Park public facility adaptive scheduling and operation and maintenance system based on artificial intelligence
By using an AI-based adaptive scheduling and operation system to dynamically adjust pricing and parking space zoning, the system solves the problems of resource idleness and demand imbalance in existing parking management systems, improves park operation efficiency and user satisfaction, and achieves efficient parking resource management and emergency response.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-01-05
- Publication Date
- 2026-04-17
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
The existing park parking management system cannot adaptively adjust according to real-time traffic flow and historical patterns, resulting in idle resources or excessive demand, low system throughput efficiency, lack of quantitative consideration of user experience, and lack of predictive maintenance and proactive intervention mechanisms, which increases the pressure of operation and maintenance and security risks.
An AI-based adaptive scheduling and operation system is adopted. Through real-time data fusion, predictive models and multi-objective optimization mathematical models, pricing strategies and parking space zoning guidance are dynamically adjusted. Combined with reinforcement learning algorithms, the optimal scheduling scheme is generated to achieve adaptive scheduling of parking resources in time and space.
It increased the park's total parking revenue by 15%-30%, parking space turnover rate by over 20%, vehicle throughput efficiency during peak hours by 25%, reduced the average parking time for users by 40%, shortened the walking distance to core destinations by 25%, and reduced emergency response time to within 5 minutes.
Smart Images
Figure CN121882591A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of public facility management technology, and in particular to an artificial intelligence-based adaptive scheduling and operation and maintenance system for public facilities in industrial parks. Background Technology
[0002] With the acceleration of urbanization, the density of vehicles in parks (including commercial parks, science parks, large campuses, hospitals, etc.) continues to rise, and "difficulty in finding parking spaces, slow parking location search, and chaotic management" have become prominent issues affecting operational efficiency and user experience. To address this challenge, parking management technology has evolved from manual management to intelligent management.
[0003] Currently, mainstream intelligent parking systems primarily rely on Internet of Things (IoT) technology, using devices such as geomagnetic sensors and video parking cameras to achieve real-time monitoring of parking space status and guidance to available spaces. Some advanced systems have also introduced reservation parking and fixed-price zone management functions. These technologies have improved information transparency to some extent and alleviated congestion caused by blindly searching for parking spaces. However, existing management systems still have the following significant defects and limitations. First, the pricing strategies (such as fixed rates per visit or per hour) and zoning rules of existing systems are usually pre-set and remain unchanged over a long period, and cannot be adaptively adjusted according to real-time traffic flow, historical patterns (such as weekday morning rush hour), or special events (such as large conferences or VIP visits). This leads to idle resources during periods of low demand, while during peak periods, the failure of price leverage results in rapid saturation of parking spaces and severe queues, resulting in low overall system throughput efficiency. Secondly, the existing system's operation and maintenance heavily relies on a passive "fault reporting - manual handling" model. For potential equipment failures (such as wear and tear on barrier motors or sensor malfunctions) and sudden operational events (such as illegal parking or obstruction of fire lanes), the system lacks predictive maintenance and proactive intervention mechanisms based on data analysis. This not only increases the pressure of sudden maintenance and safety risks but also reduces the overall lifespan of equipment and system reliability. Furthermore, the existing system's main focus remains on the basic need of "having parking spaces available," lacking quantitative consideration and proactive optimization of user experience. This prevents the system from scientifically and dynamically balancing and optimizing multi-dimensional experience indicators such as "user walking distance," "parking cost," and "parking search time" with economic indicators such as "revenue" and "turnover rate" for operators.
[0004] Therefore, there is an urgent need for an innovative technological solution that can break through the constraints of the existing static management model and achieve truly adaptive, predictive, and global scheduling and operation of parking resources in the time and space dimensions, thereby systematically improving the park's operational efficiency, economic benefits, and user satisfaction. To this end, an artificial intelligence-based adaptive scheduling and operation system for park public facilities is proposed. Summary of the Invention
[0005] The main objective of this invention is to provide an artificial intelligence-based adaptive scheduling and operation and maintenance system for public facilities in parks, which can effectively solve the problems in the background art.
[0006] To achieve the above objectives, the technical solution adopted by the present invention is as follows: An AI-based adaptive scheduling and operation and maintenance method for public facilities in a park, comprising the following steps: S1: Real-time acquisition and fusion of multi-source data from the park's parking system, including at least historical parking space usage data, real-time parking space status data, and real-time demand event information from external input; S2: Based on the historical parking space usage data, use a pre-trained prediction model to predict the parking space demand of each parking zone within a preset future time period. S3: Construct a multi-objective optimization mathematical model with the goal of optimizing the overall operational efficiency of the park's parking system and embedding the real-time demand event information as a constraint; S4: Solve the multi-objective optimization mathematical model to obtain an adaptive scheduling scheme including dynamic pricing strategy, parking space zoning guidance strategy and emergency resource reservation strategy; S5: Execute the adaptive scheduling scheme and collect feedback data after the scheme is executed, which is used to update the prediction model and optimize the parameters of the multi-objective optimization mathematical model; The objective function of the multi-objective optimization mathematical model includes at least: minimizing overall user parking dissatisfaction, maximizing the total revenue of the parking system, and minimizing the penalty for unmet emergency needs; its decision variables include at least: the unit time parking rate of each parking zone at different time periods, the indicator variable for recommending or allocating vehicles to each parking zone, and the state variable for reserving parking spaces for emergency events in each zone.
[0007] Furthermore, the prediction model is a time-series prediction model based on long short-term memory networks. Its input features also include calendar events, weather data, and real-time traffic flow data, and the output is the probability distribution of parking space occupancy in each time period and parking zone in the future.
[0008] Furthermore, the overall user parking dissatisfaction is assessed through dissatisfaction with walking distance. Dissatisfaction with economic costs Dissatisfaction with the time spent searching for mooring and dissatisfaction with forced redistribution The weighted sum is used for quantification, and is expressed as: = + + +
[0009] ;in, = , This refers to the set of vehicles awaiting allocation to parking zones during the current decision-making cycle. Indicates vehicle Assigned to parking zone State parameters, ={0,1}, when When =1, it indicates that the vehicle Assigned to parking zone ,when When =0, it indicates that the vehicle Not assigned to a parking zone , For vehicles From the parking zones marked by the system The distance from the starting point of the walk to its destination. The maximum walking distance is used for normalization. The penalty coefficient is, and >1; = , To perform the maximum value operation, Parking zones Current rates, in yuan / hour. For vehicles The expected rate, The maximum rate is used for normalization; = , Parking zones The number of currently occupied parking spaces. Parking zones Total capacity To enter the parking zone The number of vehicles, For congestion sensitivity coefficient, and >1; = , For vehicles Duration of parking The maximum billing duration is used for normalization. For urgency level indication function, , All are weighting coefficients; , , , All are weighted coefficients.
[0010] Furthermore, the total revenue of the parking system is calculated based on actual demand using the price elasticity coefficient, the parking fee per unit time, and the historical parking space turnover rate of each zone, and is expressed as follows: = ; in, For the total revenue of the parking system, Parking zones During the period Parking fees, in yuan / hour. To perform the minimum value operation, Parking zones Total number of parking spaces Parking zones The benchmark rate, Parking zones under the base rate During the period The predicted number of vehicles for parking demand. Parking zones During the period The price elasticity of demand, Parking zones During the period Historical average parking space turnover rate, unit: number of vehicles / hour / parking space The total number of time periods in a day. This represents the total number of parking zones.
[0011] Furthermore, the constraints also include: A capacity constraint used to limit the number of vehicles allocated to any parking zone during any given time period to not exceed its available capacity; Demand allocation constraint used to require all vehicles entering the park to be assigned to a single zone; Price constraints used to ensure that parking rates are within a reasonable range; Emergency reservation constraints are used to ensure that sufficient parking spaces are reserved for emergency vehicles within a specified time period and area. And, logical constraints that limit a parking space to be allocated to only one vehicle at a time.
[0012] Furthermore, based on the quantitative definition of overall user parking dissatisfaction, the calculation method of total parking system revenue, and the specific content of the constraints, the objective function of the multi-objective optimization mathematical model is defined as:
[0013] ;in, , , All are weighted coefficients. It is a large positive number. To characterize a specified region During the period For emergency vehicles Reserved state parameters, ={0,1}, when =0 indicates the specified area During the period Not an emergency vehicle Reserved, when =1 indicates the specified area During the period For emergency vehicles Reserved; The constraints of the multi-objective optimization mathematical model are expressed as follows: Capacity constraints: , Indicates vehicle During the period Assigned to parking zone State parameters, ={0,1}, when When =1, it indicates that the vehicle During the period Assigned to parking zone ,when When =0, it indicates that the vehicle During the period Not assigned to a parking zone , For the time period Parking zones Available capacity For time period The set of all vehicles to be assigned within the scope; Demand allocation constraints: ; Price constraints: , , These are the minimum and maximum parking fee rates, respectively. Emergency reserve constraints: , To be in the specified area For emergency vehicles The reserved parking spaces are gathered together. For the time period All emergency vehicles awaiting allocation within the area were assembled. For emergency vehicles The specified time period For vehicles During the period Assigned to a designated area parking spaces State parameters, ={0,1}, when When =1, it indicates that the vehicle During the period Assigned to a designated area parking spaces ,when When =0, it indicates that the vehicle During the period Not assigned to the designated area parking spaces ; Logical constraints: .
[0014] Furthermore, the solution to the multi-objective optimization mathematical model is achieved through a reinforcement learning algorithm, specifically as follows: Using the parking lot status as the state space, price adjustment and parking space allocation actions as the action space, and quantitative indicators of comprehensive operational efficiency as reward signals, the agent is trained to learn the optimal scheduling strategy. The state space includes at least the real-time occupancy rate of each partition, the demand forecast value, and the emergency event status; The reward signal is a weighted value of the combined parking revenue, average parking search time, user dissatisfaction index, and emergency response compliance reward items, expressed as: = - - + In the formula, As a reward signal, For parking revenue, The average search time, User dissatisfaction index This is an emergency response compliance reward item. , , , All are weighted coefficients. For time periods.
[0015] Furthermore, when the real-time demand event information indicates a priority vehicle visit, the generation of the emergency resource reservation strategy includes the following steps: Based on the event time, number of parking spaces, and location requirements, the system automatically searches within the target parking zone to determine the optimal set of parking spaces to be reserved. If there are occupied parking spaces in the optimal parking space set, a vehicle relocation request and compensation plan will be sent to the owner of the occupied vehicle, and a new parking space will be automatically allocated to him in the alternative zone. Before the incident occurs, the smart parking locks on the optimal parking space set can be remotely controlled to enter the locked state or the reserved indicator light can be lit.
[0016] Furthermore, the implementation of the adaptive scheduling scheme specifically includes: synchronizing the dynamic pricing strategy to the parking fee system, publishing the parking space zoning guidance strategy through variable information signs and / or user mobile terminals, and controlling the physical status of relevant parking spaces according to the emergency resource reservation strategy.
[0017] An AI-based adaptive scheduling and operation and maintenance system for park parking systems, used to implement an AI-based adaptive scheduling and operation and maintenance method for park parking systems, includes: The data sensing and aggregation module is used to collect the multi-source data; The prediction and optimization decision module integrates the prediction model and the reinforcement learning algorithm, and is used to execute steps S2 to S4 to generate the adaptive scheduling scheme. The strategy execution and interaction module is used to distribute the scheduling plan to the parking lot terminal equipment and interact with users. The simulation verification and learning closed-loop module is used to verify the scheduling scheme in a digital twin environment and continuously optimize the model using feedback data. The strategy execution and interaction module includes an intelligent parking lock control unit, a variable information sign management unit, and a user mobile terminal service unit; the user mobile terminal service unit is used to provide users with functions such as parking space reservation, dynamic price viewing, optimal route guidance, vehicle relocation request reception, and compensation collection.
[0018] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements an artificial intelligence-based adaptive scheduling and operation and maintenance method for a park parking system.
[0019] The present invention has the following beneficial effects: Compared with existing technologies, this solution uses a dynamic pricing model based on price elasticity demand to automatically find the optimal revenue price point for different time periods and zones, avoiding demand imbalance under fixed prices. During peak demand periods, moderate price increases can suppress non-urgent demand and increase revenue per vehicle; during off-peak periods, moderate price reductions can attract customers and reduce vacancy. Simulation results show that under the same traffic conditions, this system can increase the total parking revenue of the park by 15%-30%.
[0020] Compared with existing technologies, this solution utilizes big data technology, with turnover rate as the key optimization target. Through dynamic price signals and precise guidance, it directs long-term parked vehicles to the edge area, reserving high-turnover parking spaces in the core area. This increases the average daily turnover rate of parking spaces in the core parking area by more than 20%, improves the overall parking lot's peak-hour vehicle throughput efficiency by 25%, and significantly reduces entrance queue length and waiting time.
[0021] Compared with existing technologies, this solution can quickly generate the best global plan when facing emergencies such as VIP visitors and fire emergency rescue. This greatly reduces the time required for parking space preparation and traffic management in special events from an average of more than 30 minutes for manual coordination to less than 5 minutes, thereby ensuring zero delays in critical emergency tasks.
[0022] Compared with existing technologies, this solution uses quantified user dissatisfaction as an optimization target. When recommending parking spaces to each user, it systematically reduces the combined burden of walking distance and parking search time, resulting in a 40% reduction in average parking search time and a 25% reduction in average walking distance to core destinations, thereby improving user satisfaction. Attached Figure Description
[0023] Figure 1 This is a flowchart illustrating the adaptive scheduling and operation and maintenance method for public facilities in a park based on artificial intelligence, as described in this invention. Figure 2 This is a schematic diagram of the multi-source data acquisition and processing flow of the present invention; Figure 3 This is a schematic diagram of one form of the prediction model of the present invention; Figure 4 This is a schematic diagram of the adaptive scheduling and operation and maintenance system for public facilities in a park based on artificial intelligence, as described in this invention. Detailed Implementation
[0024] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.
[0025] Example 1: See Figure 1 The flowchart shown is a method for adaptive scheduling and operation and maintenance of public facilities in a park based on artificial intelligence, according to the present invention, which includes the following steps: Step 1: Real-time acquisition and fusion of multi-source data from the park's parking system The multi-source data includes historical parking space usage data, real-time parking space status data, and real-time demand event information input from external sources. Specifically, in one possible implementation, the sources and acquisition methods of various types of data are described in [reference needed]. Figure 2 The technical solution flowchart shown can be followed in the following steps: Step 1.1: Historical Parking Space Usage Data Source systems: Parking fee management system, historical database of parking space sensors.
[0026] How to obtain: API interface synchronization: Through the standardized data interface (such as RESTful API) provided by the parking management system, historical transaction records, vehicle entry and exit flow data, and parking space occupancy change logs are retrieved periodically (such as at midnight every day) or in real time.
[0027] Direct database connection: Under authorized and secure conditions, data is extracted from a backup database or data warehouse of the production database via data middleware (such as Kettle or DataX). Key fields include: 1) timestamp 2) space id (Parking space number) occupancy status (Occupied status: 0 / 1) vehicle id (License plate number, after anonymization) duration (parking time) fee (amount charged) zone id (Partition) Step 1.2: Real-time parking space status data Source device layer: Video parking camera / video stake: It acquires real-time video streams via RTSP or SDK, and uses built-in AI algorithms to identify parking space status (empty / occupied) and license plate number in real time, generating structured data reports.
[0028] Geomagnetic / ultrasonic sensors: Using IoT protocols such as LoRa and NB-IoT, they transmit the detected changes in parking space status to the IoT platform in real time.
[0029] Barrier gate / entry / exit camera: Captures vehicle entry and exit events, license plates, and timestamps, and uploads them in real time via TCP / IP protocol.
[0030] How to obtain: IoT platform convergence: All IoT devices are connected to a unified IoT platform (such as Alibaba Cloud IoT or Huawei OceanConnect). The platform manages devices, parses protocols, and standardizes data. Then, it pushes real-time data streams to the AI platform through message queues (such as Kafka and MQTT).
[0031] Edge computing preprocessing: Deploy an edge computing gateway in the parking lot to perform local analysis on the video stream, and only upload the structured recognition results (rather than the original video), which greatly reduces bandwidth consumption and cloud processing pressure.
[0032] Step 1.3: Real-time demand event information from external input Source and input method: Manual data entry on the park management platform: Administrators create events in the form of structured forms through the web management backend or mobile app.
[0033] Event Type: VIP Visit / Large Conference / Fire Drill / Facility Maintenance (Select from dropdown menu) Time: Start time, End time (selectable via time control) Requirements: Number of reserved parking spaces, designated area, list of license plate numbers (can be uploaded using OCR recognition). Priority: High / Medium / Low (Single choice) Third-party system API integration: Visitor appointment system: Automatically synchronizes appointment records and extracts visitor vehicle information, visited units, and estimated arrival time.
[0034] Enterprise calendar / meeting system: Synchronize large meeting schedules and predict the number of vehicles attending meetings.
[0035] Emergency command platform: Receives emergency instructions from fire, public security and other departments.
[0036] Natural Language Processing Interface: Receives text information from telephone recordings and customer service chatbots, automatically extracts event elements through NLP models (such as entity recognition), and transforms them into structured instructions.
[0037] Step 1.4: Fusion of multi-source data By constructing a unified spatiotemporal data model, and performing correlation, completion, and semantic enhancement under a unified spatiotemporal benchmark, a panoramic view of "parking space-vehicle-event" is formed.
[0038] Core converged technology architecture: based on data middle platform and integrated stream and batch processing Access Layer (Unified Entry Point): Set up an API gateway to handle all external system requests.
[0039] Deploy a message queue cluster (such as Apache Kafka) as the "central nervous system" of real-time data, receiving all data streams from the IoT platform and management platform.
[0040] Processing layer (stream and batch processing): Real-time stream processing: Clean, transform, and correlate real-time data streams using Flink or Spark Streaming.
[0041] Example: Correlate an event stream of "Geomagnetic sensor reports that parking space A is occupied" with an event stream of "Entrance camera reports that license plate Jing A12345 enters" within a 3-second time window according to location logic to obtain an accurate record of "License plate Jing A12345 parks in parking space A".
[0042] Batch data processing: Use Spark or Hive to perform offline mining on historical data, calculate features (such as historical turnover rate, user preferences), and use them for model training.
[0043] Fusion logic (key algorithms and rules): Spatio-temporal alignment: All data is tagged with a unified timestamp (UTC) and location label (parking space GPS coordinates or logical partition encoding).
[0044] Vehicle trajectory stitching: Use the multiple hypothesis tracking algorithm to combine discrete entry and exit events and capture records from different cameras, and combine with the parking lot topology map to reconstruct the complete driving and parking trajectory of the vehicle in the park.
[0045] Event-parking space association: When receiving an event instruction of "Reserve 5 parking spaces in area A from 14:00 to 16:00", the system immediately queries the current and predicted future occupancy status of area A.
[0046] Through the combinatorial optimization algorithm, select a set of 5 parking spaces that are continuously available in time, as adjacent as possible in space, and have the least impact on the current user from all parking spaces in area A, and strongly bind this event ID to these 5 parking space IDs.
[0047] Storage layer (unified data model after fusion): Real-time storage: Use Redis or in-memory database to store the latest parking space status, currently effective events, and dynamic price list for millisecond-level queries.
[0048] Historical details storage: Use a distributed database (such as HBase / Cassandra) or a time-series database (such as InfluxDB) to store the full volume of detailed data after fusion for analysis and playback.
[0049] Feature and label storage: Use a feature platform to store standardized features (such as "Occupancy change sequence of parking space A in the past 1 hour") prepared for AI models.
[0050] Step 2: Based on historical parking space usage data, use a pre-trained prediction model to predict the parking space demand in each parking area within a preset future period.
[0051] The prediction model is a time-series prediction model based on long short-term memory networks. Its input features include calendar events, weather data, and real-time traffic flow data. The output is the probability distribution of parking space occupancy in each time period and parking zone in the future.
[0052] In one possible implementation, the prediction model can be defined as follows: =
[0053] ;in, For the output of the prediction model, , , , These include historical parking space usage data, calendar events, weather data, and real-time traffic flow data.
[0054] This embodiment presents one possible structural form of the prediction model; see [link to relevant documentation]. Figure 3 As shown: Model input: Historical multi-source time-series features + external features (calendar events, weather data, and real-time traffic flow data) Model output: The probability distribution of parking space occupancy in each parking zone (K zones) within a future preset time period (e.g., the next 24 hours, with an interval of 15 minutes), i.e., the predicted occupancy rate (between 0 and 1).
[0055] Specifically, the model structure includes: 1) Input layer and feature engineering The model receives two parallel time series inputs (both aligned to the same future forecast period): A. Main time-series feature sequence (historical parking space data): Shape: [lookback] steps , K, feature dim ] lookback steps Lookback step size (e.g., one point per hour over the past 48 hours). steps =48) K: Number of partitions feature dim Basic feature dimensions, including: occupancy rate Historical parking space occupancy rate (0-1) inflow / outflow: Vehicle inflow / outflow rates for each zone cumulative occupancy Cumulative number of occupancy B. External Feature Sequence: Shape: [lookback] steps + forecast steps external feature-dim ] forecast steps Predicted step size (e.g., the next 24 hours) external feature-dim External feature dimensions, including: Time characteristics (One-Hot or Sine coding): hour_of_day, day_of_week, is_weekend, is_holiday Calendar events (numerical): event indicator Are there any large conferences / events? (0 / 1) event scale Event size (estimated number of attendees, normalized) event location Event occurrence partition (K-dimensional One-Hot vector) Weather data (from the Weather API): temperature, precipitation, weather type (Sunny / Rainy / Snowy, etc.) Real-time traffic flow data (from traffic management platforms or map APIs): road_congestion_index: Congestion index (0-1) of the main roads leading to the park. estimated_travel_time: Estimated travel time from the main entrance to the park (ratio to the baseline time). 2) Core Network Module Ⅰ: External Feature Embedding Network (processing input B): Objective: To encode high-dimensional, heterogeneous external features into a low-dimensional, dense context vector.
[0056] structure: Fully connected embedding layer: embeds categorical features (such as weather type, day of the week).
[0057] Feature concatenation: concatenate all numerical values and the embedded features.
[0058] Temporal fully connected layers: Use 1D convolutions or fully connected layers to process along the time axis and capture the changing patterns of external features.
[0059] Output: A shape of [lookback] is obtained. steps + forecast steps embedding dim The external context sequence of ].
[0060] II: Multi-level LSTM encoder (processes input A and fuses external features): Objective: To encode the spatiotemporal dependence of historical parking space data and dynamically introduce the influence of external factors.
[0061] Structure (Innovation points: spatiotemporal separation and feature gating): Spatial attention layer (optional but recommended): Before inputting into the LSTM, the features of the K partitions are weighted, so that the model focuses on other partitions that are more relevant to the target partition.
[0062] First layer LSTM (time-dependent modeling): Use a gated cyclic unit or a standard LSTM.
[0063] At each time step t, the input is the feature vectors of all K partitions at the current time.
[0064] Output the hidden state of each partition. .
[0065] Feature fusion gating mechanism: At each time step of the LSTM, the vector e corresponding to the time step in the external context sequence is... t Introduction.
[0066] Implementation method: e can be used t After concatenating with parking space features, input into an LSTM; or for more precise processing, use e t This can be used to modulate the input gate or forget gate (conditional gating) of the LSTM, allowing external information to directly influence the updating of memory cells. For example: = ; Second-layer LSTM (spatial dependency and higher-order temporal modeling): Hide the states of each partition output by the first LSTM layer. Reorganize by partition dimension.
[0067] Use another set of LSTMs (or ConvLSTMs) to encode along the partition dimension or a deeper time dimension to capture the inter-partition interactions (such as overflow to adjacent partitions when one partition is full) and more complex time patterns.
[0068] III: Decoding and Prediction Output Layer: State initialization: The hidden state of the encoder at the last time step is used as the initial state of the decoder.
[0069] Autoregressive decoding (or direct multi-step prediction): For multi-step prediction, a fully connected layer is used to directly map the final fused features to the [forecast] layer. steps The shape of [K].
[0070] Output layer: The final feature is mapped to the predicted occupancy (probability distribution) of each partition through a Dense(K) + Sigmoid layer.
[0071] Output shape: [forecast] steps [, K], where each value is between 0 and 1.
[0072] Loss function: Since the output is a probability, a weighted mean squared error is used to give higher weight to the prediction error during peak periods in order to optimize the prediction accuracy during critical periods. Specifically: =
[0073] ;in, The weight is determined by the time period. It can be set to 2.0 during peak hours (such as the morning of a weekday) and 1.0 during off-peak hours.
[0074] Step 3: Construct a multi-objective optimization mathematical model with the goal of optimizing the overall operational efficiency of the park's parking system and embedding real-time demand event information as constraints.
[0075] in, The objective function of a multi-objective optimization mathematical model should include at least the following: minimizing overall user parking dissatisfaction, maximizing the total revenue of the parking system, and minimizing the penalty for unmet emergency needs. The decision variables include at least: the parking rate per unit time for each parking zone at different times, the indicator variables for recommending or allocating vehicles to each parking zone, and the status variables for reserving parking spaces for emergency events in each zone.
[0076] The constraints also include: A capacity constraint used to limit the number of vehicles allocated to any parking zone during any given time period to not exceed its available capacity; Demand allocation constraint used to require all vehicles entering the park to be assigned to a single zone; Price constraints used to ensure that parking rates are within a reasonable range; Emergency reservation constraints are used to ensure that sufficient parking spaces are reserved for emergency vehicles within a specified time period and area. And, logical constraints that limit a parking space to be allocated to only one vehicle at a time.
[0077] Overall user dissatisfaction with parking was assessed based on dissatisfaction with walking distance. Dissatisfaction with economic costs Dissatisfaction with the time spent searching for mooring and dissatisfaction with forced redistribution The weighted sum is used for quantification, and is expressed as: = + + +
[0078] ;in, = , This refers to the set of vehicles awaiting allocation to parking zones during the current decision-making cycle. Indicates vehicle Assigned to parking zone State parameters, ={0,1}, when When =1, it indicates that the vehicle Assigned to parking zone ,when When =0, it indicates that the vehicle Not assigned to a parking zone , For vehicles From the parking zones marked by the system The distance from the starting point of the walk to its destination. The maximum walking distance is used for normalization. The penalty coefficient is, and >1; = , To perform the maximum value operation, Parking zones Current rates, in yuan / hour. For vehicles The expected rate, The maximum rate is used for normalization; = , Parking zones The number of currently occupied parking spaces. Parking zones Total capacity To enter the parking zone The number of vehicles, For congestion sensitivity coefficient, and >1; = , For vehicles Duration of parking The maximum billing duration is used for normalization. For urgency level indication function, , All are weighting coefficients; , , , All are weighted coefficients.
[0079] The total revenue of the parking system is calculated based on actual demand using the price elasticity coefficient, the parking fee per unit time, and the historical parking space turnover rate of each zone, and is expressed as follows: =
[0080] ;in, For the total revenue of the parking system, Parking zones During the period Parking fees, in yuan / hour. To perform the minimum value operation, Parking zones Total number of parking spaces Parking zones The benchmark rate, Parking zones under the base rate During the period The predicted number of vehicles for parking demand. Parking zones During the period The price elasticity of demand, Parking zones During the period Historical average parking space turnover rate, unit: number of vehicles / hour / parking space The total number of time periods in a day. This represents the total number of parking zones.
[0081] Based on the quantitative definition of overall user parking dissatisfaction, the calculation method of total parking system revenue, and the specific content of the constraints, the objective function of the multi-objective optimization mathematical model is defined as:
[0082] ;in, , , All are weighted coefficients. It is a large positive number. To characterize a specified region During the period For emergency vehicles Reserved state parameters, ={0,1}, when =0 indicates the specified area During the period Not an emergency vehicle Reserved, when =1 indicates the specified area During the period For emergency vehicles Reserved; The constraints of the multi-objective optimization mathematical model are expressed as follows: Capacity constraints: , Indicates vehicle During the period Assigned to parking zone State parameters, ={0,1}, when When =1, it indicates that the vehicle During the period Assigned to parking zone ,when When =0, it indicates that the vehicle During the period Not assigned to a parking zone , For the time period Parking zones Available capacity For time period The set of all vehicles to be assigned within the scope; Demand allocation constraints: ; Price constraints: , , These are the minimum and maximum parking fee rates, respectively. Emergency reserve constraints: , To be in the specified area For emergency vehicles The reserved parking spaces are gathered together. For the time period All emergency vehicles awaiting allocation within the area were assembled. For emergency vehicles The specified time period For vehicles During the period Assigned to a designated area parking spaces State parameters, ={0,1}, when When =1, it indicates that the vehicle During the period Assigned to a designated area parking spaces ,when When =0, it indicates that the vehicle During the period Not assigned to the designated area parking spaces ; Logical constraints: .
[0083] Step 4: Solve the multi-objective optimization mathematical model using reinforcement learning algorithms to obtain an adaptive scheduling scheme that includes dynamic pricing strategy, parking space zoning guidance strategy, and emergency resource reservation strategy.
[0084] Specifically, when the real-time demand event information indicates the arrival of a priority vehicle, the generation of the emergency resource reservation strategy includes the following steps: Based on the event time, number of parking spaces, and location requirements, the system automatically searches within the target parking zone to determine the optimal set of parking spaces to be reserved. If there are occupied parking spaces in the optimal parking space set, a request to move the vehicle and a compensation plan will be sent to the owner of the occupied vehicle, and a new parking space will be automatically allocated to him in the alternative zone. Before the incident occurs, the smart parking locks on the optimal parking space set can be remotely controlled to enter the locked state or the reserved indicator light can be lit.
[0085] The specific steps for solving multi-objective optimization mathematical models using reinforcement learning algorithms are as follows: Using the parking lot status as the state space, price adjustment and parking space allocation actions as the action space, and quantitative indicators of comprehensive operational efficiency as reward signals, the agent is trained to learn the optimal scheduling strategy. The state space includes at least the real-time occupancy rate of each partition, the demand forecast, and the status of emergency events. The reward signal is a weighted value of the combined parking revenue, average parking search time, user dissatisfaction index, and emergency response compliance rewards, expressed as: = - - + In the formula, As a reward signal, For parking revenue, The average search time, User dissatisfaction index This is an emergency response compliance reward item. , , , All are weighted coefficients. For time periods.
[0086] Specifically, the solution process can be carried out using the following steps: Step 4.1: Parameter Definition Time: Discretize a day into T time periods, t∈{1,2,...,T}; Parking Area: The parking lot is divided into K areas (such as core area, buffer zone, and edge area), k∈ {1,2,...,K}; Parking spaces: The set of parking spaces within region k is S. k The total number of parking spaces is N k ; Vehicle type: Frequent riders: Registered, relatively insensitive to price, and prefer fixed areas.
[0087] Temporary visitors: sensitive to price and walking distance, with highly flexible behavior.
[0088] Priority vehicles: Designated vehicles for emergency events (VIP, ambulance, fire), which must be absolutely guaranteed.
[0089] Step 4.2: Decomposition and Iteration: Outer layer (pricing optimization): With a fixed vehicle allocation x, optimize the price p. Since the demand function is differentiable, gradient descent can be used.
[0090] Inner layer (allocation optimization): With a fixed price p, optimize vehicle allocation x. This is a constrained allocation problem, which can be solved using an integer programming solver or a variant of the Hungarian algorithm.
[0091] Two iterations are performed until convergence.
[0092] Step 4.3: Reinforcement Learning Methods (More Suitable for Dynamic Environments): State S t : Current occupancy rate, demand forecast, priority event status, and current price for each region.
[0093] Action A t Price adjustment vector Δp t And allocation suggestions.
[0094] Reward R tThe weighted average of parking revenue, average parking search time, user dissatisfaction index, and emergency response compliance rewards is expressed as: = - - + In the formula, As a reward signal, For parking revenue, The average search time, User dissatisfaction index This is an emergency response compliance reward item. , , , All are weighted coefficients. For a period of time; Training: Using algorithms such as Deep Deterministic Policy Gradient (DDPG) or Proximal Policy Optimization (PPO), the agent learns a policy that maximizes long-term cumulative rewards under complex constraints.
[0095] Step 4.4: Simulation-driven optimization: Establish a digital twin simulation environment for the parking system to simulate vehicle arrival, parking, departure, and driver responses to pricing and guidance.
[0096] The above model was used as a strategy and run in simulation to evaluate its long-term effects.
[0097] Use Bayesian optimization or genetic algorithms to search the policy parameter space to find the combination that performs best in simulation.
[0098] Step 4.5: Model Output and Scheduled Execution After solving the model, the system will obtain: Optimal pricing strategy { → Publish to the payment system.
[0099] Optimal allocation scheme { → This translates into guidance instructions for newly arriving vehicles ("Please proceed to Zone B") and suggestions for moving parked vehicles ("A new parking space has been reserved for you in Zone D. You can receive a discount for moving your vehicle").
[0100] Reserved plan { → Control the raising and lowering of the smart parking lock and update the parking space status.
[0101] Step 5: Execute the adaptive scheduling scheme and collect feedback data after the scheme is executed, which is used to update the prediction model and optimize the parameters of the multi-objective optimization mathematical model.
[0102] Specifically, this includes: synchronizing dynamic pricing strategies to the parking fee system, publishing parking space zoning guidance strategies through variable message signs and / or user mobile terminals, and controlling the physical status of relevant parking spaces according to emergency resource reservation strategies.
[0103] The technical effects of the present invention will be illustrated below with examples of several specific application scenarios: Scenario 1: Weekday morning rush hour (optimized for routine use) Forecast: The model predicts that demand in the core area will be extremely high from 8:30 to 9:30.
[0104] Action taken: Starting at 7:45, the parking fare in the core area will increase by 30%, and a notification will be sent via the app: "The core area is busy. We recommend that you park in the buffer zone, a 5-minute walk away, where you will receive a 40% discount." Results: Turnover rate in the core area increased by 15%, overall traffic flow distribution was more balanced, and queuing time at entrances was reduced.
[0105] Scenario 2: Emergency Fire Drill (Emergency Dispatch) Input: The administrator marked "The fire exit area (Zone C) must be completely cleared within 15 minutes".
[0106] action: The system immediately locked area C, prohibiting new vehicles from entering.
[0107] Send the highest priority vehicle relocation alert (App push + SMS) to all parked vehicles in Zone C.
[0108] These vehicles will be automatically assigned spare parking spaces in Zone D, and the parking will be free this time.
[0109] A large screen guides you to move your car.
[0110] Result: The area was cleared within 10 minutes, ensuring that emergency access routes remained unobstructed.
[0111] Scenario 3: Multiple teams visit simultaneously (hybrid scheduling) Forecast and Input: The system predicts a regular peak in the afternoon and has received reservations for "Company A's tour group (10 vehicles)" and "Client B's negotiation (3 VIP vehicles)".
[0112] action: Reserve the best parking spaces in the core area for 3 VIP vehicles.
[0113] The 10 vehicles of Company A's tour group were guided to the reserved "group bus area" and adjacent buffer zone parking spaces, and a group parking discount was offered.
[0114] For regular visitors, fine-tune the pricing to guide them away from these reserved clusters and avoid localized congestion.
[0115] Result: Despite the tight overall capacity, key receptions were ensured and the overall operational order was maintained.
[0116] Example 2: This invention also provides an AI-based adaptive scheduling and operation and maintenance system for a park parking system to implement the above method, see [link to relevant documentation]. Figure 4 The system architecture diagram shown includes: The data sensing and aggregation module is used to collect data from multiple sources; The prediction and optimization decision module, which integrates a prediction model and a reinforcement learning algorithm, is used to execute steps S2 to S4 to generate an adaptive scheduling scheme. The strategy execution and interaction module is used to distribute the scheduling plan to the parking lot terminal equipment and interact with users. The simulation verification and learning closed-loop module is used to verify the scheduling scheme in a digital twin environment and continuously optimize the model using feedback data. The strategy execution and interaction module includes an intelligent parking lock control unit, a variable information sign management unit, and a user mobile terminal service unit; The user mobile terminal service unit is used to provide users with functions such as parking space reservation, dynamic price viewing, optimal route guidance, receiving vehicle relocation requests, and receiving compensation.
[0117] Example 3: The present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, can realize the above-mentioned adaptive scheduling and operation and maintenance method for a park parking system based on artificial intelligence.
[0118] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of this invention is defined by the appended claims and their equivalents.
Claims
1. An AI-based adaptive scheduling and operation and maintenance method for public facilities in a park, characterized in that: The method includes the following steps: S1: Real-time acquisition and fusion of multi-source data from the park's parking system, including at least historical parking space usage data, real-time parking space status data, and real-time demand event information from external input; S2: Based on the historical parking space usage data, use a pre-trained prediction model to predict the parking space demand of each parking zone within a preset future time period. S3: Construct a multi-objective optimization mathematical model with the goal of optimizing the overall operational efficiency of the park's parking system and embedding the real-time demand event information as a constraint; S4: Solve the multi-objective optimization mathematical model to obtain an adaptive scheduling scheme including dynamic pricing strategy, parking space zoning guidance strategy and emergency resource reservation strategy; S5: Execute the adaptive scheduling scheme and collect feedback data after the scheme is executed, which is used to update the prediction model and optimize the parameters of the multi-objective optimization mathematical model; The objective function of the multi-objective optimization mathematical model includes at least: minimizing overall user parking dissatisfaction, maximizing the total revenue of the parking system, and minimizing the penalty for unmet emergency needs; its decision variables include at least: the unit time parking rate of each parking zone at different time periods, the indicator variable for recommending or allocating vehicles to each parking zone, and the state variable for reserving parking spaces for emergency events in each zone.
2. The adaptive scheduling and operation and maintenance method for park public facilities based on artificial intelligence according to claim 1, characterized in that, In step S2, the prediction model is a time-series prediction model based on long short-term memory networks. Its input features also include calendar events, weather data, and real-time traffic flow data. The output is the probability distribution of parking space occupancy in each time period and parking zone in the future.
3. The adaptive scheduling and operation and maintenance method for park public facilities based on artificial intelligence according to claim 1, characterized in that, The overall user parking dissatisfaction is measured by dissatisfaction with walking distance. Dissatisfaction with economic costs Dissatisfaction with the time spent searching for mooring and dissatisfaction with forced redistribution The weighted sum is used for quantification, and is expressed as: = + + + ； in, = , This refers to the set of vehicles awaiting allocation to parking zones during the current decision-making cycle. Indicates vehicle Assigned to parking zone State parameters, ={0,1}, when When =1, it indicates that the vehicle Assigned to parking zone ,when When =0, it indicates that the vehicle Not assigned to a parking zone , For vehicles From the parking zones marked by the system The distance from the starting point of the walk to its destination. The maximum walking distance is used for normalization. The penalty coefficient is, and >1; = , To perform the maximum value operation, Parking zones Current rates, in yuan / hour. For vehicles The expected rate, The maximum rate is used for normalization; = , Parking zones The number of currently occupied parking spaces. Parking zones Total capacity To enter the parking zone The number of vehicles, For congestion sensitivity coefficient, and >1; = , For vehicles Duration of parking The maximum billing duration is used for normalization. For urgency level indication function, , All are weighting coefficients; , , , All are weighted coefficients.
4. The adaptive scheduling and operation and maintenance method for park public facilities based on artificial intelligence according to claim 1, characterized in that, The total revenue of the parking system is calculated based on actual demand using the price elasticity coefficient, the parking fee per unit time, and the historical parking space turnover rate of each zone, and is expressed as follows: ,in, For the total revenue of the parking system, Parking zones During the period Parking fees, in yuan / hour. To perform the minimum value operation, Parking zones Total number of parking spaces Parking zones The benchmark rate, Parking zones under the base rate During the period The predicted number of vehicles for parking demand. Parking zones During the period The price elasticity of demand, Parking zones During the period Historical average parking space turnover rate, unit: number of vehicles / hour / parking space The total number of time periods in a day. This represents the total number of parking zones.
5. The adaptive scheduling and operation and maintenance method for park public facilities based on artificial intelligence according to claim 1, characterized in that, In step S4, solving the multi-objective optimization mathematical model is achieved through a reinforcement learning algorithm, specifically as follows: Using the parking lot status as the state space, price adjustment and parking space allocation actions as the action space, and quantitative indicators of comprehensive operational efficiency as reward signals, the agent is trained to learn the optimal scheduling strategy. The state space includes at least the real-time occupancy rate of each partition, the demand forecast value, and the emergency event status; The reward signal is a weighted value of the combined parking revenue, average parking search time, user dissatisfaction index, and emergency response compliance reward items, expressed as: = - - + In the formula, As a reward signal, For parking revenue, The average search time, User dissatisfaction index This is an emergency response compliance reward item. , , , All are weighted coefficients. For time periods.
6. The adaptive scheduling and operation and maintenance method for park public facilities based on artificial intelligence according to claim 1, characterized in that, When the real-time demand event information indicates a priority vehicle visit, the generation of the emergency resource reservation strategy includes the following steps: Based on the event time, number of parking spaces, and location requirements, the system automatically searches within the target parking zone to determine the optimal set of parking spaces to be reserved. If there are occupied parking spaces in the optimal parking space set, a vehicle relocation request and compensation plan will be sent to the owner of the occupied vehicle, and a new parking space will be automatically allocated to him in the alternative zone. Before the incident occurs, the smart parking locks on the optimal parking space set can be remotely controlled to enter the locked state or the reserved indicator light can be lit.
7. The adaptive scheduling and operation and maintenance method for public facilities in a park based on artificial intelligence according to claim 1, characterized in that, In step S5, the implementation of the adaptive scheduling scheme specifically includes: synchronizing the dynamic pricing strategy to the parking fee system, publishing the parking space zoning guidance strategy through variable information signs and / or user mobile terminals, and controlling the physical status of relevant parking spaces according to the emergency resource reservation strategy.
8. The adaptive scheduling and operation and maintenance method for public facilities in a park based on artificial intelligence according to claim 1, characterized in that, The constraints also include: A capacity constraint used to limit the number of vehicles allocated to any parking zone during any given time period to not exceed its available capacity; Demand allocation constraint used to require all vehicles entering the park to be assigned to a single zone; Price constraints used to ensure that parking rates are within a reasonable range; Emergency reservation constraints are used to ensure that sufficient parking spaces are reserved for emergency vehicles within a specified time period and area. And, logical constraints that limit a parking space to be allocated to only one vehicle at a time.
9. An AI-based adaptive scheduling and operation and maintenance system for a park parking system, used to implement the method as described in any one of claims 1-8, characterized in that, include: The data sensing and aggregation module is used to collect the multi-source data; The prediction and optimization decision module integrates the prediction model and the reinforcement learning algorithm, and is used to execute steps S2 to S4 to generate the adaptive scheduling scheme. The strategy execution and interaction module is used to distribute the scheduling plan to the parking lot terminal equipment and interact with users. The simulation verification and learning closed-loop module is used to verify the scheduling scheme in a digital twin environment and continuously optimize the model using feedback data. The strategy execution and interaction module includes an intelligent parking lock control unit, a variable information sign management unit, and a user mobile terminal service unit; the user mobile terminal service unit is used to provide users with functions such as parking space reservation, dynamic price viewing, optimal route guidance, vehicle relocation request reception, and compensation collection.
10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the AI-based adaptive scheduling and operation and maintenance method for a park parking system as described in any one of claims 1-8.