Optimized scheduling method for intelligent parking lot of intelligent community

Abstract

The invention relates to the technical field of parking lot operation, and discloses an optimal scheduling method for an intelligent parking lot of an intelligent community, the core of the method is to construct an intelligent parking decision model, the model models the scheduling process of the parking lot into a Markov decision process, and intelligent scheduling is realized by defining a state space, an action space, rewards and strategies. The reward function design comprehensively considers the parking space utilization rate, the vehicle access speed, the user satisfaction degree and the operation cost, and supports dynamic adjustment to meet actual requirements and user feedback. And carrying out model training by adopting a deep Q network algorithm, and evaluating a model effect through an A / B test. In the real-time scheduling stage, the method can collect the state information of the parking lot in real time, selects the optimal action according to the current state and strategy, and achieves efficient and intelligent parking management. The implementation of the invention is helpful for improving the overall intelligent level of the smart community, enhancing the attraction and competitiveness of the community, and providing powerful support for the construction and development of the smart community.

Description

Technical Field

[0001] The present invention relates to the technical field of parking lot operation, and in particular to an optimization scheduling method for an intelligent parking lot in a smart community. Background Art

[0002] In the construction and development of smart communities, smart parking lots are key components, and their operational efficiency and user experience are directly related to the overall intelligence level of the community. Currently, most parking lots still use traditional manual management or basic automated control systems during operation. These methods have significant defects and fail to fully tap the application potential of smart technology in parking lot management.

[0003] Existing technologies lack intelligent management decision-making capabilities. Traditional parking lot management systems often operate based on fixed rules and processes and cannot be dynamically adjusted according to the real-time status and needs of the parking lot. For example, during peak hours, the system cannot automatically identify and prioritize vacant parking spaces for vehicles in urgent need of parking, resulting in a waste of parking space resources and vehicles waiting in line. This lack of flexibility and intelligence in management has seriously affected the operational efficiency of the parking lot and the user's parking experience. Summary of the invention

[0004] The purpose of the present invention is to provide an optimization scheduling method for smart parking lots in smart communities to solve the problems raised in the above-mentioned background technology.

[0005] To achieve the above object, the present invention provides the following technical solution: an optimization scheduling method for a smart parking lot in a smart community, the method comprising:

[0006] Construct an intelligent parking decision model, model the scheduling process of the intelligent parking lot as a Markov decision process, and specifically define the state space, action space, reward and strategy;

[0007] Design a reward function based on the goals of smart parking lots, including reward mechanisms for increased parking space utilization, faster vehicle entry and exit, improved user satisfaction, and reduced operating costs;

[0008] Dynamic reward function adjustment: dynamically adjust the weight coefficient in the reward function according to the actual needs of parking lot operations and user feedback, including increasing rewards for environmental protection and energy-saving measures and providing incentives for preferential policies for specific user groups;

[0009] Train the intelligent parking decision model, select the deep Q network algorithm, perform iterative training on actual data, and tune the hyperparameters;

[0010] Evaluate the intelligent parking decision model and compare the model recommendation with the existing parking management solution through A / B testing.

[0011] Real-time scheduling is performed based on the trained intelligent parking decision model, parking lot status information is collected in real time, the optimal action is selected according to the current status and strategy, and the decision results are transmitted to the parking lot management system for execution.

[0012] Preferably, the state space is represented in the form of a matrix, including the occupancy status of each parking space in the parking lot, historical data of vehicle entry and exit, whether the current time is a peak period, and current weather conditions;

[0013] The action space is designed to optimize parking space utilization, improve parking efficiency, and reduce operating costs, including allocating parking spaces, guiding vehicles in and out, adjusting parking space prices, and turning on or off lighting and ventilation systems in specific areas;

[0014] The strategy is a rule for selecting actions in a given state, and the goal is to maximize the cumulative reward by learning and optimizing the strategy.

[0015] Preferably, the reward function is designed as R(s,a), where s represents the current state, a represents the action taken, and R is the immediate reward obtained after executing action a in state s;

[0016] The reward function R(s,a) is composed of multiple weighted sub-reward items, including:

[0017] Parking space utilization bonus: Among them, N occupied is the number of currently occupied parking spaces, N total is the total number of parking spaces, α is the utilization weight coefficient, β is the optimal utilization threshold, and a positive reward is given when the utilization rate is higher than β, and a negative reward or zero reward is given when the utilization rate is lower than β;

[0018] Vehicle entry and exit speed bonus: Among them, T in and T out are the average entry and exit times of vehicles, γ is the speed weight coefficient, and δ is the inverse of the ideal average time threshold set, which encourages reducing vehicle waiting time;

[0019] User Satisfaction Award: R satisfaction =∈×(S survey -ζ); where S survey is the user satisfaction survey result, ∈ is the satisfaction weight coefficient, ζ is the satisfaction benchmark value, and positive rewards are given if the value is higher than the benchmark value;

[0020] Operational cost saving bonus: R cost =-η×C; where C is the operating cost caused by the current action, η is the cost saving weight coefficient, and the negative sign indicates that the cost reduction is regarded as a positive reward;

[0021] The final reward function is the sum of all sub-reward items, that is, R(s, a) = R utilization +R speed +R satisfaction +R cost ; By adjusting the weight coefficients α, γ, ∈, η and thresholds β, δ, ζ, preference setting and balance optimization for different operation objectives can be achieved.

[0022] Preferably, the steps of training the intelligent parking decision model include:

[0023] S1: Collect and preprocess actual parking lot operation data, including vehicle entry and exit records, parking space occupancy status, user feedback, and operation cost information, and clean and preprocess the collected data to form a training data set;

[0024] S2: A deep Q network is used as the decision model. The model consists of two parts: a deep neural network (DNN) and a Q value estimation. The DNN is used to extract state features and output the Q value estimation of each action.

[0025] S3: Use the preprocessed training data set to iteratively train the DQN model. During each training, a batch of samples are randomly selected from the data set. The loss is calculated through the loss function and the DNN parameters are updated. At the same time, the target network parameters θ-=θ are regularly updated. The calculation formula of the loss function is:

[0026]

[0027] Where r is the immediate reward, γ is the discount factor, s′ is the next state to be transferred after executing the action, and a′ is the optional action for the next state; θ is the DNN parameter, θ- is the target network parameter, and N is the batch sample size, which is used to optimize the DNN to reduce the gap between the predicted Q value and the true Q value;

[0028] S4: Evaluate the tuned DQN model on an independent test dataset to verify the effectiveness and practicality of the model by comparing the consistency of the parking decision predicted by the model with the actual optimal decision, user satisfaction, parking space utilization, vehicle entry and exit speed, and operating cost.

[0029] Preferably, in step S1, the method of cleaning the collected data includes:

[0030] Missing value processing: For missing values, linear interpolation algorithm is used to fill them;

[0031] Denoising: For noisy data, a sliding average filtering algorithm is used for data denoising. The algorithm formula is:

[0032]

[0033] Among them, y[n] is the denoised data, x[n] is the original data, and N is the size of the sliding window.

[0034] Preferably, in step S1, the method of preprocessing the collected data includes:

[0035] Standardization: Standardize continuous data to eliminate dimensional differences. Use the Z-score method to standardize the data. Subtract the mean from each numerical data and divide it by its standard deviation so that the processed data conforms to the standard normal distribution. The formula for standardization is: Among them, X is the original data, μ is the mean of the original data, σ is the standard deviation of the original data, and Z is the standardized data;

[0036] Encoding processing: Perform one-hot encoding conversion on categorical data.

[0037] Preferably, the steps of evaluating the performance of the intelligent parking decision model by using an A / B test comparative experimental method include:

[0038] Collect a certain amount of independent parking lot data as a test set to ensure that the test set has no intersection with the training set data;

[0039] Design an A / B test plan, where Group A uses the existing parking management plan for parking management, and Group B uses the model recommendation plan for parking management, and the parking lots of Group A and Group B maintain similar distributions in key features; the key features include the number of parking spaces, parking demand, and traffic flow;

[0040] Conduct tests in a simulated environment to record the parking management process and results of two groups of parking lots; design evaluation indicators, including parking efficiency, parking space utilization, customer satisfaction, and parking fees, to quantify the performance differences between the model-recommended solution and the existing parking management solution;

[0041] The chi-square test method was used to calculate and compare the statistical differences between group A and group B in various evaluation indicators.

[0042] Preferably, the method further comprises: establishing an adaptive feedback mechanism, regularly collecting parking lot operation data and user satisfaction survey results, and automatically adjusting the weight coefficient in the reward function.

[0043] Compared with the prior art, the present invention has the following beneficial effects:

[0044] By introducing an intelligent parking decision model, the present invention enables the parking lot to be dynamically adjusted according to real-time status and demand, optimizes parking space allocation and vehicle entry and exit processes, effectively reduces vehicle queuing waiting time, improves parking space utilization, and thus significantly improves the operational efficiency of the parking lot.

[0045] The intelligent parking decision model not only considers parking space utilization, vehicle entry and exit speed, and user satisfaction, but also incorporates an incentive mechanism for reducing operating costs. By optimizing the scheduling strategy and reducing unnecessary energy consumption and labor costs, the operating costs of parking lots can be reduced. As a key component of smart communities, the proposal and implementation of intelligent parking lot optimization scheduling methods will help improve the overall intelligence level of the community, and enhance the attractiveness and competitiveness of the community by improving parking lot operation efficiency and user experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 A step diagram of an optimization scheduling method for a smart parking lot in a smart community according to the present invention;

[0047] Figure 2 Schematic diagram of reward function design scheme;

[0048] Figure 3 A diagram showing the steps for conducting an A / B test comparison experiment. DETAILED DESCRIPTION

[0049] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0050] See also Figure 1-3 The present invention provides a technical solution: an optimization scheduling method for a smart parking lot in a smart community, the method comprising:

[0051] An intelligent parking decision model is constructed, and the scheduling process of the intelligent parking lot is modeled as a Markov decision process (MDP). Specifically, the state space is defined as the state set of all parking spaces in the parking lot. The action space is defined as the set of all actions that the parking lot management system can take. The state space is represented in the form of a matrix, including the occupancy status of each parking space in the parking lot, the historical data of vehicle entry and exit, whether the current time is a peak period, and the current weather conditions. The action space aims to optimize the utilization of parking spaces, improve parking efficiency, and reduce operating costs, including allocating parking spaces, guiding vehicles in and out, adjusting parking space prices, and turning on or off the lighting and ventilation systems in specific areas.

[0052] The reward function is designed according to the goals of the smart parking lot, including reward mechanisms for improving parking space utilization, accelerating vehicle entry and exit speeds, improving user satisfaction, and reducing operating costs. The strategy is a rule that selects the best action to maximize long-term cumulative rewards based on the current state and reward function.

[0053] The reward function is the core part of the intelligent parking decision model, which directly affects the decision-making effect of the model. When designing the reward function, multiple factors need to be considered comprehensively, including parking space utilization, vehicle entry and exit speed, user satisfaction, and operating costs. Specifically, a weight coefficient can be set for each factor and adjusted according to actual conditions. For example, in order to improve parking space utilization, a higher reward can be set for successfully allocating an empty parking space; in order to speed up vehicle entry and exit, a reward can be set for quickly guiding vehicles in and out; in order to improve user satisfaction, a reward can be set for providing high-quality services and preferential policies; in order to reduce operating costs, a reward can be set for reducing manpower and material inputs.

[0054] In the actual operation process, the demand and user feedback of the parking lot may change, so the weight coefficient in the reward function needs to be adjusted dynamically. Specifically, by collecting parking lot operation data and user feedback information, the current needs and pain points can be analyzed, and then the weight coefficient in the reward function can be adjusted accordingly. For example, if environmental protection and energy conservation have become important issues in the community, rewards for environmental protection and energy conservation measures can be increased; if specific user groups (such as the elderly, the disabled, etc.) have special needs for parking services, preferential policies can be provided for them and corresponding rewards can be set.

[0055] The Deep Q Network (DQN) algorithm is used to train the intelligent parking decision model. First, historical data of the parking lot is collected, including parking space status, vehicle entry and exit records, user feedback and other information. Then, the DQN is iteratively trained using this data, and the model is able to learn the optimal decision strategy by continuously adjusting the network parameters. During the training process, the hyperparameters also need to be tuned to improve the performance and stability of the model.

[0056] In order to verify the effectiveness of the intelligent parking decision model, an evaluation experiment is required. Specifically, the A / B test comparative experiment method can be used to compare the model recommendation scheme with the existing parking management scheme. During the experiment, relevant data needs to be collected and analyzed, including indicators such as parking space utilization, vehicle entry and exit speed, user satisfaction, and operating costs. By comparing the experimental results, the advantages and disadvantages of the model recommendation scheme can be evaluated, and a basis for subsequent optimization can be provided.

[0057] After the evaluation experiment is passed, the trained intelligent parking decision model is deployed to the parking lot management system for real-time scheduling. Specifically, it is necessary to collect parking lot status information in real time, including parking space status, vehicle entry and exit, user needs, etc. Then, the optimal action is selected based on the current status and strategy, and the decision result is transmitted to the parking lot management system for execution. During the real-time scheduling process, it is also necessary to continuously collect feedback data and fine-tune the model to adapt to the actual changes in parking lot operations.

[0058] The present invention will be further described below in conjunction with Examples 1 to 3:

[0059] Embodiment 1:

[0060] The reward function is used to drive the intelligent parking decision model to make the optimal scheduling decision. The following describes in detail the implementation of the reward function with specific examples:

[0061] The reward function is designed as R(s,a), where s represents the current state, a represents the action taken, and R is the immediate reward obtained after executing action a in state s;

[0062] The reward function R(s,a) is composed of multiple weighted sub-reward items, including:

[0063] Parking space utilization bonus: Among them, N occupied is the number of currently occupied parking spaces, N total is the total number of parking spaces, α is the utilization weight coefficient, β is the set optimal utilization threshold, and a positive reward is given when the utilization rate is higher than β, and a negative reward or zero reward is given when the utilization rate is lower than β.

[0064] Assume that a smart community parking lot has 100 parking spaces and the optimal utilization threshold is set to 80%. At a certain moment, 75 parking spaces in the parking lot are occupied. According to the reward function formula, the parking space utilization reward is calculated as follows: If α is 10, then R utilization =-0.5, indicating that the current utilization rate is lower than the optimal threshold, and a negative reward is given.

[0065] Vehicle entry and exit speed bonus: Among them, T in and T out are the average entry and exit times of vehicles, γ is the speed weight coefficient, and δ is the inverse of the ideal average time threshold, which encourages reducing vehicle waiting time.

[0066] Assuming that the average vehicle entry time is 30 seconds and the average exit time is 45 seconds, the ideal average time threshold is set to 60 seconds (i.e., the total entry and exit time of the vehicle is expected to be no more than 1 minute). According to the reward function formula, the vehicle entry and exit speed reward is calculated as follows: If γ is 50, then R speed ≈-0.11, indicating that the current vehicle entry and exit speed is lower than the ideal value, and a negative reward is given (here, since the time unit is seconds, the calculated reward value is small, and in actual applications, it may be necessary to adjust the time unit or introduce a magnification factor).

[0067] User Satisfaction Award: R satisfaction =∈×(S survey -ζ); where S survey is the user satisfaction survey result, ∈ is the satisfaction weight coefficient, ζ is the satisfaction benchmark value, and positive rewards are given if it is higher than the benchmark value.

[0068] Assume that the satisfaction score obtained through the user survey is 4.5 (out of 5 points), and the satisfaction benchmark value is set to 4 points. According to the reward function formula, the user satisfaction reward is calculated as follows: R satisfaction =∈×(4.5-4)=∈×0.5; if ∈ is 20, then R satisfaction =10, indicating that user satisfaction is higher than the benchmark value, and a positive reward is given.

[0069] Operational cost saving bonus: R cost =-η×C; where C is the operating cost caused by the current action, η is the cost saving weight coefficient, and the negative sign indicates that cost reduction is regarded as a positive reward.

[0070] Assume that the operating cost caused by the current action is 10 yuan (including energy consumption, labor costs, etc.), and the cost saving weight coefficient is set to -0.5 (the negative sign indicates that cost reduction is regarded as a positive reward). According to the reward function formula, the operating cost saving reward is calculated as follows: R cost =-η×10=-(-0.5)×10=5; it means that the current action reduces the operating cost and gives a positive reward.

[0071] The final reward function is the sum of all sub-reward items, that is, R(s, a) = R utilization +R speed +R satisfaction +R cost ;Right now:

[0072] R(s,a)=R utilization +R speed +R satisfaction +R cost

[0073] =-0.5-0.11+10+5≈14.39

[0074] By adjusting the weight coefficients α, γ, ∈, η and the thresholds β, δ, ζ, preference setting and balance optimization for different operation objectives can be achieved.

[0075] Embodiment 2:

[0076] The present invention provides a method for training an intelligent parking decision model based on a deep Q network, and the specific steps include:

[0077] S1: Collect the operation data of a smart parking lot in a smart community for one month, including vehicle entry and exit timestamps, parking space occupancy status records, satisfaction scores submitted by users through the APP, and daily operation costs (such as electricity consumption and manpower expenses), etc., and clean and preprocess the collected data to form a training data set.

[0078] Methods for cleaning the collected data include:

[0079] Missing value processing: For missing values, linear interpolation algorithm is used to fill them.

[0080] Denoising: For noisy data, a sliding average filtering algorithm is used for data denoising. The algorithm formula is:

[0081]

[0082] Among them, y[n] is the denoised data, x[n] is the original data, and N is the size of the sliding window.

[0083] The methods for preprocessing the collected data include:

[0084] Standardization: Standardize continuous data to eliminate dimensional differences. Use the Z-score method to standardize the data. Subtract the mean from each numerical data and divide it by its standard deviation so that the processed data conforms to the standard normal distribution. The formula for standardization is: Among them, X is the original data, μ is the mean of the original data, σ is the standard deviation of the original data, and Z is the standardized data.

[0085] Encoding processing: Perform one-hot encoding conversion on categorical data.

[0086] S2: Design a deep Q network (DQN) model, where the deep neural network (DNN) part contains 3 fully connected layers, each with 128, 64 and 32 neurons respectively, using ReLU as the activation function, and the output layer is linearly activated, corresponding to the Q value estimation of each possible action.

[0087] Initialize the DQN model parameters, including learning rate α = 0.01, discount factor γ = 0.99, batch size N = 32, etc., and randomly initialize the DNN weights.

[0088] S3: In each training iteration, a batch of samples are randomly selected from the preprocessed data set and input into the DQN model to calculate the Q value estimate of each action in the current state s. Based on the actual action a, the immediate reward r, and the next state s′ transferred to, the Q value update formula is used to update the Q value estimate in the DNN; the Q value update formula is:

[0089] Q(s,a)←Q(s,a)+α[r+γmax a′ Q(s′ i , a′; θ - )-Q(s,a)]

[0090] Calculate the loss function and optimize the DNN parameter θ through the back propagation algorithm. The algorithm formula of the loss function is:

[0091]

[0092] Among them, s′ is the next state transferred to after executing the action, a′ is the optional action of the next state; θ is the DNN parameter, θ - is the target network parameter and N is the batch sample size, which is used to optimize DNN to reduce the gap between the predicted Q value and the true Q value.

[0093] S4: Set the number of training rounds to 10,000. Perform model evaluation after each round of training and record changes in key indicators (such as cumulative rewards, parking space utilization, user satisfaction, etc.).

[0094] The grid search method was used to tune the hyperparameters such as learning rate α, discount factor γ, batch size N, etc. A total of 10 different parameter combinations were tried, and finally α=0.005, γ=0.95, N=64 were selected as the optimal parameter combination.

[0095] Under the tuned parameters, retrain the DQN model until convergence, that is, the loss function value is stable and no longer decreases significantly.

[0096] S5: Use an independent test data set (containing another month of parking lot operation data) to evaluate the performance of the tuned DQN model. Compare the parking decisions recommended by the model with the decisions taken in actual operations, and calculate indicators such as decision consistency rate, parking space utilization rate improvement ratio, user satisfaction improvement score, and operating cost savings ratio.

[0097] Embodiment 3:

[0098] The steps to evaluate the performance of the smart parking decision model through the A / B test comparative experimental method include:

[0099] Collect test set data: Select independent parking lots with a certain number of parking lots from multiple smart community parking lots as the test set. Make sure that these test set data do not overlap with the training set data used in model training to avoid data leakage and overfitting problems. Assume that 20 parking lots are selected as the test set, which are representative in terms of geographical location, number of parking spaces, parking demand, etc.

[0100] Design an A / B test plan: randomly divide the parking lots in the test set into Group A and Group B, with 10 parking lots in each group. Group A uses the existing parking management plan for parking management as the control group; Group B uses the plan recommended by the intelligent parking decision model proposed in this invention for parking management as the experimental group. Ensure that the parking lots in Group A and Group B maintain similar distributions in key features, including the number of parking spaces, parking demand, and traffic flow, to eliminate the impact of these factors on the experimental results.

[0101] Execute the test and record the data: In the simulation environment, perform the parking management process for the parking lots in Group A and Group B respectively. Record the parking management process and results of the two groups of parking lots, including data such as vehicle entry and exit time, parking space occupancy, and customer satisfaction survey. These data will be used for subsequent analysis and evaluation.

[0102] Design evaluation indicators: In order to quantify the performance difference between the model recommendation scheme and the existing parking management scheme, the following evaluation indicators are designed: parking efficiency (measured by the average vehicle entry and exit time), parking space utilization (measured by parking space occupancy rate), customer satisfaction (measured by satisfaction survey scores) and parking fees (measured by average parking fees). These indicators can fully reflect the effectiveness of parking management.

[0103] Statistical analysis: The chi-square test method was used to conduct statistical analysis on the data of Group A and Group B on various evaluation indicators. The statistical differences between the two groups in parking efficiency, parking space utilization, customer satisfaction and parking fees were calculated and compared. Assume that in terms of parking efficiency, the average entry and exit time of Group A is 60 seconds, while the average entry and exit time of Group B is 45 seconds; in terms of parking space utilization, the parking space occupancy rate of Group A is 75%, while the parking space occupancy rate of Group B is 85%; in terms of customer satisfaction, the satisfaction score of Group A is 4.0, while the satisfaction score of Group B is 4.5; in terms of parking fees, the average parking fee of Group A is 10 yuan, while the average parking fee of Group B is 8 yuan. Through the chi-square test, it can be concluded that Group B is better than Group A in all evaluation indicators, and the difference is statistically significant.

[0104] The present invention establishes an adaptive feedback mechanism, regularly collects parking lot operation data and user satisfaction survey results, and automatically adjusts the weight coefficient in the reward function. Assume that in the initial operation stage of a smart community parking lot, due to the limited number of parking spaces and large parking demand, the average parking time is long and the user satisfaction is low. Through the adaptive feedback mechanism, after the system detects this problem, it automatically increases the weight coefficient related to parking efficiency and user satisfaction. In subsequent training, the model pays more attention to recommending efficient parking solutions and meeting user needs, thereby gradually reducing the average parking time and improving user satisfaction. With the continuous accumulation of data and the continuous effect of the feedback mechanism, the model performance has been significantly improved, achieving a dual improvement in the efficient operation of the parking lot and user satisfaction.

[0105] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0106] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. An optimization scheduling method for a smart community smart parking lot, characterized in that: The method comprises: Construct an intelligent parking decision model, model the scheduling process of the intelligent parking lot as a Markov decision process, and specifically define the state space, action space, reward and strategy; Design a reward function based on the goals of smart parking lots, including reward mechanisms for increased parking space utilization, faster vehicle entry and exit, improved user satisfaction, and reduced operating costs; Dynamic reward function adjustment: dynamically adjust the weight coefficient in the reward function according to the actual needs of parking lot operations and user feedback, including increasing rewards for environmental protection and energy-saving measures and providing incentives for preferential policies for specific user groups; Train the intelligent parking decision model, select the deep Q network algorithm, perform iterative training on actual data, and tune the hyperparameters; Evaluate the intelligent parking decision model and compare the model recommendation with the existing parking management solution through A / B testing. Real-time scheduling is performed based on the trained intelligent parking decision model, parking lot status information is collected in real time, the optimal action is selected according to the current status and strategy, and the decision results are transmitted to the parking lot management system for execution.

2. The optimization scheduling method of a smart community intelligent parking lot according to claim 1 is characterized by: The state space is represented in the form of a matrix, including the occupancy status of each parking space in the parking lot, historical data of vehicle entry and exit, whether the current time is a peak period, and the current weather conditions; The action space is designed to optimize parking space utilization, improve parking efficiency, and reduce operating costs, including allocating parking spaces, guiding vehicles in and out, adjusting parking space prices, and turning on or off lighting and ventilation systems in specific areas; The strategy is a rule for selecting actions in a given state, and the goal is to maximize the cumulative reward by learning and optimizing the strategy.

3. The optimization scheduling method of a smart community intelligent parking lot according to claim 2 is characterized by: The reward function is designed as R(s, a), where s represents the current state, a represents the action taken, and R is the immediate reward obtained after executing action a in state s; The reward function R(s, a) is composed of multiple weighted sub-reward items, including: Parking space utilization bonus: Among them, N occupied is the number of currently occupied parking spaces, N total is the total number of parking spaces, α is the utilization weight coefficient, β is the set optimal utilization threshold, and a positive reward is given when the utilization rate is higher than β, and a negative reward or zero reward is given when the utilization rate is lower than β; Vehicle entry and exit speed bonus: Among them, T in and T out are the average entry and exit times of vehicles, γ is the speed weight coefficient, and δ is the inverse of the ideal average time threshold set, which encourages reducing vehicle waiting time; User Satisfaction Rewards: Among them, S survey is the user satisfaction survey result, ∈ is the satisfaction weight coefficient, The satisfaction level is the benchmark value, and positive rewards are given if the value is higher than the benchmark value; Operational cost saving bonus: R cost =-η×C; where C is the operating cost caused by the current action, η is the cost saving weight coefficient, and the negative sign indicates that the cost reduction is regarded as a positive reward; The final reward function is the sum of all sub-reward items, that is, R(s, a) = R utilization +R speed +R satisfaction +R cost ; By adjusting the weight coefficients α, γ, ∈, η and the thresholds β, δ, Achieve preference setting and balance optimization for different operational objectives.

4. The optimization scheduling method of a smart community intelligent parking lot according to claim 3 is characterized in that: The steps to train the smart parking decision model include: S1: Collect and preprocess actual parking lot operation data, including vehicle entry and exit records, parking space occupancy status, user feedback, and operation cost information, and clean and preprocess the collected data to form a training data set; S2: A deep Q network is used as the decision model. The model consists of two parts: a deep neural network (DNN) and a Q value estimation. The DNN is used to extract state features and output the Q value estimation of each action. S3: Use the preprocessed training data set to iteratively train the DQN model. During each training, a batch of samples are randomly selected from the data set. The loss is calculated through the loss function and the DNN parameters are updated. At the same time, the target network parameters θ-=θ are regularly updated. The calculation formula of the loss function is: Among them, r is the immediate reward, γ is the discount factor, s′ is the next state transferred to after executing the action, a ′ is the optional action for the next state; θ is the DNN parameter, θ- is the target network parameter, and N is the batch sample size, which is used to optimize the DNN to reduce the gap between the predicted Q value and the true Q value; S4: Evaluate the tuned DQN model on an independent test dataset to verify the effectiveness and practicality of the model by comparing the consistency of the parking decision predicted by the model with the actual optimal decision, user satisfaction, parking space utilization, vehicle entry and exit speed, and operating cost.

5. The optimization scheduling method of a smart community intelligent parking lot according to claim 4 is characterized in that: In step S1, the method of cleaning the collected data includes: Missing value processing: For missing values, linear interpolation algorithm is used to fill them; Denoising: For noisy data, a sliding average filtering algorithm is used for data denoising. The algorithm formula is: Among them, y[n] is the denoised data, x[n] is the original data, and N is the size of the sliding window.

6. The optimization scheduling method of a smart community intelligent parking lot according to claim 4 is characterized in that: In step S1, the method of preprocessing the collected data includes: Standardization: Standardize continuous data to eliminate dimensional differences. Use the Z-score method to standardize the data. Subtract the mean from each numerical data and divide it by its standard deviation so that the processed data conforms to the standard normal distribution. The formula for standardization is: Among them, X is the original data, μ is the mean of the original data, σ is the standard deviation of the original data, and Z is the standardized data; Encoding processing: Perform one-hot encoding conversion on categorical data.

7. The optimization scheduling method for a smart parking lot in a smart community according to claim 1 is characterized in that: The steps to evaluate the performance of the smart parking decision model through the A / B test comparative experimental method include: Collect a certain amount of independent parking lot data as a test set to ensure that the test set has no intersection with the training set data; Design an A / B test plan, where Group A uses the existing parking management plan for parking management, and Group B uses the model recommendation plan for parking management, and the parking lots of Group A and Group B maintain similar distributions in key features; the key features include the number of parking spaces, parking demand, and traffic flow; Conduct tests in a simulated environment to record the parking management process and results of two groups of parking lots; design evaluation indicators, including parking efficiency, parking space utilization, customer satisfaction, and parking fees, to quantify the performance differences between the model-recommended solution and the existing parking management solution; The chi-square test method was used to calculate and compare the statistical differences between group A and group B in various evaluation indicators.

8. The optimization scheduling method for a smart parking lot in a smart community according to claim 1 is characterized in that: The method also includes: establishing an adaptive feedback mechanism, regularly collecting parking lot operation data and user satisfaction survey results, and automatically adjusting the weight coefficient in the reward function.

Images