Intelligent property parking lot energy consumption optimization and management system based on big data analysis

Abstract

This invention discloses a smart property parking lot energy consumption optimization and management system based on big data analysis, belonging to the field of smart property and energy management technology. The system includes thermal imaging sensors and environmental sensors deployed in the parking lot, as well as multiple intelligent agents corresponding to lighting, ventilation, and parking space lock equipment. These intelligent agents form a distributed intelligent agent network through a wireless mesh network. Intelligent agents in adjacent physical spaces establish local communication connections. Based on multimodal perception data and local observation information broadcast by neighboring intelligent agents, each intelligent agent autonomously makes decisions and executes corresponding control actions through a pre-trained multi-agent reinforcement learning model, while simultaneously achieving information interaction and collaborative linkage across subsystems. This invention enables refined collaborative control of parking lot energy consumption, effectively reducing operating energy consumption and costs while ensuring operational safety and service quality. It is applicable to energy consumption optimization and intelligent management of various smart property parking lots.

Description

Technical Field

[0001] This invention relates to the field of smart property and energy management technology, specifically to a smart property parking lot energy consumption optimization and management system based on big data analysis. Background Technology

[0002] With the acceleration of urbanization and the continuous growth of car ownership, parking lots, as an important component of building infrastructure, are experiencing a year-on-year increase in their energy consumption as a proportion of property operating costs. Parking lot energy consumption mainly comes from lighting systems, ventilation systems, and charging facilities, with lighting and ventilation systems typically operating for extended periods and thus accounting for a significant portion of energy consumption. How to effectively reduce overall parking lot energy consumption while ensuring parking safety, air quality, and user experience has become a key issue of concern in smart property management and green building. In recent years, with the development of IoT, big data, and artificial intelligence technologies, parking lot energy management has gradually evolved from traditional timed control towards intelligent and refined management.

[0003] In the existing technology, several intelligent solutions for parking lot energy consumption management have emerged. For example, patent document CN119091518A discloses a comprehensive management backend operation and maintenance system, which includes an energy management module for automatically adjusting the status of lighting and ventilation systems. This solution can control lighting and ventilation equipment according to preset rules or simple sensor trigger signals, achieving a certain degree of automated management. Another example is patent document CN118351708B, which discloses a carbon emission energy-saving optimization scheduling system and method based on energy conservation and environmental protection. This system collects real-time video of the parking lot, plans driving routes for vehicles entering the parking lot, calculates carbon emission impact coefficients, and then determines parking space allocation through a priority index, achieving energy-saving scheduling. This solution focuses on route planning and carbon emission optimization during the vehicle entry phase, providing a new approach to parking lot energy-saving management. However, the existing technologies mentioned above still have the following shortcomings: First, the sensing methods are relatively simple, mainly relying on video or basic sensors, making it difficult to obtain dynamic information such as vehicle movement trends, personnel density, and pollutant diffusion; second, the control architecture mostly adopts centralized or independent control, and there is a lack of coordination mechanism between subsystems such as lighting, ventilation, and ground locks, making it difficult to form an overall optimization effect; third, the control strategies are mostly responsive or rule-based, resulting in problems such as response lag, local over-ventilation, or ineffective lighting.

[0004] A comprehensive analysis of existing technologies reveals that current parking lot energy management systems face the following technical challenges: a lack of effective collaborative control mechanisms among different functional subsystems such as lighting, ventilation, and parking locks. These subsystems operate independently and do not communicate with each other, leading to a lack of data sharing, conflicting control decisions, and difficulty in achieving spatiotemporal collaborative optimization of multiple physical fields (light field, flow field, and thermal field). Specifically, this manifests as: lighting systems only activating when vehicles arrive, resulting in response lag; ventilation systems activating all fans in the area during localized pollution, causing energy waste; and charging space management failing to integrate with ventilation and lighting, thus hindering the formation of a unified energy consumption optimization strategy. These problems restrict further improvement in the overall energy efficiency of parking lots, necessitating a solution capable of achieving collaborative control of multiple subsystems. This invention is proposed against this backdrop. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of existing technologies and provide a smart property parking lot energy consumption optimization and management system based on big data analysis. By constructing a distributed intelligent agent network composed of lighting intelligent agents, ventilation intelligent agents, and parking lock intelligent agents, and based on multimodal perception data and multi-agent reinforcement learning models, it realizes the coordinated control of light, wind, and lock in a three-in-one manner to solve the problems of control conflicts and low energy efficiency caused by the independent operation of multiple subsystems.

[0006] To address the aforementioned technical problems, this invention provides the following technical solution: a smart property parking lot energy consumption optimization and management system based on big data analysis, comprising: Multiple thermal imaging sensors are deployed at parking lot entrances and main roads to collect vehicle thermal radiation data and generate vehicle thermal feature trajectories. Multiple environmental sensors are deployed inside the parking lot to collect environmental data, including pollutant concentration and parking space occupancy status. Multiple lighting intelligent agents, each lighting intelligent agent corresponds to an LED lamp, and each lighting intelligent agent includes a first wireless communication module, a first edge computing module and a brightness adjustment module; Multiple ventilation intelligent agents, each ventilation intelligent agent corresponds to a jet fan or an exhaust valve, and each ventilation intelligent agent includes a second wireless communication module, a second edge computing module and a fan control module; Multiple smart lock agents, each smart lock agent corresponds to a parking space lock, and each smart lock agent includes a third wireless communication module, a third edge computing module and a lock driving module; The lighting agent, the ventilation agent, and the lock agent form a distributed agent network through a wireless mesh network, and local communication connections are established between agents in adjacent physical spaces. The first edge computing module of the lighting agent autonomously decides the brightness adjustment amount of the lighting agent based on the vehicle's thermal feature trajectory and the local observation information broadcast by neighboring lighting agents, through a pre-trained first multi-agent reinforcement learning model. The second edge computing module of the ventilation agent, based on pollutant concentration, vehicle thermal trajectory and local observation information broadcast by neighboring ventilation agents, autonomously decides the speed adjustment amount or valve opening adjustment amount of the ventilation agent through a pre-trained second multi-agent reinforcement learning model. The third edge computing module of the parking lock agent, based on the parking space occupancy status and local observation information broadcast by neighboring parking lock agents, autonomously decides whether to raise, lower, or maintain the parking space through a pre-trained third multi-agent reinforcement learning model. By constructing a distributed agent network consisting of lighting agents, ventilation agents, and parking lock agents, and making collaborative decisions based on a multi-agent reinforcement learning model, it is possible to achieve joint control of light field, flow field, and parking lock.

[0007] Furthermore, the thermal imaging sensor is a low-resolution thermal imaging sensor, and the vehicle thermal feature trajectory includes the coordinates of the center point of the heat cluster contour, the rate of change of the heat cluster area, the peak temperature of the heat cluster, the thermal radiation distribution entropy, and motion vector data, wherein the motion vector data includes velocity, acceleration, and orientation angle. By using low-resolution thermal imaging sensors to collect vehicle thermal radiation data and extracting vehicle thermal feature trajectories, including the coordinates of the center point of the heat cluster outline and the rate of change of the heat cluster area, it is possible to obtain information on vehicle movement trends and personnel density without infringing on privacy, providing a data foundation for advanced predictive control.

[0008] Furthermore, the system also includes a voiceprint recognition sensor deployed above the parking space to collect vehicle start-up voiceprint data; The environmental sensor includes a gas sensor for collecting carbon monoxide concentration, PM2.5 concentration, nitrogen oxide concentration, or TVOC concentration. The system locates the coordinates of parking spaces at pollution sources based on pollutant concentration, vehicle start-up soundprint data, and vehicle engine type data. By deploying voiceprint recognition sensors to collect vehicle start-up voiceprint data and combining it with pollutant concentration data collected by gas sensors to locate the coordinates of parking spaces at the source of pollution, the location of local pollution events can be accurately identified, avoiding energy waste caused by starting full-area ventilation.

[0009] Furthermore, the system also includes a gate data interface for obtaining vehicle entry time, vehicle engine type, and license plate information; The system constructs a parking space behavior profile based on the parking space occupancy status and vehicle entry time. The parking space behavior profile includes parking duration distribution, parking time period characteristics, and turnover rate data. By setting up a gate data interface to obtain vehicle entry time and engine type, and combining this with parking space occupancy status to construct a parking space behavior profile including parking duration distribution and turnover rate, differentiated lighting and ventilation strategies can be formulated for different areas to achieve on-demand energy supply.

[0010] Furthermore, the first edge computing module of the lighting agent also receives pollutant diffusion path information broadcast by the adjacent ventilation agent. When the area covered by the lighting agent is marked as a pollutant diffusion path, the lighting agent adjusts the brightness to a preset warning brightness value. By integrating the function of receiving pollutant diffusion path information broadcast by adjacent ventilation agents into the lighting agent and adjusting the brightness to a warning value when the covered area is marked as a diffusion path, it is possible to prompt people to avoid polluted areas through light signals while ensuring ventilation efficiency.

[0011] Furthermore, the second edge computing module of the ventilation smart agent also receives charging status information broadcast by adjacent parking lock smart agents. When the charging space within the coverage area of ​​the ventilation smart agent is fully charged and the vehicle has not left for more than a preset threshold, the ventilation smart agent reduces the speed adjustment amount to the baseline operating value. By integrating the function of receiving charging status information broadcast by neighboring smart locks into the ventilation agent, and reducing the speed to the baseline operating value when charging is completed in the charging parking space and the vehicle has not left, ineffective ventilation of the area where the vehicle has completed charging can be avoided.

[0012] Furthermore, the third edge computing module of the ground lock intelligent agent also receives the pollutant source tracing results broadcast by the adjacent ventilation intelligent agent. When the parking space corresponding to the ground lock intelligent agent is marked as a pollution source and the vehicle has completed charging, the ground lock intelligent agent performs a lifting action. By integrating the function of receiving pollutant source tracing results broadcast by adjacent ventilation smart agents into the parking lock smart agent, and performing the lifting action when the corresponding parking space is marked as a pollution source and the vehicle has completed charging, the parking lock and ventilation system can be linked to guide users to cooperate in the rapid discharge of pollutants.

[0013] Furthermore, the local observation information broadcast in the local communication connection includes the agent's current execution parameters, the agent's cumulative energy consumption, the predicted probability of vehicle arrival in the agent's coverage area, the pollutant concentration gradient in the agent's coverage area, or the parking space occupancy time in the agent's coverage area. By broadcasting local observation information, including vehicle arrival prediction probability, pollutant concentration gradient, and parking space occupancy duration, in local communication connections, neighboring agents can share environmental situation, providing information support for distributed collaborative decision-making.

[0014] Furthermore, the first multi-agent reinforcement learning model, the second multi-agent reinforcement learning model, and the third multi-agent reinforcement learning model adopt a centralized training-distributed execution framework. After the model training is completed on the cloud server, the execution network is deployed on the edge computing modules of each agent. By employing a centralized training-distributed execution framework, the model is trained in the cloud and then the execution network is deployed on the edge computing modules of each agent, enabling distributed control with millisecond-level response while ensuring model performance.

[0015] Furthermore, the system also includes a cloud server, which is communicatively connected to edge nodes of multiple parking lots. Each edge node uploads the parameters of its locally trained multi-agent reinforcement learning model to the cloud server. The cloud server generates a globally optimized model through a federated averaging algorithm and distributes it to each edge node to update its local model. By setting up communication connections between cloud servers and edge nodes of multiple parking lots, and using a federated averaging algorithm to generate a global optimization model for distribution and updates, cross-scenario knowledge sharing and rapid model adaptation can be achieved while protecting the data privacy of each parking lot.

[0016] Compared with existing technologies, this smart property parking energy consumption optimization and management system based on big data analysis has the following beneficial effects: I. This invention constructs a distributed intelligent agent network composed of lighting intelligent agents, ventilation intelligent agents, and parking lock intelligent agents. Based on a wireless mesh network, it realizes local communication and observation information sharing among intelligent agents in adjacent physical spaces. Through a multi-agent reinforcement learning model, it enables autonomous collaborative decision-making among the intelligent agents. This breaks down information barriers and control silos between lighting, ventilation, and parking management subsystems, achieving spatiotemporal collaborative optimization of light field, flow field, and parking management. It avoids control conflicts and response lags caused by the independent operation of each subsystem, thereby significantly improving the coordination and refinement of parking lot energy consumption management and effectively reducing ineffective energy consumption across the entire scenario.

[0017] Second, this invention collects vehicle thermal trajectory, environmental parameters, and equipment operation data across all dimensions through a multimodal perception system. Combined with a centralized training and distributed execution multi-agent reinforcement learning framework and federated learning mechanism, it can predict vehicle movement trends and accurately locate pollution sources without infringing on user privacy. At the same time, it achieves a balance between global optimization and real-time response through a cloud-edge collaborative scheduling architecture. It can also realize knowledge sharing and rapid model adaptation across parking lots, thereby continuously improving the system's energy-saving optimization effect and scenario adaptability while ensuring parking lot operation safety and service quality.

[0018] Other advantages, objectives and features of the invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art from the following examination or study, or may be learned from the practice of the invention. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without any creative effort.

[0020] Figure 1 This is a diagram of the overall system architecture of the present invention; Figure 2 This is a schematic diagram of the parking lot energy consumption data collection and preprocessing process of the present invention; Figure 3 This is a schematic diagram of the parking lot energy consumption optimization scheduling execution logic of the present invention. Detailed Implementation

[0021] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided below.

[0022] Example

[0023] In related technologies, energy consumption management in property parking lots often adopts a fixed-time equipment start-stop control mode, which cannot adapt to the dynamic changes in actual traffic flow and environmental parameters, easily leading to a large amount of ineffective energy consumption. Furthermore, it lacks comprehensive energy consumption data collection and refined analysis and scheduling capabilities, making it difficult to achieve global optimization and intelligent management of parking lot energy consumption. This invention utilizes big data analytics to achieve in-depth mining and dynamic optimization scheduling of energy consumption data across all parking lot scenarios, comprehensively improving the intelligence level and energy utilization efficiency of parking lot energy consumption management.

[0024] like Figure 1 As shown, the overall system architecture, from bottom to top, consists of a data acquisition layer, a data preprocessing layer, a big data analysis and prediction layer, an energy consumption optimization and scheduling layer, and a visualization management and interaction layer. Each layer communicates with the others via wired or wireless communication networks to achieve bidirectional data transmission and command execution, forming a complete closed-loop management system.

[0025] Specifically, the data acquisition layer is used to realize the real-time collection and reporting of energy consumption-related data for the entire parking lot scenario, and it is the data foundation of the entire system. In this embodiment, the data acquisition layer includes multiple data acquisition units such as smart meters, smart water meters, environmental sensors, vehicle detection equipment, equipment operation status acquisition terminals, and access control gate data acquisition units.

[0026] Specifically, smart meters are deployed at the incoming ends of various power circuits within the parking lot, including lighting circuits, ventilation circuits, elevator circuits, charging pile circuits, gate equipment circuits, and monitoring equipment circuits. The smart meters collect parameters including real-time voltage, real-time current, active power, reactive power, power factor, cumulative electricity consumption, and peak / off-peak electricity consumption. In this embodiment, the smart meter's data collection frequency is set to once per minute. For circuits corresponding to high-power equipment, the collection frequency can be set to once every 10 seconds to ensure the real-time nature and completeness of energy consumption data. The smart meters establish a connection with the field data gateway via an RS485 communication interface, uploading the collected electrical parameter data to the data preprocessing layer in real time.

[0027] Specifically, smart water meters are deployed at water supply pipeline nodes within the parking lot, including cleaning water pipelines and fire-fighting water pipelines. The smart water meters collect parameters including real-time flow rate, cumulative water consumption, and water supply pressure. The smart water meters are set to collect data once per hour, and the collected data is uploaded to the on-site data gateway via LoRa wireless communication.

[0028] Specifically, environmental sensors are deployed in different areas within the parking lot, including the entrance area, parking space areas, driveway areas, and equipment room areas. These sensors include temperature and humidity sensors, carbon dioxide concentration sensors, illuminance sensors, and air quality sensors. The temperature and humidity sensors collect ambient temperature and relative humidity data every 5 minutes. The carbon dioxide concentration sensor collects real-time carbon dioxide concentration data within the parking lot every 5 minutes. The illuminance sensor collects real-time light intensity data for each area within the parking lot every minute. The air quality sensor collects PM2.5 and PM10 concentration data within the parking lot every 10 minutes. All environmental sensors are connected to a field data gateway via a ZigBee wireless communication network for unified reporting of environmental parameter data.

[0029] Specifically, vehicle detection equipment is deployed at the entrance, exit, and above each parking space in the parking lot. This equipment includes geomagnetic vehicle detectors, video parking space detectors, and license plate recognition cameras at entrances and exits. The geomagnetic and video parking space detectors collect real-time occupancy status, occupancy duration, and area occupancy rate for each parking space. The license plate recognition cameras at entrances and exits collect data on vehicle entry and exit times, vehicle type, daily traffic flow, and time-of-day traffic flow. The data collection frequency of the vehicle detection equipment is set according to its function; the parking space status detection frequency is set to once every 2 seconds, while the traffic flow statistics frequency is set to real-time triggered reporting, uploading corresponding data immediately upon vehicle entry and exit. The vehicle detection equipment connects to the on-site data gateway via Ethernet communication to upload vehicle-related data.

[0030] Specifically, the equipment operation status acquisition terminal is deployed at the control terminals of various energy-consuming equipment within the parking lot, including lighting equipment controllers, ventilation fan frequency converters, elevator operation controllers, charging pile management units, and fire equipment monitoring units. The acquisition parameters collected by the equipment operation status acquisition terminal include the equipment's operating status, operating power, operating time, number of start-stop cycles, equipment fault status, and equipment operating speed. The acquisition frequency of the equipment operation status acquisition terminal is set to once every 10 seconds, and it establishes a communication connection with the field data gateway via industrial Ethernet to upload comprehensive data on equipment operation in real time.

[0031] Specifically, the access control gate data acquisition unit is deployed at the parking lot entrance and exit gate equipment. The acquired parameters include the number of gate start / stop cycles, gate operating time, real-time power consumption of the gate equipment, and gate fault status. The data acquisition unit is set to acquire data once every 30 seconds and connects to the field data gateway via an RS485 communication interface to upload relevant data from the gate equipment.

[0032] It is understandable that the data acquisition layer uses a variety of types and wide coverage of acquisition devices to achieve full-dimensional collection of energy consumption data, environmental data, vehicle data, and equipment operation data in the parking lot, providing complete and reliable data support for subsequent big data analysis and energy consumption optimization, and avoiding optimization and scheduling deviations caused by data gaps.

[0033] like Figure 2 As shown, the data preprocessing process includes six steps in sequence: data reception and aggregation, data cleaning, outlier handling, data normalization, data fusion, and data storage. The data preprocessing layer is deployed in the field edge gateway and the cloud server. The preprocessing steps with high real-time requirements are executed in the edge gateway, while the non-real-time batch data preprocessing is executed in the cloud server.

[0034] Specifically, the data receiving and aggregation step is used to receive multi-source heterogeneous data uploaded by various acquisition units in the data acquisition layer, classify and aggregate it according to data type, and establish a unified data index. In this embodiment, data receiving and aggregation are implemented using a message queue to ensure that data is not lost and is processed in an orderly manner under high-concurrency scenarios. The received raw data is classified and stored according to four main categories: energy consumption data, environmental data, vehicle data, and equipment operation data. At the same time, a timestamp and a unique device identifier are added to each data item to facilitate subsequent data traceability and processing.

[0035] Specifically, the data cleaning step is used to remove invalid data and handle missing values ​​from the aggregated raw data. It is understood that invalid and missing data may be generated during data collection due to equipment communication interruptions, equipment failures, and network fluctuations. Directly using such data will affect the accuracy of subsequent analysis results. In this embodiment, the invalid data removal rules are as follows: delete invalid data that does not conform to the data value range, and delete redundant data that is repeatedly reported. Missing value handling is implemented using a combination of linear interpolation and historical average methods. For continuous time-series data, when the duration of missing data is less than a preset first duration threshold, linear interpolation is used to fill in the missing values. When the duration of missing data is greater than or equal to the preset first duration threshold, the historical average method for the same time period is used to fill in the missing values. In this embodiment, the preset first duration threshold is set to 30 minutes, and this threshold can be adjusted according to the actual operation of the parking lot.

[0036] Specifically, the outlier handling step is used to identify and correct outliers in the cleaned data. In this embodiment, outlier identification is achieved by combining the 3σ criterion and the Isolation Forest algorithm. First, the 3σ criterion is used to perform preliminary outlier screening on the time-series data, identifying outlier data points that exceed the normal fluctuation range. Then, the Isolation Forest algorithm is used to perform secondary identification on the pre-screened outlier data points, distinguishing between real outlier data and pseudo-outlier data. For pseudo-outlier data, the original data values ​​are retained. For real outlier data, the average value of normal data from adjacent time periods is used for correction and replacement. It can be understood that by using a dual outlier identification method, misjudgments caused by data fluctuations can be effectively avoided, improving the accuracy of data processing.

[0037] Specifically, the data normalization step is used to standardize the processed data, eliminate dimensional differences between different data types, and improve the convergence speed and analysis accuracy of subsequent big data analysis algorithms. In this embodiment, data normalization is implemented using the min-max normalization method, and the corresponding normalization formula is:

[0038] in, These are the normalized data values. The actual values ​​of the original data. It is the minimum value in this data sequence. This is the maximum value in the data sequence.

[0039] It is understandable that the above normalization process can map the value range of all data to the interval between 0 and 1, eliminate the influence of different dimensions between parameters, and lay the foundation for subsequent multi-dimensional data fusion and algorithm analysis.

[0040] Specifically, the data fusion step is used to fuse the normalized multi-source data in terms of time and spatial dimensions. Time dimension fusion involves aligning data from different collection frequencies to the same time step, generating a time-series dataset with a unified time dimension. Spatial dimension fusion involves correlating and fusing energy consumption data, environmental data, vehicle data, and equipment operation data within the same parking lot area according to its regional division, generating a multi-dimensional fused dataset based on the region. In this embodiment, the unified time step for time dimension fusion is set to 1 minute. For data collected less than 1 minute, time dimension alignment is performed by keeping the values ​​unchanged.

[0041] Specifically, the data storage step is used to classify and store the fused dataset. In this embodiment, a time-series database is used to store time-series data with high real-time requirements, a relational database is used to store basic device information, regional configuration information, and structured user information, and a distributed file storage system is used to store historical batch data and analysis results. Through a layered storage architecture, both data read / write performance and storage capacity requirements are considered, ensuring secure and reliable data storage and fast retrieval.

[0042] Specifically, the big data analysis and prediction layer is used to perform in-depth mining and analysis of preprocessed parking lot data across all dimensions, enabling the analysis of correlation factors of parking lot energy consumption, traffic flow prediction, and energy demand prediction, providing data support and decision-making basis for subsequent energy consumption optimization and scheduling. The big data analysis and prediction layer is deployed on a cloud server and includes three core functional modules: an energy consumption correlation factor analysis module, a traffic flow prediction module, and an energy consumption prediction module.

[0043] Specifically, the energy consumption correlation factor analysis module is used to explore the correlation between parking lot energy consumption and various influencing factors, identifying core factors that significantly affect energy consumption changes. In this embodiment, the energy consumption correlation factor analysis is implemented using the Pearson correlation coefficient analysis method. By calculating the Pearson correlation coefficient between energy consumption data and the data of each influencing factor, the degree of influence of each factor on energy consumption is quantified. The formula for calculating the Pearson correlation coefficient is:

[0044] in, The Pearson correlation coefficient between variables X and Y. The total number of data samples. Let the i-th sample of variable X take its value. For the i-th sample of variable Y, Let X be the arithmetic mean of all sample values ​​of variable X. Y is the arithmetic mean of all sample values ​​for variable Y.

[0045] It is understood that the Pearson correlation coefficient ranges from -1 to 1. The closer the absolute value of the correlation coefficient is to 1, the stronger the linear correlation between the two variables. Conversely, the closer the absolute value is to 0, the weaker the linear correlation. In this embodiment, factors with correlation coefficients greater than or equal to a preset threshold are identified as the core influencing factors of parking lot energy consumption. In this embodiment, the preset threshold is set to 0.6, which can be adjusted based on the actual operating data of the parking lot. Through energy consumption correlation factor analysis, traffic flow, ambient temperature, light intensity, and time period type can be identified as the core influencing factors of parking lot energy consumption, providing clear input dimensions for subsequent prediction model construction and optimized scheduling.

[0046] Specifically, the traffic flow prediction module is used to predict the trend of traffic flow and parking space occupancy changes in parking lots within a preset time period based on historical traffic flow data and related influencing factor data. In this embodiment, the traffic flow prediction is implemented using a long short-term memory network model. This model can effectively capture long-term dependencies in time-series data, adapt to the nonlinear and temporal characteristics of traffic flow data, and improve prediction accuracy.

[0047] Specifically, the construction and training process of the Long Short-Term Memory (LSTM) network model is as follows. First, the input features and output target of the model are determined. Input features include historical traffic flow data, historical parking space occupancy rate data, time feature data, weather feature data, and holiday feature data. Time feature data includes hourly features, weekday features, and weekday / restday features. The output target is the traffic flow and parking space occupancy rate for each time period within a preset future timeframe. In this embodiment, the preset future timeframe is set to 24 hours, and the time step for each time period is set to 1 hour, meaning the model outputs the predicted traffic flow and parking space occupancy rate for each hour within the next 24 hours.

[0048] The input feature data is then preprocessed, and the preprocessed dataset is divided into a training set, a validation set, and a test set. The training set is used for model parameter training, the validation set is used for hyperparameter tuning during model training, and the test set is used to verify the model's prediction accuracy. In this embodiment, the ratio of the training set, validation set, and test set is set to 7:2:1.

[0049] Next, the network structure of the Long Short-Term Memory (LSTM) network model was constructed. The model structure consists of an input layer, hidden layers, a fully connected layer, and an output layer. The hidden layer is configured with two LSM units, each with 128 neurons. A Dropout layer is placed after the hidden layers to prevent overfitting. The dropout rate of the Dropout layer is set to 0.2. The fully connected layer has 64 neurons and is used to integrate and map the features output from the hidden layers. The number of neurons in the output layer is consistent with the number of time steps in the prediction duration, and it is used to output the final prediction result.

[0050] During model training, mean squared error is used as the loss function, and an adaptive moment estimator (AME) is used as the optimization algorithm. The formula for calculating the loss function is:

[0051] in, This is the mean squared error loss value. To predict the total number of samples, This represents the actual value of traffic flow or parking space occupancy rate. This refers to the predicted traffic flow or parking space occupancy rate output by the model.

[0052] The model training iterations are set to 200, and the batch size is set to 32. If the loss value on the validation set does not decrease for 10 consecutive iterations, model training is terminated early, and the optimal model parameters are saved. The trained model is then validated using a test set. When the model's prediction accuracy meets the preset accuracy requirements, the model is deployed online for real-time traffic flow and parking space occupancy prediction. In this embodiment, the preset prediction accuracy requirement is a mean absolute percentage error of less than or equal to 10%.

[0053] Specifically, the energy consumption prediction module is used to predict changes in the parking lot's energy consumption demand over a preset period of time, based on historical energy consumption data, traffic flow prediction results, environmental parameter prediction data, and equipment operation data. In this embodiment, the energy consumption prediction is implemented using a model combining a bidirectional long short-term memory network and an attention mechanism. This model can simultaneously capture the forward and backward dependencies of time-series data, and strengthens the weights of features that significantly affect energy consumption changes through the attention mechanism, further improving the accuracy and stability of energy consumption prediction.

[0054] Specifically, the input features of the energy consumption prediction model include historical energy consumption data, traffic flow prediction results, environmental parameter prediction data, time characteristic data, historical equipment operating status data, and peak-valley electricity price data. Environmental parameter prediction data includes predicted ambient temperature, predicted light intensity, and predicted carbon dioxide concentration for a preset future time period. The output target is the time-segmented energy consumption prediction values ​​for each area and each equipment loop in the parking lot for a preset future time period. In this embodiment, the preset future time period is set to 24 hours, and the time step for each time period is set to 1 hour, meaning the model outputs the energy consumption prediction values ​​for each area and each loop for each hour within the next 24 hours.

[0055] The training process for the energy consumption prediction model is the same as that for the traffic flow prediction model, encompassing the entire process of dataset partitioning, model structure construction, model training, accuracy verification, and model deployment. The loss function for both models is mean squared error, and the optimization algorithm uses an adaptive moment estimator (AME). After training, the prediction accuracy is verified using a test set. The model is deployed when the mean absolute percentage error (MAE) is less than or equal to a preset accuracy threshold. In this embodiment, the preset accuracy threshold is set to 8%.

[0056] It is understandable that by predicting traffic flow and energy consumption, we can gain an advance understanding of the future operating status and energy demand changes of parking lots, providing a basis for decision-making in order to optimize energy consumption scheduling and upgrading the energy management model from passive response to proactive prediction.

[0057] like Figure 3 As shown, the energy consumption optimization scheduling execution logic includes six stages: optimization target determination, constraint setting, optimization scheduling strategy solution, strategy feasibility verification, instruction issuance and execution, and operation effect feedback. The energy consumption optimization scheduling layer is deployed in the edge gateway and the cloud server. The solution of the global optimization strategy is executed in the cloud server, while real-time device control and emergency scheduling are executed in the edge gateway, ensuring the real-time performance and reliability of scheduling instructions.

[0058] Specifically, in the optimization objective determination stage, multiple optimization objectives are established, including minimizing the total energy consumption, minimizing the total operating cost, and maximizing the equipment lifespan throughout the parking lot's entire lifecycle. A multi-objective optimization function is constructed. In this embodiment, the multiple optimization objectives are first normalized and then transformed into a single-objective optimization function through linear weighted summation, facilitating the solution of the optimization algorithm.

[0059] The objective function expression for minimizing total energy consumption is:

[0060] in, This represents the total energy consumption of the parking lot during the scheduling cycle. This represents the total number of time steps within the scheduling period. Let t be the time step within the scheduling period. This represents the total number of energy-consuming devices in the parking lot. This is the k-th energy-consuming device in the parking lot. Let be the operating power of the k-th energy-consuming device in the t-th time step. The duration of a single time step.

[0061] The objective function expression for minimizing total operating cost is:

[0062] in, The total energy consumption operating cost of the parking lot during the scheduling cycle. Let be the total energy consumption of the parking lot during the t-th time step. Let t be the grid electricity price corresponding to the t-th time step.

[0063] The optimization objective of maximizing equipment lifespan is achieved by minimizing the number of equipment start-ups and shutdowns and the equipment's runtime under off-rated conditions. The corresponding objective function expression is:

[0064] in, This is a comprehensive quantitative value of equipment operating losses. Let k be the number of times the k-th energy-consuming device starts and stops during the scheduling cycle. Let be the rated operating power of the kth energy-consuming device.

[0065] After normalizing the three optimization objectives, a comprehensive optimization objective function is constructed by linear weighted summation, expressed as:

[0066] in, To comprehensively optimize the objective function value, The weighting coefficients for the total energy consumption optimization objective. The weighting coefficients for the total operating cost optimization objective. Weighting coefficients for optimizing equipment operating losses. This represents the total energy consumption within the scheduling cycle under the baseline scheduling mode. This represents the total operating cost within the scheduling cycle under the baseline scheduling mode. This is the comprehensive quantitative value of equipment operating losses within the scheduling cycle under the baseline scheduling mode.

[0067] In this embodiment, the values ​​of the weighting coefficients satisfy the following: The specific values ​​of each weighting coefficient can be adjusted according to the actual management needs of the parking lot. In this embodiment, when the parking lot takes energy conservation and cost reduction as its core management objective, the following settings can be configured: It is 0.4. It is 0.4. The value is 0.2. When the parking lot's core management objective is stable equipment operation, it can be set to... It is 0.3. It is 0.2. It is 0.5.

[0068] Specifically, in the constraint setting stage, multiple constraints are set for solving the objective function, including safety operation constraints, service quality constraints, equipment operation constraints, and power grid operation constraints, to ensure the feasibility and security of the optimized scheduling strategy.

[0069] Specifically, safety operation constraints include environmental parameter safety constraints within the parking lot and fire equipment operation constraints. Environmental parameter safety constraints stipulate that environmental parameters in all areas of the parking lot must be maintained within preset safety ranges. The ambient temperature range is set to 5 degrees Celsius to 35 degrees Celsius. The upper limit for carbon dioxide concentration is set at 1000 ppm. The upper limit for PM2.5 concentration is set at 75 micrograms per cubic meter. Fire equipment must be kept in normal operating condition 24 / 7 and must not be shut down for energy-saving purposes.

[0070] Specifically, the service quality constraints are as follows: the lighting brightness of the parking lot must meet the safety requirements for vehicle driving and pedestrian passage. The minimum illuminance for parking space areas is set at 50 lux, for driveway areas at 100 lux, and for entrance / exit areas at 200 lux. The parking lot's ventilation system must ensure that the indoor air quality meets standards and must not cause the indoor air quality to exceed safe limits due to energy-saving scheduling. The parking lot's vehicle flow service must not be affected by scheduling strategies, and the barrier gates, monitoring, and license plate recognition equipment must be kept in normal operating condition.

[0071] Specifically, the equipment operation constraints are as follows: the operating power of each energy-consuming device must be maintained between the device's minimum and maximum permissible operating power. The number of device start-ups and shutdowns must not exceed the device's maximum permissible number of start-ups and shutdowns per day. The continuous operating time of the device must not exceed the device's maximum permissible continuous operating time. The start-stop interval of the device must not be less than the preset minimum start-stop interval to avoid damage caused by frequent device start-ups and shutdowns. In this embodiment, the preset minimum start-stop interval is set to 15 minutes, and this parameter can be adjusted according to the operating specifications of different devices.

[0072] Specifically, the power grid operation constraints are as follows: the total power consumption of the parking lot must not exceed the rated maximum power supply of the distribution system to avoid overload operation. The power factor of the parking lot must be maintained above 0.9 to meet the power quality requirements of the power grid.

[0073] Specifically, in the optimization scheduling strategy solution stage, based on the constructed comprehensive optimization objective function and constraints, an improved non-dominated sorting genetic algorithm is used to generate the globally optimal energy consumption optimization scheduling strategy. The optimized scheduling strategy includes the time-segmented operation levels, start-stop times, and operating power settings of various lighting equipment, ventilation equipment, and elevator equipment in the parking lot, as well as the orderly charging scheduling strategy for charging piles.

[0074] Specifically, the solution process of the improved non-dominated sorting genetic algorithm is as follows. First, the algorithm population is initialized with a size of 100. Individuals in the population represent combinations of operating parameters for each device at each time step within the scheduling cycle. The maximum number of iterations is set to 200. The crossover probability is set to 0.8, and the mutation probability is set to 0.1. Then, non-dominated sorting and crowding calculation are performed on the individuals in the population. A new population is generated through selection, crossover, and mutation operations. An adaptive crossover and mutation probability adjustment mechanism is introduced during the iteration process. When the fitness value of the population does not improve for several consecutive generations, the mutation probability is increased to avoid the algorithm getting trapped in local optima. After iteration, the solution with the smallest comprehensive optimization objective function value is selected from the final non-dominated solution set as the optimal energy consumption optimization scheduling strategy.

[0075] Specifically, in the strategy feasibility verification stage, the optimized scheduling strategy generated by the solution is subjected to constraint verification to confirm whether the strategy meets all preset constraints. If the strategy does not meet any of the constraints, the optimization solution is re-run to generate a new scheduling strategy. When the strategy meets all constraints, it is determined to be an executable optimized scheduling strategy.

[0076] Specifically, in the instruction issuance and execution phase, the validated optimized scheduling strategy is broken down into specific control instructions for each device, which are then sent to the corresponding device operation controllers via the field communication network. This controls the devices to operate according to the parameters set in the scheduling strategy. In this embodiment, an execution mode combining cloud-based global scheduling and edge real-time scheduling is adopted. The cloud server generates the global optimized scheduling strategy for the next day 24 hours in advance and sends it to the edge gateway. The edge gateway performs real-time fine-tuning and correction of the global scheduling strategy based on the parking lot's real-time operating data, real-time traffic flow changes, and real-time environmental parameter changes, ensuring that the scheduling strategy adapts to real-time changes in on-site conditions and improves the optimization effect.

[0077] Specifically, in the operational performance feedback phase, real-time data on equipment operation, energy consumption, environment, and traffic flow are collected after the scheduling strategy is implemented. The actual operational performance is compared and analyzed with the predicted results to calculate the actual energy-saving effect of the optimization strategy. Simultaneously, new data generated during operation is added to the historical dataset for iterative updates to the prediction model and optimization algorithm, continuously improving the model's prediction accuracy and the algorithm's solution performance, thus forming a complete closed-loop optimization system.

[0078] Specifically, the visualization management and interaction layer provides property management personnel with a visual system management interface and interactive entry point, enabling functions such as visual display of parking lot energy consumption data, remote monitoring and control of equipment, alarm information management, data report generation, and system configuration management. The visualization management and interaction layer includes two ports: a web-based management platform and a mobile management application, meeting the usage needs of management personnel in different scenarios.

[0079] Specifically, the web-based management platform includes a data visualization dashboard module to display core parking lot operational data, including real-time total energy consumption, daily cumulative energy consumption, monthly cumulative energy consumption, year-on-year energy consumption change rate, month-on-month energy consumption change rate, real-time traffic flow, daily cumulative traffic flow, real-time parking space occupancy rate, operating status of various equipment, and real-time environmental parameters. The data visualization dashboard uses a zoned layout, displaying core data in the form of charts, numbers, and dynamic trend graphs, allowing managers to intuitively grasp the overall operational status and energy consumption of the parking lot.

[0080] Specifically, the web-based management platform includes a remote equipment monitoring and control module. This module displays the real-time operating status, parameters, and energy consumption data of all energy-consuming equipment within the parking lot. It allows administrators to remotely issue start / stop commands, gear adjustment commands, and parameter setting commands for the equipment. The module also features an equipment fault alarm function. When equipment malfunctions or exhibits abnormal parameters, an alarm message immediately pops up on the platform, recording the alarm time, affected equipment, alarm type, and abnormal parameters, facilitating timely fault handling by administrators.

[0081] Specifically, the web-based management platform includes an energy consumption analysis and report management module. This module supports generating energy consumption statistical reports based on daily, weekly, monthly, and yearly time dimensions. It also supports the breakdown, statistical analysis, and comparative analysis of energy consumption data by region, equipment type, and equipment circuit. Reports can be exported to Excel format files for easy archiving and reporting by management personnel. Furthermore, the energy consumption analysis and report management module supports energy consumption benchmarking analysis, comparing actual parking lot energy consumption data with industry benchmarks, historical data for the same period, and preset energy consumption targets. This displays the progress towards energy-saving goals and identifies areas and equipment with abnormal energy consumption.

[0082] Specifically, the web-based management platform includes a system configuration management module for administrators to configure and manage system parameters. This includes adding and deleting data acquisition devices, configuring device parameters, setting acquisition frequency, adjusting prediction model parameters, setting optimization scheduling weight coefficients, setting alarm thresholds, managing user permissions, and modifying passwords. The system configuration management module implements multi-level user permissions, with different permissions granting access to different functional modules, ensuring system operational security and data security.

[0083] Specifically, the mobile management application possesses the core functions of the web-based management platform, including real-time data viewing, remote device control, alarm information reception, and energy consumption report viewing. It enables managers to monitor the parking lot's operational status anytime, anywhere, handle device alarms and emergencies, and improve the flexibility and response efficiency of parking lot energy consumption management.

[0084] In this embodiment, the complete implementation process of the smart property parking lot energy consumption optimization and management system based on big data analysis of the present invention is as follows.

[0085] The first step is to complete the deployment and debugging of data acquisition equipment, control equipment, and communication gateways in the parking lot, establish the hardware foundation for the data acquisition layer, and realize the real-time acquisition and reporting of energy consumption-related data in the entire parking lot scenario.

[0086] The second step is to deploy the relevant programs for the data preprocessing layer, configure the relevant parameters for data cleaning, outlier handling, data normalization, and data fusion, and complete the preprocessing and standardization of multi-source heterogeneous data to provide high-quality datasets for subsequent big data analysis.

[0087] The third step is to build and train traffic flow prediction models and energy consumption prediction models, complete the accuracy verification and optimization of the models, and deploy the trained models to the big data analysis and prediction layer to achieve real-time prediction of parking lot traffic flow and energy consumption demand.

[0088] The fourth step is to construct a multi-objective energy consumption optimization function and constraints, deploy an improved non-dominated sorting genetic algorithm, and realize the automatic solution and feasibility verification of the global optimal energy consumption optimization scheduling strategy.

[0089] The fifth step is to deploy the scheduling execution unit on the edge side, establish a two-way communication mechanism between the cloud and the edge side, break down the optimized scheduling strategy into device control commands for execution, and dynamically fine-tune the strategy based on real-time on-site operating data to ensure scheduling effectiveness.

[0090] The sixth step is to deploy the web-based management platform and mobile management application for the visualization and interaction layer, complete the configuration of system functions and the setting of user permissions, and provide property management personnel with a visual management and interaction entry point.

[0091] The seventh step is to continuously collect full-dimensional data during the system's operation after it goes live, and to regularly iterate and update the prediction model and optimization algorithm to continuously improve the system's prediction accuracy and optimization effect, forming a closed-loop optimization management system.

[0092] In this embodiment, the present invention is applied to an underground intelligent property parking lot supporting a commercial complex in a certain city. The total floor area of the parking lot is 20,000 square meters, with 500 parking spaces. The energy-consuming equipment in the parking lot includes 800 sets of lighting fixtures, 12 ventilators, 4 elevators, 60 charging piles, as well as gate barriers, monitoring, and fire protection supporting equipment. The original energy consumption management mode of the parking lot is the start-stop control of equipment at fixed time periods. The lighting equipment is fully powered on from 6:00 to 24:00 every day, and the ventilation equipment is fixed to be turned on 4 times a day, running for 2 hours each time. It cannot adapt to the dynamic changes of traffic flow and environmental parameters, resulting in serious energy consumption waste problems.

[0093] Specifically, after applying the system of the present invention, the deployment of intelligent electricity meters, intelligent water meters, environmental sensors, vehicle detection equipment, and equipment operation status acquisition terminals is completed in the parking lot to achieve real-time acquisition of full-scenario data. Through the data preprocessing layer, the cleaning and standardization processing of multi-source data are completed. Through the big data analysis and prediction layer, the accurate prediction of traffic flow and energy consumption demand for 24 hours is achieved. Through the energy consumption optimization scheduling layer, the optimized operation strategy of equipment for sub-periods is generated. The lighting equipment dynamically adjusts the number of lights turned on and the operating power according to the real-time parking space occupancy rate and light intensity in each area, reducing the ineffective energy consumption while ensuring the lighting demand. The ventilation equipment dynamically adjusts the number of operating units and the operating gears according to the real-time carbon dioxide concentration and traffic flow changes in the parking lot, achieving ventilation on demand. The elevator equipment dynamically adjusts the number of operating units and the operating mode according to the traffic flow prediction result, reducing the energy consumption of no-load operation. The charging pile equipment realizes the orderly charging scheduling according to the peak-valley electricity price periods and the grid load conditions, reducing the electricity cost.

[0094] After the system runs stably, the monthly total energy consumption of the parking lot decreases by 22% compared with the original mode, the monthly energy consumption operation cost decreases by 26%, the average start-stop times of the equipment decrease by 35%, effectively extending the service life of the equipment. At the same time, the environmental parameters in the parking lot always remain within the preset safe range, ensuring the service quality and operation safety of the parking lot.

[0095] It is understood that this invention, through a multi-dimensional, multi-source data acquisition system, achieves comprehensive and real-time collection of parking lot energy consumption data, solving the problems of single data collection dimensions and insufficient data completeness in related technologies. Through big data analysis and deep learning algorithms, it enables accurate prediction of parking lot traffic flow and energy consumption demand, allowing for advance understanding of changes in parking lot operating status and providing a preliminary decision-making basis for energy consumption optimization. Through multi-objective optimization algorithms, it achieves global collaborative optimization of parking lot energy consumption, operating costs, and equipment operating losses, effectively improving energy utilization efficiency and reducing property operating costs while ensuring parking lot operational safety and service quality. Through a cloud-edge collaborative scheduling execution mode, it balances the rationality of global optimization with the flexibility of real-time scheduling, ensuring the effective implementation of optimized scheduling strategies. This invention can effectively improve the intelligence and precision of smart property parking lot energy consumption management.

[0096] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A smart property parking lot energy consumption optimization and management system based on big data analysis, characterized in that, The system consists of: Multiple thermal imaging sensors are deployed at parking lot entrances and main roads to collect vehicle thermal radiation data and generate vehicle thermal feature trajectories. Multiple environmental sensors are deployed inside the parking lot to collect environmental data, including pollutant concentration and parking space occupancy status. Multiple lighting intelligent agents, each lighting intelligent agent corresponds to an LED lamp, and each lighting intelligent agent includes a first wireless communication module, a first edge computing module and a brightness adjustment module; Multiple ventilation intelligent agents, each ventilation intelligent agent corresponds to a jet fan or an exhaust valve, and each ventilation intelligent agent includes a second wireless communication module, a second edge computing module and a fan control module; Multiple smart lock agents, each smart lock agent corresponds to a parking space lock, and each smart lock agent includes a third wireless communication module, a third edge computing module and a lock driving module; The lighting agent, the ventilation agent, and the lock agent form a distributed agent network through a wireless mesh network, and local communication connections are established between agents in adjacent physical spaces. The first edge computing module of the lighting agent autonomously decides the brightness adjustment amount of the lighting agent based on the vehicle's thermal feature trajectory and the local observation information broadcast by neighboring lighting agents, through a pre-trained first multi-agent reinforcement learning model. The second edge computing module of the ventilation agent, based on pollutant concentration, vehicle thermal trajectory and local observation information broadcast by neighboring ventilation agents, autonomously decides the speed adjustment amount or valve opening adjustment amount of the ventilation agent through a pre-trained second multi-agent reinforcement learning model. The third edge computing module of the parking lock agent autonomously decides whether to raise, lower, or maintain the parking space based on the parking space occupancy status and local observation information broadcast by neighboring parking lock agents, through a pre-trained third multi-agent reinforcement learning model.

2. The smart property parking lot energy consumption optimization and management system based on big data analysis according to claim 1, characterized in that, The thermal imaging sensor is a low-resolution thermal imaging sensor. The vehicle thermal feature trajectory includes the coordinates of the center point of the heat cluster contour, the rate of change of the heat cluster area, the peak temperature of the heat cluster, the thermal radiation distribution entropy, and motion vector data. The motion vector data includes velocity, acceleration, and orientation angle.

3. The smart property parking lot energy consumption optimization and management system based on big data analysis according to claim 1, characterized in that, The system also includes a voiceprint recognition sensor, deployed above the parking space, for collecting vehicle start-up voiceprint data; The environmental sensor includes a gas sensor for collecting carbon monoxide concentration, PM2.5 concentration, nitrogen oxide concentration, or TVOC concentration. The system locates the coordinates of parking spaces at pollution sources based on pollutant concentration, vehicle start-up soundprint data, and vehicle engine type data.

4. The smart property parking lot energy consumption optimization and management system based on big data analysis according to claim 1, characterized in that, The system also includes a gate data interface for obtaining vehicle entry time, vehicle engine type and license plate information; The system constructs a parking space behavior profile based on the parking space occupancy status and vehicle entry time. The parking space behavior profile includes parking duration distribution, parking time period characteristics, and turnover rate data.

5. The smart property parking lot energy consumption optimization and management system based on big data analysis according to claim 1, characterized in that, The first edge computing module of the lighting agent also receives pollutant diffusion path information broadcast by the adjacent ventilation agent. When the area covered by the lighting agent is marked as a pollutant diffusion path, the lighting agent adjusts the brightness to a preset warning brightness value.

6. The smart property parking lot energy consumption optimization and management system based on big data analysis according to claim 1, characterized in that, The second edge computing module of the ventilation smart agent also receives charging status information broadcast by adjacent parking lock smart agents. When the charging space within the coverage area of ​​the ventilation smart agent is fully charged and the vehicle has not left for more than a preset threshold, the ventilation smart agent reduces the speed adjustment amount to the baseline operating value.

7. The smart property parking lot energy consumption optimization and management system based on big data analysis according to claim 1, characterized in that, The third edge computing module of the parking lock intelligent agent also receives the pollutant source tracing results broadcast by the adjacent ventilation intelligent agent. When the parking space corresponding to the parking lock intelligent agent is marked as a pollution source and the vehicle has finished charging, the parking lock intelligent agent performs a lifting action.

8. The smart property parking lot energy consumption optimization and management system based on big data analysis according to claim 1, characterized in that, The local observation information broadcast in the local communication connection includes the agent's current execution parameters, the agent's cumulative energy consumption, the predicted probability of vehicle arrival in the agent's coverage area, the pollutant concentration gradient in the agent's coverage area, or the parking space occupancy time in the agent's coverage area.

9. The smart property parking lot energy consumption optimization and management system based on big data analysis according to claim 1, characterized in that, The first multi-agent reinforcement learning model, the second multi-agent reinforcement learning model, and the third multi-agent reinforcement learning model adopt a centralized training-distributed execution framework. After the model training is completed on the cloud server, the execution network is deployed on the edge computing modules of each agent.

10. The smart property parking lot energy consumption optimization and management system based on big data analysis according to claim 1, characterized in that, The system also includes a cloud server, which is communicatively connected to edge nodes of multiple parking lots. Each edge node uploads the parameters of its locally trained multi-agent reinforcement learning model to the cloud server. The cloud server generates a globally optimized model using a federated averaging algorithm and distributes it to each edge node to update its local model.

Images