Intelligent parking space management method based on multi-source data fusion and related equipment

Abstract

The invention relates to the technical field of vehicles, in particular to an intelligent parking space management method based on multi-source data fusion and related equipment, and the method comprises the steps: collecting multi-source data, fusing the multi-source data into a feature vector, inputting the feature vector into a parking space prediction model obtained through training, and outputting a predicted parking space occupancy rate vector; based on the parking space occupancy rate vector and a preset multi-target reward function, dynamically adjusting a parking rate through a deep reinforcement learning method; the multi-target reward function is established based on a difference value of parking lot revenue, a change value of a parking space utilization rate and user complaint times; the parking space prediction information and the dynamic pricing information are displayed through a user end interaction interface, and adjusted system parameters are obtained through a management end interaction interface; according to the invention, multi-objective optimization balance of parking lot resource configuration efficiency, operation income and user satisfaction can be achieved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of vehicles, in particular to a smart parking space management method based on multi-source data fusion and related equipment. BACKGROUND

[0002] The parking lot management mode in the related art exposes multi-dimensional structural contradictions. On the level of parking space utilization, due to the limitation of static planning thinking, most parking lots have the contradiction that "one spot is difficult to find" in peak hours and "high vacancy rate" in off-peak hours coexist. In terms of pricing mechanism, the rigid time division pricing strategy cannot match the real-time supply and demand relationship, which cannot maximize the income in the demand peak period, nor can it adjust the idle traffic through the price lever. In the resource allocation link, the lack of intelligent scheduling system leads to resource mismatch between parking lots, forming a vicious cycle of local congestion and resource waste.

[0003] The parking space occupation prediction technology in the related art mainly relies on historical parking data to establish a time series model. This kind of method does not include real-time traffic flow, sudden weather changes, large-scale activities, and other external variables in the analysis framework, resulting in a sharp drop in prediction accuracy in complex scenarios. In the dynamic pricing field, most systems still use fixed time period rate adjustment strategies, which are difficult to cope with sudden traffic control, holiday passenger flow fluctuations and other irregular events, causing the price to deviate seriously from the actual supply and demand curve. SUMMARY

[0004] Therefore, the purpose of the embodiments of the present application is to provide a smart parking space management method based on multi-source data fusion and related equipment to solve one or more technical problems existing in the prior art and provide at least one beneficial option or create conditions.

[0005] In one aspect, the embodiments of the present application provide a smart parking space management method based on multi-source data fusion, which comprises the following steps: Collecting multi-source data, the multi-source data including parking lot internal parking space state information, parking lot surrounding traffic flow data, weather data and date type information; Fusing the multi-source data into a feature vector, inputting the feature vector into a trained parking space prediction model, and outputting a predicted parking space occupancy rate vector; wherein the parking space occupancy rate vector includes the occupancy rate prediction value of each parking space in each time interval in the future time; the parking space prediction model is trained based on a space-time graph neural network, and the parking space prediction model includes an input layer, a spatial graph convolution layer, a time feature extraction layer, a space-time feature fusion layer and an output layer; Based on the parking space occupancy rate vector and a preset multi-objective reward function, a dynamic parking fee rate is adjusted through a deep reinforcement learning method; the multi-objective reward function is established based on a difference value of parking lot revenue, a change value of parking space utilization rate and a number of user complaints; The parking space prediction information and the dynamic pricing information are displayed through a user terminal interactive interface, and the adjusted system parameters are obtained through a management terminal interactive interface.

[0006] Optionally, the parking space prediction model is obtained through training in the following manner: Historical multi-source data is collected, and the historical multi-source data is arranged in chronological order, and the data set is divided according to a set time unit; A spatio-temporal graph neural network is established, and the data set is input into the spatio-temporal graph neural network for iterative training; the spatio-temporal graph neural network includes an input layer, a spatial graph convolution layer, a time feature extraction layer, a spatio-temporal feature fusion layer and an output layer; The input layer receives a feature vector obtained by fusing the historical multi-source data, the spatial graph convolution layer processes the feature vector, and the spatial features between the parking spaces are extracted through graph convolution operation; the time feature extraction layer processes the feature vector in the time dimension by using a long short-term memory network, and extracts the time features of the parking demand in different time periods; The spatio-temporal feature fusion layer fuses the spatial features and the time features to obtain a comprehensive feature vector through an attention mechanism; The comprehensive feature vector and a weight matrix of the output layer are multiplied, and then a bias vector is added, and then an activation process is performed, and a predicted parking space occupancy rate vector is output; the parking space occupancy rate vector includes a predicted occupancy rate value of each parking space in each time interval in the future time; A loss value of the spatio-temporal graph neural network is obtained based on a real parking space occupancy rate and a predicted parking space occupancy rate, and when it is determined that the loss value is lower than a loss threshold value, the iterative training of the spatio-temporal graph neural network is stopped, and a parking space prediction model is obtained.

[0007] Optionally, the spatio-temporal feature fusion layer fuses the spatial features and the time features to obtain a comprehensive feature vector through an attention mechanism, including: An adaptive spatio-temporal weight mechanism dynamically adjusts the weights of the spatial graph convolution layer and the time sequence processing layer in the spatio-temporal features through learnable parameters, and outputs a fusion feature vector; A multi-scale modeling method is used to perform graph convolution operation on the fusion feature vector in different scales, and then a splicing operation is performed, and an output of multi-scale spatio-temporal modeling is obtained; The output of multi-scale spatio-temporal modeling and the feature vector are added to obtain a comprehensive feature vector.

[0008] Optionally, the dynamic adjustment of the parking fee rate based on the parking space occupancy rate vector and the preset multi-objective reward function comprises: defining a state space, combining the parking space occupancy rate vector, traffic flow data around the parking lot, time characteristics, and date type information into a state vector; defining an action space, the action space comprising a plurality of parking fee rate adjustment strategies; taking the state vector as an input of a deep reinforcement learning model, calculating a Q value corresponding to each action, the Q value representing a long-term cumulative reward expected to be obtained after performing the action in the state; selecting the action with the maximum Q value for execution and pushing a price adjustment instruction to a parking lot charging system.

[0009] Optionally, the dynamic adjustment of the parking fee rate based on the parking space occupancy rate vector and the preset multi-objective reward function further comprises: at the end of each preset time step, determining a parking lot revenue based on the parking fee rate adjustment strategy with the maximum Q value, and calculating a reward value according to the preset multi-objective reward function based on a difference in the parking lot revenue, a change in the parking space utilization rate, and a number of user complaints in the time step; storing the current state, the executed parking fee rate adjustment strategy, the obtained reward value, and the next state as a sample in an experience replay pool, and when the number of samples accumulated in the replay pool reaches a preset value, randomly selecting the samples to form a small batch of data for parameter updating of the deep reinforcement learning model; taking a mean square error as a loss function to measure a difference between a predicted Q value and a target Q value, adjusting a learning rate through an optimizer, iteratively updating weights of the deep reinforcement learning model, gradually reducing a prediction error, and making the Q value continuously approach an optimal value to obtain the maximum Q value.

[0010] Optionally, the parking space prediction information and the dynamic pricing information are displayed through a user-side interactive interface, comprising: filtering nearby parking lot information according to a user location and displaying the information through the user-side interactive interface, the parking lot information comprising a distance, a number of remaining parking spaces, a current parking fee rate, and an estimated arrival time; receiving a parking space reservation request of a user, generating order information, and notifying the user through the user-side interactive interface; providing a parking space guidance service through the user-side interactive interface when the user arrives at the parking lot.

[0011] Optionally, the adjusted system parameters are obtained through a management-side interactive interface, comprising: real-time displaying parking space occupancy and revenue of the parking lot in the form of a visual chart; The system parameters include weight coefficients of a modified reward function, a prediction time range, and upper and lower limits of price adjustment.

[0012] In another aspect, an embodiment of the present application provides a smart parking space management device based on multi-source data fusion, comprising: A data acquisition module is configured to acquire multi-source data, including parking space state information inside a parking lot, traffic flow data around the parking lot, weather data, and date type information. A parking space prediction module is configured to fuse the multi-source data into a feature vector, input the feature vector into a trained parking space prediction model, and output a predicted parking space occupancy rate vector. A dynamic pricing strategy module is configured to dynamically adjust a parking fee rate based on the parking space occupancy rate vector and a preset multi-objective reward function through a deep reinforcement learning method. A human-computer interaction module is configured to display parking space prediction information and dynamic pricing information through a user-side interactive interface and acquire adjusted system parameters through a management-side interactive interface.

[0013] In another aspect, an embodiment of the present application provides a smart parking space management system based on multi-source data fusion, comprising: At least one processor; At least one memory configured to store at least one program; When the at least one program is executed by the at least one processor, the at least one processor implements the above method.

[0014] In another aspect, an embodiment of the present application provides a computer readable storage medium, which stores a program executable by a processor.

[0015] The embodiments of the present application include the following beneficial effects: the present application provides a smart parking space management method and related equipment based on multi-source data fusion, which integrates multi-dimensional information of the internal and external environment of the parking lot through multi-source data fusion technology, constructs a feature system that comprehensively reflects the parking supply and demand relationship, and overcomes the limitations of traditional methods that rely on only a single data dimension. Among them, the introduction of the spatio-temporal graph neural network realizes the collaborative modeling of the spatial correlation of parking spaces and the time sequence dependence. The spatial graph convolution layer captures the topological relationship between parking spaces through the graph structure, and the time feature extraction layer deeply excavates the periodicity and trend rules in the historical data. After dynamic weighted fusion through the attention mechanism, the parking occupancy rate prediction accuracy in complex scenarios is significantly improved. The dynamic pricing module constructs a closed-loop decision system based on deep reinforcement learning. The multi-objective reward function integrates the maximization of revenue, optimization of resource utilization, and user experience guarantee into a unified framework. Through the experience replay mechanism and gradient descent algorithm, the strategy continuously evolves, so that the parking rate can respond to dynamic factors such as traffic flow fluctuations and weather changes in real time. The two-way interactive interface of the user end and the management end forms a smart management closed loop. Users can obtain accurate parking recommendation and real-time pricing information, and managers can realize flexible configuration of system parameters through a visual data panel, so as to ultimately achieve multi-objective optimization balance of parking lot resource allocation efficiency, operating income and user satisfaction. BRIEF DESCRIPTION OF DRAWINGS

[0016] Figure 1 is a step flow diagram of a smart parking space management method based on multi-source data fusion provided by the embodiments of the present application; Figure 2 is a structural block diagram of a smart parking space management device based on multi-source data fusion provided by the embodiments of the present application; Figure 3 is a schematic diagram of multi-source data provided by the embodiments of the present application; Figure 4 is a structural block diagram of a parking space prediction module provided by the embodiments of the present application; Figure 5 is an architecture diagram of a parking space prediction model provided by the embodiments of the present application; Figure 6 is an architecture diagram of a dynamic pricing strategy model provided by the embodiments of the present application; Figure 7 is a structural block diagram of a human-computer interaction module provided by the embodiments of the present application. DETAILED DESCRIPTION

[0017] In order to make the purpose, technical scheme and advantages of the present application more clear, the present application is further described in detail below in combination with the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and not to limit the present application.

[0018] It should be noted that although the module division is made in the device schematic diagram, the logical order is shown in the flowchart, in some cases, the steps shown or described can be performed in a manner different from the module division in the device, or the order in the flowchart. The terms "first", "second", etc. in the description and claims and the above-mentioned drawings are used to distinguish similar objects, and do not necessarily describe a specific order or sequence.

[0019] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used in the specification herein is for describing the embodiments of the application only and is not intended to limit the application.

[0020] In addition, the described features, structures or characteristics can be combined in any suitable manner in one or more embodiments. In the following description, many specific details are provided to give a thorough understanding of embodiments of the application. However, one skilled in the art will recognize that the technical solutions of the application can be practiced without one or more of the specific details, or can employ other methods, components, devices, steps, etc. In other cases, well-known methods, devices, implementations or operations are not shown or described in detail to avoid obscuring aspects of the application.

[0021] The block diagram shown in the accompanying drawings is only a functional entity, which does not necessarily correspond to a physically independent entity. That is, the functional entity can be implemented in the form of software, or in one or more hardware modules or integrated circuits, or in different network and / or processor devices and / or microcontroller devices.

[0022] The flowchart shown in the accompanying drawings is only an exemplary illustration, which does not necessarily include all contents and operations / steps, and is not necessarily executed in the order described. For example, some operations / steps can be further divided, and some operations / steps can be combined or partially combined, so the actual execution order can be changed according to the actual situation.

[0023] As Figure 1 shown, Figure 1 A multi-source data fusion-based intelligent parking space management method is provided for embodiments of the application, which comprises the following steps: S100, collecting multi-source data, the multi-source data including parking space state information inside the parking lot, traffic flow data around the parking lot, weather data and date type information; S200, fuse the multi-source data into a feature vector, input the feature vector into a trained parking space prediction model, and output a predicted parking occupancy rate vector; wherein the parking occupancy rate vector includes a predicted occupancy rate value of each parking space at each time interval in the future; the parking space prediction model is trained based on a spatio-temporal graph neural network, and the parking space prediction model includes an input layer, a spatial graph convolution layer, a temporal feature extraction layer, a spatio-temporal feature fusion layer, and an output layer; S300, dynamically adjusting the parking rate based on the parking occupancy rate vector and a preset multi-objective reward function through a deep reinforcement learning method; the multi-objective reward function is established based on the difference of parking lot revenue, the change value of parking space utilization rate, and the number of user complaints; S400, displaying the parking space prediction information and the dynamic pricing information through a user terminal interactive interface, and obtaining the adjusted system parameters through a management terminal interactive interface.

[0024] In steps S100 to S300 shown in the embodiments of the present application, the intelligent parking space management scheme based on multi-source data fusion, when predicting the parking occupancy rate, deeply fuses weather data (such as precipitation, wind grade), holiday and special event data (such as concerts, sports events) with parking lot internal data and traffic flow data, and multi-dimensional data integration significantly improves the adaptability to complex scenarios.

[0025] Based on the multi-objective dynamic pricing scheme, the pricing strategy is dynamically adjusted, the pricing strategy comprehensively considers the change of revenue, the change of parking space utilization rate, and the number of user complaints, and an interface is reserved to adaptively adjust the weight coefficient, facilitating the administrator to dynamically adjust the pricing strategy, and realizing the multi-objective of “revenue-efficiency-experience”.

[0026] The perfect client and management terminal interactive module, the client in the man-machine interactive interface mentioned in the present application can intuitively feed back the prediction result and the pricing strategy to the user, can provide accurate parking space guidance service for the user, help the user quickly find the reserved parking space, and improve the user parking experience. The management terminal provides a visual chart (such as a heat map and a column chart) to real-time display the parking space occupancy of the parking lot, so that the management personnel can intuitively understand the real-time state of each parking space and the overall parking space utilization rate; the management personnel can adjust the system parameters in the management interface according to the actual operation situation, to prevent unreasonable large fluctuations of the parking rate.

[0027] In some embodiments, the collection of multi-source data includes: Collecting parking space internal parking space state information, the parking space state information including real-time occupancy state and historical parking records of each parking space in the parking lot; Collecting real-time traffic flow data of the parking lot surrounding road; Collecting weather data, the weather data including real-time and forecast weather data of the area where the parking place is located, including precipitation, wind level, etc. Collecting date type information, integrating date type of current and future time period, and special event information affecting parking demand around the parking place; Forming multi-source data by using parking space status information inside the parking place, traffic flow data around the parking place, weather data and date type information.

[0028] As shown in Figure 2 The intelligent parking space management device based on multi-source data fusion mentioned in the application comprises a data collection module, a parking space prediction module, a dynamic pricing module and a man-machine interaction module.

[0029] The data collection module mentioned in the application is as shown in Figure 3 The data collection module collects data affecting whether the parking space of the parking place is occupied, including but not limited to parking space status information inside the parking place, traffic flow data around the parking place, weather data and holiday or special event information, etc., and specifically: (I) parking place internal data collection; The data collection module mentioned in the application first needs to collect parking space and vehicle information inside the parking place, for example, in the application, cameras are deployed at the entrance and exit of the parking place and parking spaces to identify the license plate of passing vehicles, obtain vehicle entry and exit time, license plate number and other information, and transmit whether the parking space is occupied to the server in real time.

[0030] (II) traffic flow data acquisition; The data collection module mentioned in the application needs to collect traffic flow data of the road around the parking place. For example, in the application, the application obtains traffic flow data of the road around the parking place from the traffic management department server every T1 minutes (in the application, T1 is 15 minutes) by means of 4G / 5G communication network, including vehicle flow, average speed, congestion index and other data of each road section. These data can directly reflect the real-time traffic condition of the surrounding road, and provide important external influencing factor data support for parking space demand prediction.

[0031] (III) weather data collection; The data acquisition module mentioned in the present application needs to collect real-time weather data in the area where the parking place is located. For example, the present application sends a data request to the data interface of the meteorological department at regular intervals, and obtains real-time weather data in the area where the parking place is located every T2 (in the present application, T2 is 30 minutes). Weather data covers temperature, precipitation, wind force level, weather conditions (sunny, rainy, snowy, etc.). By analyzing the potential correlation between weather data and parking demand, the accuracy of parking space prediction and dynamic pricing can be improved, for example, people may prefer indoor parking lots on rainy days, resulting in increased demand for specific parking lots.

[0032] (Four) holiday and special event data integration; The data acquisition module mentioned in the present application needs to collect holidays and special events. For example, the system enters the date type information of previous years into the local database, and collects the time, location, etc. Information of special events (such as concerts, sports events, large exhibitions, etc.) that have been predicted and enters the database. The system regularly updates and maintains the database to ensure the timeliness and completeness of the data. When a special event occurs, combined with the type, size and expected number of participants, etc. Factors, comprehensive evaluation of its impact on the parking demand of the surrounding parking lot, provide comprehensive data basis for system decision-making.

[0033] The parking space prediction module mentioned in the present application is shown in Figure 4 The schematic diagram includes data set construction, model building and training, and model prediction application.

[0034] When the model training is completed and the expected performance is reached, the real-time acquisition and fusion of good multi-source data are input into the trained model. The model cooperates in each layer such as the spatial graph convolution layer, the time series processing layer, and the multi-modal spatio-temporal feature fusion layer, and outputs the occupancy rate prediction value of each parking space in each time interval in the future time.

[0035] In some embodiments, the parking space prediction model is trained by the following way: Collect historical multi-source data, arrange the historical multi-source data in chronological order, and divide the data set according to the set time unit; Establish a spatio-temporal graph neural network, input the data set into the spatio-temporal graph neural network for iterative training; the spatio-temporal graph neural network includes an input layer, a spatial graph convolution layer, a time feature extraction layer, a spatio-temporal feature fusion layer and an output layer; The input layer receives the feature vector obtained by fusing the historical multi-source data, the spatial graph convolution layer processes the feature vector, and extracts the spatial features between the parking spaces through graph convolution operation; the time feature extraction layer adopts long short-term memory network to process the feature vector in time dimension, and extracts the time features of parking demand in different time periods; The spatio-temporal feature fusion layer fuses the spatial features and the temporal features through an attention mechanism to obtain a comprehensive feature vector; After multiplying the comprehensive feature vector and the weight matrix of the output layer and adding the bias vector, an activation process is performed, and a predicted parking occupancy rate vector is output; the parking occupancy rate vector includes a predicted value of the occupancy rate of each parking space in each time interval in the future time; Based on the real parking occupancy rate and the predicted parking occupancy rate, a loss value of the spatio-temporal graph neural network is obtained, and when the loss value is determined to be lower than a loss threshold, the iterative training of the spatio-temporal graph neural network is stopped, and a parking space prediction model is obtained.

[0036] The model training module mentioned in the application is shown in the schematic diagram as Figure 5 The model prediction is shown in the schematic diagram as

[0037] (I) Data set construction and division; For example, the collected multi-source data is arranged in time sequence, and the data set is divided in weeks as a time unit. 80% of the data is used as a training set for model training, and the model learns the change rule of parking space demand through historical data; 10% of the data is used as a validation set to adjust the hyperparameters during model training to prevent model overfitting and ensure the generalization ability of the model under different data distributions; the remaining 10% of the data is used as a test set to evaluate the prediction performance of the trained model on unknown data and test the accuracy and reliability of the model.

[0038] (II) Parking space prediction model building and training; For example, the application is based on a spatio-temporal graph neural network (ST-GNN) parking space prediction model, as shown in Figure 4 The prediction model mentioned in the application is divided into five core structures, namely: input layer, spatial graph convolution layer, temporal feature extraction layer, spatio-temporal feature fusion layer and output layer. The hierarchical processing realizes the capture and prediction of the spatio-temporal dynamic characteristics of the parking space.

[0039] Input layer: The input layer receives the feature vector where N represents the number of parking spaces, and F represents the dimension of the feature. These fused feature vectors contain parking lot internal parking state information, surrounding traffic flow data, weather data and holidays, providing a data basis for subsequent model processing.

[0040] Spatial graph convolution layer: In the spatial graph convolution layer, the input features are processed through the formula where represents the learnable parameters in the spatial graph convolution operation, An activation function is used to perform a nonlinear transformation on the calculation result, so that the model can learn more complex spatial feature relationships. Spatial features between parking spaces are extracted through graph convolution operations, such as patterns of mutual influence of the occupancy states of adjacent parking spaces, thereby capturing the correlation of parking demand in the spatial dimension.

[0041] The time feature extraction layer adopts a long short-term memory network (LSTM), and the calculation process can be represented as: ; In the formula, represents the hidden state of the previous time step, represents the spatial feature output by the spatial graph convolution layer, represents the hidden state of the current time step, and the system processes the time series data (selectively remembers and forgets information) through the collaborative control of the forgetting gate, the input gate, and the output gate, thereby learning the periodicity, trend, and other time features of parking demand in different time periods. For example, the difference between weekday and weekend parking demand in terms of time distribution is learned.

[0042] The spatio-temporal feature fusion layer: on the basis of the spatial graph convolution layer and the time series processing layer, a multi-modal spatio-temporal feature fusion layer is introduced, and the fusion of spatial and temporal features is realized through the following formula: ; In the formula, are weight matrices, represents element-wise multiplication, represents the spatial feature output by the spatial graph convolution layer, represents the time feature output by the time series processing layer, represents the spatio-temporal feature output by the spatio-temporal feature fusion layer, and this layer automatically learns different data sources (parking lot internal data, traffic flow data, weather data, holiday and special event data) through an attention mechanism to adapt to complex parking scenarios.

[0043] In some embodiments, the spatio-temporal feature fusion layer fuses the spatial feature and the time feature through an attention mechanism to obtain a comprehensive feature vector, including: An adaptive spatio-temporal weight mechanism dynamically adjusts the weights of the spatial graph convolution layer and the time series processing layer in the spatio-temporal feature through learnable parameters, and outputs a fused feature vector; After a multi-scale modeling method is used to perform graph convolution operations on the fused feature vector at different scales, a splicing operation is performed to obtain the output of multi-scale spatio-temporal modeling; The output of multi-scale spatio-temporal modeling and the feature vector are added to obtain a comprehensive feature vector.

[0044] Specifically, the spatio-temporal graph neural network introduces an adaptive spatio-temporal weight mechanism, dynamically adjusting the weights of the spatial graph convolution layer and the time series processing layer through learnable parameters.

[0045] ; In the formula, represents the fusion feature vector after fusion, and respectively represent the spatial weight and the time weight , , , representing the splicing of spatial features and time features, respectively are learnable parameters, representing an activation function. During peak hours, the model automatically increases the weight of the time series processing layer, paying more attention to the time variation of parking demand; while during special events, the model strengthens the role of the spatial graph convolution layer, highlighting the mutual influence between adjacent parking spaces. This adaptive mechanism enables the model to better adapt to various complex scenarios, improving the flexibility and accuracy of the prediction.

[0046] Multi-scale spatio-temporal modeling: In order to more comprehensively capture the spatio-temporal features of parking demand, the model adopts a multi-scale modeling method.

[0047] ; In the formula, represents different scale graph convolution operations, representing a splicing operation. In the spatial dimension, not only the direct adjacent relationship between parking spaces is considered, but also the long-distance dependency relationship between parking spaces in a wider area is captured by constructing adjacency matrices of different scales; in the time dimension, short-term (10 minutes), medium-term (1 hour) and long-term (day, week) parking demand change patterns are modeled simultaneously. Through this multi-scale processing method, the model can more comprehensively and accurately capture the spatio-temporal features of parking demand, thereby improving the prediction accuracy.

[0048] Spatio-temporal residual connection: During model training, with the increase of network layers, problems such as gradient vanishing may occur, affecting the training effect and performance of the model. In order to solve the problem of gradient vanishing in the training process of deep neural networks, the present invention realizes spatio-temporal residual connection, directly adding the output of the original features of the input layer , retaining important spatio-temporal information in the original input, while helping the model to learn more complex spatio-temporal features, improving the learning ability and generalization performance of the model.

[0049] ; Output layer: The model output is a predicted parking occupancy rate vector, each element of which corresponds to a predicted value of a parking space.

[0050] wherein, is a comprehensive feature vector after pre-layer processing (such as spatiotemporal feature fusion, feature transformation), is a weight matrix of the output layer, is a bias vector, used to adjust the baseline of the predicted value, to avoid the model output always being biased towards 0 or 1.

[0051] Training process: Set the initial learning rate to 0.001, use the optimizer to update the model parameters, and use the mean square error (MSE) as the loss function: wherein, n is the number of samples, is the real parking occupancy rate, is the model predicted parking occupancy rate. After completing one epoch of training, input the validation set data into the model to calculate the MSE on the validation set. If the MSE of the validation set does not decrease for 5 consecutive epochs, reduce the learning rate to 0.5 times the original, and continue training. At the same time, use the early stopping mechanism, when the MSE on the validation set decreases by less than 0.001 for 10 consecutive epochs, stop training to prevent overfitting.

[0052] (Three) Model prediction application When the model training is completed and reaches the expected performance, input the real-time collected and fused good multi-source data into the trained model. The model outputs the occupancy rate prediction value of each parking space at every 10-minute time interval within the next 1 hour under the collaborative operation of each layer such as the spatial graph convolution layer, the time series processing layer, and the multi-modal spatiotemporal feature fusion layer. For example, if the current time is 10:00 am, the model will output the occupancy rate prediction result of each parking space at 6 time points from 10:10 to 11:00, providing forward-looking information for parking lot management and dynamic pricing.

[0053] In some embodiments, the parking fee rate is dynamically adjusted by a deep reinforcement learning method based on the parking occupancy rate vector and a preset multi-objective reward function, comprising: S311, define a state space, combine the parking occupancy rate vector, the traffic flow data around the parking lot, the time feature, and the date type information into a state vector; ​​The occupancy rate prediction value of the parking space is divided into multiple intervals, each interval is taken as an interval state feature;The surrounding traffic flow data is divided into multiple levels, each level is taken as a level state feature;The extracted time feature and date type information are taken as supplementary features;The interval state feature, the level state feature and the supplementary feature are combined into a state vector; S312, defining an action space, the action space containing multiple parking rate adjustment strategies; S313, taking the state vector as the input of a deep reinforcement learning model, calculating the Q value corresponding to each action, the Q value representing the long-term cumulative reward expected to be obtained after performing a certain action in the state; S314, selecting the action with the maximum Q value to execute, and pushing the pricing instruction to the parking lot charging system.

[0054] The dynamic pricing strategy mentioned in the application is as shown in the figure Figure 6 The application relies on a deep Q network (DQN) to build an intelligent decision-making closed loop, comprehensively considers factors such as parking space occupancy rate, parking lot revenue, and user feedback, and realizes dynamic optimization and precise regulation and control of the pricing strategy.

[0055] The specific operation process is as follows: the parking space occupancy rate is divided into 5 intervals (0-20%, 20%-40%, 40%-60%, 60%-80%, 80%-100%), each interval is taken as a state feature;The surrounding traffic flow data is divided into 5 levels (smooth, light congestion, moderate congestion, heavy congestion, and severe congestion) as another state feature;The time information is extracted as the hour (0-23) and whether it is a weekday (0 represents a non-working day and 1 represents a working day) as supplementary features. The action space is defined as 5 parking rate adjustment strategies, which are respectively increasing by 20%, increasing by 10%, keeping unchanged, decreasing by 10%, and decreasing by 20%. These features are combined into a state vector as the input of the DQN, and the Q value corresponding to each action (5 parking rate adjustment strategies) is calculated. The Q value represents the long-term cumulative reward expected to be obtained after performing a certain action in the state. The system selects the action with the maximum Q value to execute. For example, when it is calculated that the Q value corresponding to "increasing the parking rate by 10%" is the highest, the system immediately pushes the pricing instruction to the parking lot charging system, realizing real-time updating of the parking price.

[0056] The application takes 10 minutes as a time step, and after each time step ends, the system evaluates the decision-making effect according to the actual operation data. The actual revenue of the parking lot, the change of the parking space utilization rate, and the number of user complaints in the period are counted, and the reward value is calculated according to the preset reward function.

[0057] In some embodiments, the dynamic adjustment of the parking fee rate based on the parking space occupancy rate vector and the preset multi-objective reward function through the deep reinforcement learning method further comprises: S321, at the end of each preset time step, the parking lot revenue is determined based on the parking fee rate adjustment strategy with the maximum Q value, and the reward value is calculated according to the preset multi-objective reward function according to the difference of the parking lot revenue, the change value of the parking space utilization rate and the number of user complaints in the time step; S322, the current state, the executed parking fee rate adjustment strategy, the obtained reward value and the next state are stored in the experience replay pool as samples, and when the number of samples in the replay pool accumulates to a preset value, the samples are randomly extracted to form small batch data for parameter update of the deep reinforcement learning model; S323, the mean square error is used as a loss function to measure the difference between the predicted Q value and the target Q value, and the learning rate is adjusted by an optimizer to iteratively update the weights of the deep reinforcement learning model, gradually reducing the prediction error, so that the Q value continuously approaches the optimal value to obtain the maximum Q value.

[0058] The multi-objective reward function is set in the application, and multiple targets such as revenue change, parking space utilization rate change and user satisfaction are comprehensively considered, and the calculation formula is: ; In the formula, is a weight coefficient (for example, in the application, ), is the difference between the current time step and the last time step of the parking lot revenue, reflecting the direct influence of the parking fee rate adjustment on the revenue; is the change value of the parking space utilization rate, measuring the effect of the parking fee rate adjustment on the parking space utilization efficiency; is the number of user complaints, reflecting the acceptance degree of the user to the parking fee rate.

[0059] After the calculation is completed, the system stores the information of “current state - executed action - obtained reward - next state” in the experience replay pool. The experience replay pool stores the historical experience of the interaction between the agent and the environment. When the number of samples in the pool accumulates to N (N is 1000 in the application), the system randomly extracts N1 (N1 is 32 in the application) samples from the pool to form small batch data for model parameter update.

[0060] In the parameter update process, the mean square error (MSE) is used as a loss function to measure the difference between the predicted Q value and the target Q value. The learning rate is automatically adjusted by an optimizer to iteratively update the weights of the DQN network, gradually reducing the prediction error, so that the Q value continuously approaches the optimal value. With the passage of time, the agent continuously interacts with the parking lot operating environment, continuously accumulates new experience samples to achieve the balance of “revenue - efficiency - experience”.

[0061] In some embodiments, the parking space prediction information and dynamic pricing information are displayed through a user terminal interface, comprising: According to the user's location, the nearby parking lot information is filtered and displayed through the user terminal interface, including the distance, the number of remaining parking spaces, the current parking rate, and the estimated arrival time. The user's parking space reservation request is received, order information is generated, and the user is notified through the user terminal interface. When the user arrives at the parking lot, the parking space guidance service is provided through the user terminal interface.

[0062] Reference Figure 7 The human-computer interaction module mentioned in the present application is mainly divided into user terminal and management terminal. For the user terminal, when the user opens the application, the system obtains the user's current location information through the mobile phone GPS or web page IP positioning. The user can input the destination parking lot name or location in the search box. The system filters the parking lot information within 1 km range according to the user's location. At the same time, the distance from the user's location, the number of remaining parking spaces (predicted by the prediction module), the current parking rate (pricing strategy of the dynamic pricing module), and the estimated arrival time (calculated by the map navigation algorithm) are displayed. The user can select the parking lot of interest according to his own needs and click the "reserve parking space" button to reserve the parking space. After receiving the reservation request, the system checks whether there is available parking space in the parking lot within the user's estimated arrival time period. If there is, the order information is generated, including order number, number of reserved parking spaces, estimated parking duration, and payment, and the user is reminded of the successful reservation through SMS and APP notification. When the user arrives at the parking lot, the APP provides accurate parking guidance service for the user by using map navigation and pre-set parking guidance signs inside the parking lot, helping the user to quickly find the reserved parking space and improving the user's parking experience.

[0063] In some embodiments, the adjusted system parameters are obtained through the management terminal interface, comprising: The parking lot occupancy and revenue are displayed in real time in the form of a visual chart. The adjusted system parameters are obtained through the management terminal interface, including modifying the weight coefficient of the reward function, adjusting the prediction time range, and setting the upper and lower limits of the price adjustment.

[0064] Specifically, for the management end, the management interface displays the parking lot occupancy in real time in the form of visual charts (such as heat maps, bar charts), making it easy for the management personnel to intuitively understand the real-time state of each parking space and the overall parking space utilization rate; at the same time, the income of the parking lot is displayed, including the cumulative income of the day, the week, and the month, as well as the detailed information of each transaction (license plate number, entry and exit time, parking duration, charging amount, etc.). The management personnel can adjust the system parameters in the management interface according to the actual operation situation. For example, the weight coefficients of the reward function in the dynamic pricing module are modified, the emphasis of the pricing strategy on different goals such as parking lot income, parking space utilization rate, and user satisfaction is changed; the prediction time range of the parking space prediction module is adjusted (such as from 1 hour to 2 hours); the upper and lower limits of price adjustment are set to prevent unreasonable large fluctuations in parking fees.

[0065] Compared with the related art, the present application has the following advantages: Significant improvement in parking space prediction accuracy: Traditional parking space prediction relies on single historical data and is difficult to cope with complex and variable actual situations. The present application comprehensively captures various factors affecting parking space use by fusing parking lot historical data, real-time traffic flow data, weather information, city activity calendar, and other multi-source heterogeneous data, conducts in-depth analysis on multi-source data, and digs out potential rules and trends of parking space use, effectively solving the problem of low accuracy of traditional prediction methods, and providing a reliable decision basis for parking lot operation and management.

[0066] Adaptive optimization of dynamic pricing: Existing dynamic pricing is mostly based on simple rules and cannot flexibly cope with irregular events. The present application introduces a deep reinforcement learning algorithm to construct a dynamic decision-making model. This model can continuously interact with the real operating environment, automatically adjust the pricing parameters according to real-time parking space utilization rate, user feedback, and parking lot income, and realize dynamic optimization of prices.

[0067] Efficient allocation of resources and improvement of utilization rate: Based on accurate parking space prediction and dynamic pricing, the closed-loop management system formed by the present application can reasonably allocate parking lot resources. During peak hours, through price guidance and resource allocation, vehicles are guided to relatively idle areas to avoid local congestion; during off-peak hours, flexible pricing is used to attract more vehicles and improve parking space utilization. The system can also monitor and adjust the parking space usage in different areas according to real-time data, realize dynamic reallocation of parking space resources, and improve the overall operation efficiency of the parking lot.

[0068] User experience is significantly improved: car owners can learn about the parking lot situation in advance, plan the travel route reasonably, and reduce the time cost of finding parking spaces. Under the dynamic pricing mechanism, car owners can choose the appropriate parking time and place according to their own needs and price strategy, and reduce parking fee expenditure. Through the combination with the user terminal, the car owner can also obtain real-time parking lot information, navigate to the idle parking space, and enjoy the convenient payment service, greatly improving the convenience and comfort of parking, improving the user experience, and enhancing the attraction of the parking lot to the user.

[0069] Reference Figure 2 The embodiment of the application also provides a smart parking space management device based on multi-source data fusion, comprising: A data acquisition module is configured to acquire multi-source data, wherein the multi-source data comprises parking space status information inside a parking lot, traffic flow data around the parking lot, weather data, and date type information. A parking space prediction module is configured to fuse the multi-source data into a feature vector, input the feature vector into a trained parking space prediction model, and output a predicted parking space occupancy rate vector; wherein the parking space occupancy rate vector comprises a predicted occupancy rate value of each parking space in each time interval in the future; the parking space prediction model is trained based on a space-time graph neural network, and comprises an input layer, a spatial graph convolution layer, a time feature extraction layer, a space-time feature fusion layer, and an output layer. A dynamic pricing strategy module is configured to dynamically adjust a parking fee rate based on the parking space occupancy rate vector and a preset multi-objective reward function through a deep reinforcement learning method; the multi-objective reward function is established based on a difference value of parking lot revenue, a change value of parking space utilization rate, and a number of user complaints. A human-computer interaction module is configured to display parking space prediction information and dynamic pricing information through a user terminal interaction interface, and obtain adjusted system parameters through a management terminal interaction interface.

[0070] The embodiment of the application also provides a smart parking space management system based on multi-source data fusion, comprising a memory, a processor, and a program stored in the memory and executable on the processor; when the program is executed by the processor, the method of the above embodiment is realized.

[0071] For example, the processor and the memory in the vehicle controller can be connected through a bus. The memory, as a non-transitory computer readable storage medium, can be used to store non-transitory software programs and non-transitory computer executable programs. In addition, the memory can include a high-speed random access memory, and can also include a non-transitory memory, such as at least one disk memory, a flash memory device, or other non-transitory solid-state memory device. In some embodiments, the memory can optionally include a memory remotely arranged relative to the control processor, and these remote memories can be connected to the control device through a network.

[0072] The non-transitory software programs and instructions required to implement the above-mentioned embodiments of the method are stored in the memory, and when executed by the processor, the method in the above-mentioned embodiments is executed.

[0073] The device embodiments described above are only illustrative, and the units described as separate components can or can not be physically separated, i.e., can be located in one place, or can be distributed to multiple network units. Part or all of the modules can be selected according to actual needs to achieve the purpose of the present embodiment.

[0074] The embodiment of the present application also provides a vehicle comprising the vehicle control device of the above-mentioned embodiments.

[0075] The vehicle can be a private car, such as a sedan, an SUV, an MPV, or a pickup truck, etc. The vehicle can also be an operating vehicle, such as a van, a bus, a small truck, or a large trailer, etc. The vehicle needs to have an electric motor that can output power or store mechanical energy as a generator. When the vehicle is a new energy vehicle, it can be a hybrid vehicle, or a pure electric vehicle.

[0076] Since the vehicle applies all the technical solutions of the above-mentioned control device or vehicle controller, it at least has all the beneficial effects brought by the technical solutions of the above-mentioned embodiments, which will not be repeated here.

[0077] In addition, one embodiment of the present application also provides a computer readable storage medium, which stores computer executable instructions, and the computer executable instructions are used to execute the above-mentioned method.

[0078] It is worth noting that since the computer readable storage medium of the embodiment of the present application can execute the method of any of the above-mentioned embodiments, the specific implementation and technical effects of the computer readable storage medium of the embodiment of the present application can be referred to the specific implementation and technical effects of the method of any of the above-mentioned embodiments.

[0079] In addition, an embodiment of the present application further provides a computer program product, comprising a computer program or computer instructions, the computer program or computer instructions being stored in a computer readable storage medium, a processor of a computer device reading the computer program or computer instructions from the computer readable storage medium, and the processor executing the computer program or computer instructions, so that the computer device executes the method described above.

[0080] It is worth noting that since the computer program product of the embodiment of the present application can execute the method of any of the above embodiments, the specific implementation and technical effects of the computer program product of the embodiment of the present application can refer to the specific implementation and technical effects of the method of any of the above embodiments.

[0081] Those skilled in the art can understand that all or some of the steps in the method disclosed above can be implemented as software, firmware, hardware, and appropriate combinations thereof. Some or all of the physical components can be implemented as software executed by a processor, such as a central processing unit, a digital signal processor, or a microprocessor, or as hardware, or as an integrated circuit, such as an application specific integrated circuit. Such software can be distributed on a computer readable medium, which can include computer storage media (or non-transitory media) and communication media (or transitory media). As known to those skilled in the art, the term computer storage media includes volatile and non-volatile, removable and non-removable media implemented in any method or technology for storage of information such as computer readable instructions, data structures, program modules or other data. Computer storage media includes, but is not limited to, RAM, ROM, EEPROM, flash memory or other memory technology, CD-ROM, digital versatile disks (DVD) or other optical disk storage, magnetic cassettes, magnetic tapes, magnetic disk storage or other magnetic storage devices, or any other medium that can be used to store the desired information and can be accessed by a computer. In addition, as known to those skilled in the art, communication media generally includes computer readable instructions, data structures, program modules or other data in modulated data signals such as carrier waves or other transport mechanisms, and can include any information delivery medium.

[0082] The device embodiments described above are only schematic, wherein the units described as separate components can or can not be physically separate, i.e. can be located in one place or distributed on a plurality of network units. Some or all of the modules can be selected according to actual needs to achieve the purposes of the embodiments.

Claims

1. A smart parking space management method based on multi-source data fusion, characterized in that, The method comprises: Collecting multi-source data, the multi-source data comprising parking space status information inside a parking lot, traffic flow data around the parking lot, weather data, and date type information; Fusing the multi-source data into a feature vector, inputting the feature vector into a trained parking space prediction model, and outputting a predicted parking occupancy rate vector; wherein the parking occupancy rate vector comprises predicted occupancy rate values of each parking space in each time interval in a future time; the parking space prediction model is trained based on a spatio-temporal graph neural network, and comprises an input layer, a spatial graph convolution layer, a time feature extraction layer, a spatio-temporal feature fusion layer, and an output layer; Based on the parking occupancy rate vector and a preset multi-objective reward function, dynamically adjusting a parking rate by a deep reinforcement learning method; the multi-objective reward function is established based on a difference in parking lot revenue, a change value of parking space utilization rate, and a number of user complaints; Displaying the parking space prediction information and the dynamic pricing information through a user terminal interactive interface, and obtaining adjusted system parameters through a management terminal interactive interface.

2. The method of claim 1, wherein, The parking space prediction model is trained in the following manner: Collecting historical multi-source data, arranging the historical multi-source data in chronological order, and dividing the data set in a set time unit; Establishing a spatio-temporal graph neural network, inputting the data set into the spatio-temporal graph neural network for iterative training; the spatio-temporal graph neural network comprises an input layer, a spatial graph convolution layer, a time feature extraction layer, a spatio-temporal feature fusion layer, and an output layer; The input layer receives a feature vector fused from the historical multi-source data, the spatial graph convolution layer processes the feature vector, and the spatial features between parking spaces are extracted through graph convolution operation; The time feature extraction layer processes the feature vector in the time dimension by using a long short-term memory network, and extracts time features of parking demand in different time periods; The spatio-temporal feature fusion layer fuses the spatial features and the time features to obtain a comprehensive feature vector through an attention mechanism; After multiplying the comprehensive feature vector and a weight matrix of the output layer and adding a bias vector, the predicted parking occupancy rate vector is outputted after activation processing; the parking occupancy rate vector comprises predicted occupancy rate values of each parking space in each time interval in a future time; Based on the real parking occupancy rate and the predicted parking occupancy rate, a loss value of the spatio-temporal graph neural network is obtained, and when the loss value is determined to be lower than a loss threshold value, the iterative training of the spatio-temporal graph neural network is stopped, and a parking space prediction model is obtained.

3. The method of claim 2, wherein, The spatio-temporal feature fusion layer fuses the spatial features and the time features to obtain a comprehensive feature vector through an attention mechanism, comprising: An adaptive spatio-temporal weight mechanism dynamically adjusts the weights of the spatial graph convolution layer and the time sequence processing layer in the spatio-temporal features through learnable parameters, and outputs a fused feature vector; After performing different scale graph convolution operations on the fused feature vector by using a multi-scale modeling method, a splicing operation is performed, and an output of multi-scale spatio-temporal modeling is obtained; The output of multi-scale spatio-temporal modeling and the feature vector are added to obtain a comprehensive feature vector.

4. The method of claim 1, wherein, The dynamic adjustment of the parking fee rate based on the parking space occupancy rate vector and the preset multi-objective reward function comprises the following steps: Defining a state space, combining the parking space occupancy rate vector, the traffic flow data around the parking lot, the time characteristics, and the date type information into a state vector; Defining an action space, wherein the action space comprises a plurality of parking fee rate adjustment strategies; Taking the state vector as the input of a deep reinforcement learning model, calculating the Q value corresponding to each action, wherein the Q value represents the long-term cumulative reward expected to be obtained after performing a certain action in the state; Selecting the action with the maximum Q value and pushing the price adjustment instruction to the parking lot charging system.

5. The method of claim 4, wherein, The dynamic adjustment of the parking fee rate based on the parking space occupancy rate vector and the preset multi-objective reward function further comprises the following steps: After each preset time step ends, determining the parking lot revenue based on the parking fee rate adjustment strategy with the maximum Q value, calculating the reward value according to the difference in the parking lot revenue, the change in the parking space utilization rate, and the number of user complaints in the time step, and according to the preset multi-objective reward function; Storing the current state, the executed parking fee rate adjustment strategy, the obtained reward value, and the next state as samples in an experience replay pool, randomly extracting samples to form a small batch of data for parameter updating of the deep reinforcement learning model when the number of samples in the replay pool accumulates to a preset value; Taking the mean square error as a loss function to measure the difference between the predicted Q value and the target Q value, adjusting the learning rate through an optimizer, iteratively updating the weights of the deep reinforcement learning model, gradually reducing the prediction error, and making the Q value continuously approach the optimal value to obtain the maximum Q value.

6. The method of claim 1, wherein, The parking space prediction information and the dynamic pricing information are displayed through a user-side interactive interface, which comprises the following steps: Filtering the parking lot information near the user's location according to the user's location and displaying the parking lot information through the user-side interactive interface, wherein the parking lot information comprises the distance, the number of remaining parking spaces, the current parking fee rate, and the estimated arrival time; Receiving the parking space reservation request of the user, generating order information, and notifying the user through the user-side interactive interface; Providing parking space guidance services through the user-side interactive interface when the user arrives at the parking lot.

7. The method of claim 1, wherein, The adjusted system parameters are obtained through a management-side interactive interface, which comprises the following steps: Real-time displaying the parking space occupancy and the revenue of the parking lot in the form of a visual chart; Obtaining the adjusted system parameters through the management-side interactive interface, wherein the system parameters comprise modifying the weight coefficients of the reward function, adjusting the prediction time range, and setting the upper and lower limits of the price adjustment.

8. A smart parking space management device based on multi-source data fusion, characterized in that, The device comprises: A data acquisition module for acquiring multi-source data, wherein the multi-source data comprises the parking space state information inside the parking lot, the traffic flow data around the parking lot, the weather data, and the date type information; The parking space prediction module is configured to fuse the multi-source data into a feature vector, input the feature vector into a trained parking space prediction model, and output a predicted parking space occupancy rate vector; the parking space occupancy rate vector includes predicted occupancy rate values of each parking space at each time interval in the future; the parking space prediction model is trained based on a space-time graph neural network, and includes an input layer, a space graph convolution layer, a time feature extraction layer, a space-time feature fusion layer, and an output layer; The dynamic pricing strategy module is configured to dynamically adjust a parking fee rate based on the parking space occupancy rate vector and a preset multi-objective reward function through a deep reinforcement learning method; the multi-objective reward function is established based on a difference in parking lot revenue, a change value of parking space utilization rate, and a number of user complaints; The human-computer interaction module is configured to display parking space prediction information and dynamic pricing information through a user-side interaction interface, and obtain adjusted system parameters through a management-side interaction interface.

9. A smart parking space management system based on multi-source data fusion, characterized in that, The computer program is executed by the processor to implement the method in any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, the computer program comprising instructions that, when executed by a computer, cause the computer to perform the method of any one of claims 1 to 9. The computer program is executed by the processor to implement the method in any one of claims 1 to 7.

Images