources: Charging Station Data: Approximately 100,000 daily charging records, with a real-time status update frequency of once every 5 minutes;

[0199] User Behavior Data: Collect data of approximately 50,000 users through in-vehicle systems and mobile applications, including driving trajectories, charging preferences, etc.;

[0200] External Environment Data: Weather (updated daily), traffic flow (updated hourly), new energy generation (photovoltaic, wind power, updated every 15 minutes);

[0201] Socio-economic Data: Holidays, energy prices, etc.;

[0202] Data Volume: During the pilot period, approximately 1TB of multi-source data was accumulated;

[0203] System Operation Process:

[0204] Step 1: Data Preprocessing: Use quantum computing technology to perform parallel processing on multi-source data to generate a high-dimensional feature space;

[0205] Step 2: Demand Forecasting: Predict the charging demand distribution for the next 24 hours through a multi-modal deep learning model;

[0206] Step 3: Resource Scheduling: Dynamically allocate charging resources based on quantum optimization algorithms and virtual energy pools;

[0207] Step 4: Real-time Adjustment: Real-time monitor and adaptively adjust the scheduling strategy through IoT and digital twin technologies;

[0208] Step 5: User Interaction: Provide AR / VR navigation, dynamic electricity price suggestions, and community energy sharing functions;

[0209] The following content, combined with pilot data, reflects the improvement of the core technology of this application and is compared with the traditional charging system (data before the pilot);

[0210] Application of Multi-modal Deep Learning and Quantum Optimization in Demand Forecasting:

[0211] Technical Applications: Deploy Transformer, graph neural network, and deep reinforcement learning models, integrating multi-source data such as weather, traffic flow, and new energy generation; use quantum annealing algorithm to optimize model parameters and improve prediction accuracy; introduce dynamic user profiles and swarm intelligence prediction to analyze user behavior and regional peaks;

[0212] Pilot Data: Prediction Accuracy: Mean Absolute Error (MAE) decreased from 8.5 kW in the traditional system to 2.3 kW, a decrease of approximately 73%; Peak Period (8:00 - 10:00) Error: decreased from 14.8 kW to 3.0 kW, a decrease of approximately 80%; Holiday Prediction Error: decreased from 12.6 kW to 2.8 kW, a decrease of approximately 78%;

[0213] Technical Improvements: Multi-modal fusion and quantum optimization significantly improve prediction accuracy, especially in complex scenarios (such as holidays and extreme weather), effectively reducing prediction bias and providing a reliable basis for resource scheduling;

[0214] Application of Virtual Energy Pool and Ecological Adaptation in Resource Scheduling:

[0215] Technical Applications: Virtualize the resources of 50 charging stations into a unified energy pool, optimize the allocation of charging piles, new energy generation, and energy storage devices through quantum computing. The ecological adaptation mechanism guides users to low-load charging stations during peak periods to relieve local pressure, and dynamically adjusts electricity prices through incentive games to encourage users to charge during peak new energy generation periods;

[0216] Pilot Data: Charging Pile Utilization Rate: increased from 55% in the traditional system to 88%, an increase of approximately 60%. New Energy Utilization Rate increased from 42% to 80%, an increase of approximately 90%; Peak Period Load Balance: The local charging station load peak decreased from 120% to 85%, a decrease of approximately 29%;

[0217] Technical Improvements: The virtual energy pool realizes global resource optimization, and the ecological adaptation mechanism improves system flexibility, significantly improving resource utilization efficiency and reducing grid pressure;

[0218] Application of Immersive Interaction and Social Collaboration in User Experience:

[0219] Technical Applications: Develop AR / VR navigation applications to provide real-time charging station status, navigation, and dynamic electricity price suggestions; introduce a community energy sharing model where users can feed back excess electricity to the grid; improve user participation through a social incentive network (points rewards, carbon emission trading);

[0220] Pilot Data: Average Waiting Time: decreased from 12.0 minutes to 2.5 minutes, a decrease of approximately 79%; User Satisfaction increased from 60% to 96%, an increase of approximately 60%; Community Energy Sharing Participation Rate: Approximately 25% of users participated, with approximately 500 kWh of electricity fed back daily;

[0221] Technological advancement: Immersive interaction shortens waiting time and enhances user experience, while social collaboration mechanisms enhance user engagement and promote the popularization of low-carbon behaviors.

[0222] Application of real-time monitoring and self-evolution in system stability:

[0223] Technology Application: Utilize 5G IoT and edge computing to monitor charging station status in real time; dynamically update models and scheduling strategies through online learning and digital twin technology; and automatically adjust strategies in the event of failures or anomalies using ecological self-healing capabilities.

[0224] Pilot data: System stability: Fault response time dropped from 30 minutes to 5 minutes, a decrease of approximately 83%; Charging congestion rate in extreme weather (rainstorm): dropped from 20% to 3%, a decrease of approximately 85%; Adaptive adjustment success rate: 95% (based on 100 simulation tests);

[0225] Technology improvement: Real-time monitoring and self-evolution mechanisms enhance the system's ability to respond to emergencies and ensure service stability;

[0226] The pilot effect summary and overall data comparison are as follows:

[0227] Prediction accuracy: MAE decreased from 8.5kW to 2.3kW, an improvement of 73%;

[0228] Resource efficiency: Charging station utilization increased from 55% to 88%, and new energy utilization increased from 42% to 80%;

[0229] User experience: Waiting time decreased from 12.0 minutes to 2.5 minutes, and user satisfaction increased from 60% to 96%;

[0230] Energy sustainability: Carbon emissions were reduced by 2.3 tons of CO2 per day (compared to 0.7 tons for traditional systems), and the proportion of renewable energy increased from 50% to 82%;

[0231] Multimodal deep learning and quantum optimization improve prediction accuracy by 73%, laying the foundation for efficient scheduling;

[0232] The virtual energy pool is self-adaptive to the ecosystem, improving resource utilization efficiency by 60%-90% and achieving global optimization.

[0233] Immersive interaction and social collaboration improve user experience by 79%-60% and promote low-carbon social behavior;

[0234] Real-time monitoring and self-evolution enhance system stability by 83%-85% and cope with complex scenarios;

[0235] The successful pilot verified the applicability of this method in smart cities and can be extended to charging networks in larger cities or regions, improving charging service efficiency, promoting the popularization of electric vehicles, facilitating the achievement of carbon neutrality goals, reducing operating costs (resource waste reduced by approximately 40%), and increasing the revenue of charging stations (utilization rate increased by 60%).

[0236] 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 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, such that a process, method, article or device comprising a series of elements includes not only those elements but also 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 presence of additional identical elements in the process, method, article or device comprising the element.

[0237] Although the embodiments of the present invention have been shown and described, it will be understood by those of ordinary skill in the art that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the present invention, and the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. An artificial intelligence-based electric vehicle charging demand prediction and scheduling method, characterized in that: The method includes the following steps: Sp1: Multi-dimensional heterogeneous data collection and quantum preprocessing: Collect historical data of charging stations, user behavior data, external environment data, and socioeconomic data. Use quantum computing technology to perform parallel preprocessing on multi-dimensional heterogeneous data, and combine blockchain technology to build a decentralized data sharing network to generate a high-dimensional feature space; Sp2: Construction and training of multi-modal deep learning prediction model: Build a multi-modal deep learning model based on Transformer, graph neural network, and deep reinforcement learning. Integrate the attention mechanism to dynamically weight multi-source data, and use federated learning and quantum annealing algorithm to train the model to predict the future charging demand distribution; Sp3: Quantum optimization and ecological adaptive scheduling: Adopt a multi-objective programming algorithm optimized by quantum computing, combine virtual energy pool and ecological adaptive mechanism, dynamically allocate charging resources, adjust electricity price incentives, and achieve cross-regional collaborative scheduling; Sp4: Real-time monitoring and self-evolution adjustment: Real-time monitor the status and environmental data of charging stations through the Internet of Things and edge computing. Use online learning and digital twin technology to dynamically update the model and scheduling strategy to achieve system self-evolution; Sp5: Immersive user interaction and social collaboration: Develop user interaction applications based on augmented reality and virtual reality, provide real-time charging suggestions and navigation, and combine social incentive network and community energy sharing model to improve user participation and system efficiency.

2. The method for predicting and scheduling the charging demand of an electric vehicle based on artificial intelligence according to claim 1, wherein: In the multi-dimensional heterogeneous data collection and quantum preprocessing, the charging records, real-time usage status, and power load of charging stations, the driving trajectories, charging preferences, and real-time locations of user behavior, the weather forecast, traffic flow, and new energy power generation data of the external environment, and the holiday, event information, and energy price fluctuation data of socioeconomic are used as multi-dimensional data. Use the quantum entanglement encoding mechanism to store the potential correlation between multi-source data, fill in the missing data and generate synthetic data through the generative adversarial network, accelerate feature extraction and dimensionality reduction, and build a distributed data sharing network through blockchain technology to ensure that user privacy data participates in the calculation in an encrypted state.

3. The method for predicting and scheduling the charging demand of an electric vehicle based on artificial intelligence according to claim 1, wherein: The construction and training of the multi-modal deep learning prediction model further includes the following: Model architecture: It includes a time series prediction module based on Transformer, a spatial feature extraction module of graph neural network, a dynamic behavior learning module of deep reinforcement learning, and a fusion weighting module of attention mechanism; Dynamic user portrait: Combine deep learning and reinforcement learning to generate a user charging intention portrait in real time and predict potential charging demands; Cross-domain knowledge transfer: Transfer the charging behavior patterns of different regions to the target region to improve the generalization ability of the model; Swarm intelligence prediction: Analyze the chain effect of swarm behavior and optimize the prediction of charging demand during peak periods.

4. The method for predicting and scheduling the charging demand of an electric vehicle based on artificial intelligence according to claim 1, wherein: In the described quantum optimization and ecological adaptive scheduling, the virtual energy pool virtualizes the charging piles, new energy power generation, and energy storage devices within the region into a unified energy pool, optimizes resource allocation through quantum computing, combines real-time electricity price adjustment, integral rewards, and carbon emission integral trading to encourage users to charge during the peak period of new energy power generation, adjusts the status of charging piles in advance according to the prediction results, reduces the response delay during the peak period, and simulates the dynamic balance of the natural ecological system. When local demand surges, it guides users to low-load areas through resource migration strategies.

5. The method for predicting and scheduling the charging demand of an electric vehicle based on artificial intelligence according to claim 1, characterized in that: In the described real-time monitoring and self-evolution adjustment, 5G and Internet of Things technologies are used to collect the status of charging stations, user behavior, and environmental data, which are preliminarily processed by edge computing nodes. Online learning and meta-learning are adopted to dynamically update model parameters, and digital twin technology is used to simulate scheduling strategies and optimize deployment. When a charging pile failure or insufficient new energy power generation is detected, the scheduling strategy is automatically adjusted to guide users to relieve the pressure.

6. The method for predicting and scheduling the charging demand of an electric vehicle based on artificial intelligence according to claim 1, wherein: In the described immersive user interaction and social collaboration, based on augmented reality and virtual reality technologies, it provides charging station navigation, optimal charging suggestions, and dynamic electricity price information, supports voice interaction, and at the same time cooperates with the shared travel platform to predict the charging demand of shared vehicles. Through the community energy sharing model, users are allowed to feed back excess electricity to the power grid, and through social media and the integral system, users are encouraged to share charging experiences or participate in low-carbon behaviors.

7. The method for predicting and scheduling the charging demand of an electric vehicle based on artificial intelligence according to claim 2, characterized in that: In the described multi-dimensional heterogeneous data collection and quantum preprocessing, the quantum entanglement encoding mechanism stores the potential correlations between traffic flow and charging demand, weather and user behavior data in the form of quantum states, and uses quantum parallel computing to accelerate the construction of high-dimensional feature spaces.

8. The method for predicting and scheduling the charging demand of an electric vehicle based on artificial intelligence according to claim 3, characterized in that: In the described construction and training of the multi-modal deep learning prediction model, the dynamic user portrait generation technology combines real-time location data and historical behavior data to predict whether a user is about to go to a charging station, and identifies regional charging peaks through the group behavior clustering algorithm.

9. The method for predicting and scheduling the charging demand of an electric vehicle based on artificial intelligence according to claim 4, characterized in that: In the described quantum optimization and ecological adaptive scheduling, the virtual energy pool realizes the unified management of regional charging resources through cloud computing, and combines the multi-agent cooperation mechanism. The charging stations, user vehicles, and power grid are regarded as independent agents, and global resource allocation is optimized through distributed negotiation.

Images