Parking lot intelligent operation analysis method based on Mama model

Abstract

The invention belongs to the technical field of parking lot management, and relates to a parking lot intelligent operation analysis method based on a Mama model, and the method comprises the steps: firstly collecting and processing multi-source data; performing feature extraction and preprocessing on the processed data to obtain multi-dimensional tensor data; then, constructing a Mama model, and outputting a predicted value of a future traffic flow, abnormal and normal behaviors and predicted values of future parking demands; the Mama model is trained and optimized, a solution and a report are finally generated, manual intervention is reduced, efficiency is improved, data are automatically processed and analyzed, dependence on manual operation is reduced, analysis efficiency and accuracy are improved, human errors are reduced, real-time operation reports and insights are provided, and the requirement for modern management is met.

Description

Technical Field

[0001] The invention belongs to the technical field of parking lot management and relates to a parking lot intelligent operation analysis method based on a Mamba model. Background Art

[0002] In the process of modern urbanization, parking lots have become an important part of urban traffic management. With the rapid development of smart cities and Internet of Things technologies, the market demand for intelligent parking lot operation analysis systems has become increasingly urgent. However, traditional parking lot management systems have the following technical problems in operation: First, incomplete data collection: parking lot systems can often only provide simple parking space status information, such as idle or occupied status, but cannot collect more complex operational data, such as order classification, traffic flow, length of stay, etc. Only monitoring individual indicators cannot integrate multiple data sources, and it is difficult to comprehensively analyze the overall operation of the parking lot. This limitation makes it impossible for managers to fully understand the operating status of the parking lot and formulate scientific optimization strategies; second, insufficient data analysis capabilities: the existing system lacks the ability to deeply analyze parking lot operation data, and it is difficult to provide effective operation indicator analysis and prediction, and it cannot help parking lot managers make scientific decisions. In addition, many parking lots rely on manual statistics and report generation for operational analysis. This method requires a lot of manpower and time. , and the data is prone to errors. Manual analysis methods lack the ability to conduct in-depth analysis of large-scale data, and it is difficult to identify complex patterns or abnormal behaviors, resulting in low operational efficiency; third, lack of real-time and intelligence: when faced with abnormal situations (such as invalid orders, abnormal data, etc.), traditional systems usually react slowly, lack real-time analysis and feedback capabilities, cannot respond to emergencies or abnormal situations in a timely manner, and cannot generate analysis reports and propose solutions in real time; fourth, information island phenomenon: some parking lots use simple automation systems, such as using cameras and license plate recognition technology to record vehicle entry and exit information. These systems can only perform basic data collection, and the data between equipment and systems are isolated. It is impossible to integrate and analyze multiple data sources, lack a global view of parking lot operations, and lead to information island phenomenon, which cannot fully support management decisions and form comprehensive operational insights. In summary, the existing parking lot management faces problems such as information islands, lack of real-time intelligent analysis, and inefficient manual intervention. Therefore, a new type of parking lot intelligent operation analysis method is urgently needed to improve the intelligent management level of parking lots, improve parking space utilization efficiency and user satisfaction. Summary of the invention

[0003] The purpose of the present invention is to overcome the shortcomings of the prior art and to provide a parking lot intelligent operation analysis method based on the Mamba model to improve the intelligent management level of the parking lot, and to improve the parking space utilization efficiency and user satisfaction.

[0004] To achieve the above object, the present invention adopts the following technical solution: A parking lot intelligent operation analysis method based on the Mamba model specifically includes the following steps: S1. Data collection and processing: collect multi-source data and process the data; S2, feature extraction and preprocessing: feature extraction and preprocessing are performed on the data processed in step S1 by a feature extraction and preprocessing unit to obtain multi-dimensional tensor data; S3, Mamba model construction: construct the Mamba model based on the long short-term memory network (LSTM) and the self-attention mechanism; input the multi-dimensional tensor data obtained in step S2 into the Mamba model, and output the predicted value of future traffic flow, abnormal and normal behavior, and predicted value of future parking demand; S4. Training and optimization of Mamba model: training and optimization of Mamba model; S5, solution and report generation: Based on the output results of step S4, automatically generate solutions and strategy recommendations, and generate reports with analysis results, predictions and recommended measures and distribute them to relevant decision makers and managers.

[0005] As a further technical solution of the present invention, the multi-source data described in step S1 is obtained through a sensor data acquisition unit, a video surveillance system acquisition unit, a user terminal device acquisition unit and an environment perception device acquisition unit, wherein the sensor data acquisition unit uses sensors installed at various entrances, lanes and parking spaces in the parking lot to obtain real-time vehicle entry and exit and parking space usage, the video surveillance system acquisition unit captures traffic flow and vehicle activities in the parking lot through cameras, the user terminal device acquisition unit collects user behavior data and transaction information through mobile phone applications, license plate recognition systems and payment terminals, and the environment perception device acquisition unit collects environmental data, which includes weather, temperature, and light for auxiliary analysis and prediction.

[0006] As a further technical solution of the present invention, the data processing process in step S1 includes: S11. Data cleaning: De-noise and clean the collected data to remove invalid, duplicate and abnormal data; S12. Data conversion: Format and convert data, standardize numerical data to a unified range, eliminate dimensional differences, and convert raw data into CSV or JSON standard formats for easy storage and analysis. At the same time, reorganize the data according to time series to facilitate time series analysis.

[0007] As a further technical solution of the present invention, the multidimensional tensor data in step S2 includes order information, total number of orders, normal orders, invalid orders, abnormal data, business orders, long-term rental data, length of stay, daily average parking rate in each period, daily average traffic flow in each period, equipment data and transaction source data, expressed as .

[0008] As a further technical solution of the present invention, the Mamba model constructed in step S3 includes an input layer, a time series modeling layer, a state space module (SSM), an attention calculation module, a fully connected layer and an output layer. The specific working process is as follows: Input layer: receives tensor data , and normalize and standardize them to eliminate the differences between different feature scales and ensure the training stability of the model. The normalization operation scales the feature value to the range of [0, 1], and the standardization operation adjusts the feature value to the mean. is 0, standard deviation is 1, and the formulas for normalization and standardization are as follows: , ,in is the normalized result, To standardize the results; Time series modeling layer: The time series modeling layer uses the long short-term memory network (LSTM) to model the input time series data to capture long-term dependencies and time patterns. Specifically, the number of hidden units in the time series modeling layer is 128, the number of layers is 2, and the Dropout (random inactivation) is set to 0.2. The update formula of the long short-term memory network (LSTM) is as follows: , , , , , in, They are input gate, forget gate, output gate and unit state. are network parameters, is the sigmoid function, For the moment Input, , are the outputs at the previous moment and this moment respectively; State Space Module (SSM): The self-attention mechanism is used to weight the features output by the time series modeling layer, identify important time periods and features, and improve the prediction performance of the model. The multi-head attention mechanism uses multiple attention heads to capture different time and feature dependencies. The number of attention heads is 8, and the input dimension is consistent with the output dimension of the time series modeling layer. Attention calculation module: The input sequence is transformed into query, key and value vectors through linear transformation, the similarity between query and key is calculated, attention weight is generated, and the value is weighted summed using attention weight to generate a new sequence representation. The attention calculation formula is expressed as follows: ,in, The output of LSTM Converted from is the vector K The dimensions are as follows: , , ; as well as They are the weight parameters and bias parameters in the attention calculation module respectively; Fully connected layer: The fully connected layer further nonlinearly transforms the features processed by the self-attention mechanism calculation module to generate the final prediction results; Output layer: The fully connected output layer outputs the prediction result after a linear activation function.

[0009] As a further technical solution of the present invention, the process of training and optimizing the Mamba model in step S4 is as follows: S41, data preparation: collect historical parking data and perform data preprocessing, including data cleaning, feature extraction and standardization, and generate training sets and test sets; S42. Model training: Use the Adam optimizer and the cross entropy loss function to train the model. Optimize the training efficiency by setting the learning rate and batch size, and use the early stopping strategy to prevent overfitting. S43, Model Verification: Use the test set to verify the trained model, evaluate the model's performance indicators through accuracy, recall rate, and F1-score, and adjust the model parameters according to the verification results to obtain a trained model; S44, Model deployment: Deploy the trained model to the cloud or edge computing device, and update the model regularly; S45. Optimization strategy hyperparameter optimization: Use Grid Search to adjust hyperparameters to find the best model configuration, and use L2 regularization or batch normalization technology to increase the generalization ability of the model.

[0010] As a further technical solution of the present invention, the solutions and strategy recommendations automatically generated in step S5 include dynamic adjustments based on predictions and abnormal response measures, wherein the dynamic adjustment based on predictions is to recommend adding temporary parking spaces or adjusting parking fees to manage demand when high traffic volume is predicted, and to provide discounts or promotions to attract more customers when low traffic volume is predicted; the abnormal response measures are to recommend immediate repair or replacement of detected equipment failures; for illegal parking, automatically notify security personnel or send warnings to users through applications; the generated reports use charts, trend lines and heat map visualization methods to display analysis results and predictions, and provide text-described analysis summaries, key indicator explanations and recommended action plans; the reports are distributed to relevant decision makers and managers via email, mobile application notifications or integration into management dashboards.

[0011] Compared with the prior art, the present invention has the following beneficial effects: (1) Enhanced real-time performance and intelligence. The Mamba model supports real-time data processing and can quickly identify and respond to emergencies or abnormal situations in the parking lot, improving the response speed and effectiveness of the system. In addition, by using advanced data analysis and prediction functions, the Mamba model can dynamically adjust operation strategies based on operation data, optimize resource allocation and user experience; (2) Data integration and integrity. The Mamba model is designed to support the integration of multiple data sources, including vehicle flow, parking space occupancy, payment information, etc., ensuring the integrity and comprehensiveness of the data. By integrating different data sources, the model can provide global operational insights and support more accurate and comprehensive decision-making. (3) Reduce manual intervention, improve efficiency, automatically process and analyze data, reduce reliance on manual operations, improve analysis efficiency and accuracy, reduce human errors, provide real-time operational reports and insights, and meet the needs of modern management; (4) Diversified indicators and global analysis capabilities. The Mamba model supports the analysis of diversified indicators. Using deep learning technology, the model can mine hidden patterns and associations in the data and provide deeper business insights than traditional systems. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] Figure 1 It is a flowchart of the workflow of the present invention. DETAILED DESCRIPTION

[0013] The present invention will be further described below by way of embodiments in conjunction with the accompanying drawings.

[0014] Example: like Figure 1 As shown, this embodiment provides a parking lot intelligent operation analysis method based on the Mamba model, which specifically includes the following steps: S1. Data collection and processing: Collect multi-source data and process the data; the multi-source data is obtained through the sensor data collection unit, the video surveillance system collection unit, the user terminal device collection unit and the environment perception device collection unit. The sensor data collection unit uses sensors installed at each entrance, lane and parking space of the parking lot to obtain real-time vehicle entry and exit and parking space usage. The video surveillance system collection unit uses cameras to capture traffic flow and vehicle activities in the parking lot. The user terminal device collection unit collects user behavior data and transaction information through mobile phone applications, license plate recognition systems and payment terminals. The environment perception device collection unit collects environmental data, including weather, temperature, and light, which are used to assist analysis and prediction; The data processing process includes data cleaning and data conversion. Data cleaning is to denoise and clean the collected data to remove invalid, duplicate and abnormal data; data conversion is to format and convert the data, standardize the numerical data to a unified range, eliminate dimensional differences, and convert the original data into CSV or JSON standard format for easy storage and analysis. At the same time, the data is reorganized according to time series to facilitate time series analysis.

[0015] S2. Feature extraction and preprocessing: The feature extraction and preprocessing unit extracts and preprocesses the data processed in step S1 to obtain multidimensional tensor data. The multidimensional tensor data includes order information, total number of orders, normal orders, invalid orders, abnormal data, business orders, long-term rental data, length of stay, daily average parking rate in each time period, daily average traffic flow in each time period, equipment data and transaction source data, expressed as ;

[0016] S3, Mamba model construction: construct the Mamba model based on the long short-term memory network (LSTM) and the self-attention mechanism (Self-Attention); input the multi-dimensional tensor data obtained in step S2 into the Mamba model, and output the predicted value of future traffic flow, abnormal and normal behavior, and predicted value of future parking demand; the constructed Mamba model includes an input layer, a time series modeling layer, a state space module (SSM), an attention calculation module, a fully connected layer, and an output layer. The specific working process is as follows: Input layer: receives tensor data , and normalize and standardize them to eliminate the differences between different feature scales and ensure the training stability of the model. The normalization operation scales the feature value to the range of [0, 1], and the standardization operation adjusts the feature value to the mean. is 0, standard deviation is 1, and the formulas for normalization and standardization are as follows: , ,in is the normalized result, To standardize the results; Time series modeling layer: The time series modeling layer uses the long short-term memory network (LSTM) to model the input time series data to capture long-term dependencies and time patterns. Specifically, the number of hidden units in the time series modeling layer is 128, the number of layers is 2, and the Dropout (random inactivation) is set to 0.2. The update formula of the long short-term memory network (LSTM) is as follows: , , , , , in, They are input gate, forget gate, output gate and unit state. are network parameters, is the sigmoid function, For the moment Input, , are the outputs at the previous moment and this moment respectively; The features are weighted to identify important time periods and features and improve the prediction performance of the model. The multi-head attention mechanism uses multiple attention heads to capture different time and feature dependencies. The number of attention heads is 8 and the input dimension is consistent with the output dimension of the time series modeling layer. Attention calculation module: The input sequence is transformed into query, key and value vectors through linear transformation, the similarity between query and key is calculated, attention weight is generated, and the value is weighted summed using attention weight to generate a new sequence representation. The attention calculation formula is expressed as follows: ,in, The output of LSTM Converted from is the vector K The dimensions are as follows: , , ; as well as They are the weight parameters and bias parameters in the attention calculation module respectively; Fully connected layer: The fully connected layer further nonlinearly transforms the features processed by the self-attention mechanism calculation module to generate the final prediction results; Output layer: The fully connected output layer outputs the prediction result after a linear activation function.

[0017] S4. Training and optimization of the Mamba model: Train and optimize the Mamba model, specifically: S41, data preparation: collect historical parking data and perform data preprocessing, including data cleaning, feature extraction and standardization, and generate training sets and test sets; S42. Model training: Use the Adam optimizer and the cross entropy loss function to train the model. Optimize the training efficiency by setting the learning rate and batch size, and use the early stopping strategy to prevent overfitting. S43, Model Verification: Use the test set to verify the trained model, evaluate the model's performance indicators through accuracy, recall rate, and F1-score, and adjust the model parameters according to the verification results to obtain a trained model; S44, Model deployment: Deploy the trained model to the cloud or edge computing device, and update the model regularly; S45. Optimization strategy hyperparameter optimization: Use Grid Search to adjust hyperparameters to find the best model configuration, and use L2 regularization or Batch Normalization technology to increase the generalization ability of the model.

[0018] S5. Solution and report generation: Based on the output results of step S4, solutions and strategy recommendations are automatically generated, and reports are generated based on the analysis results, predictions and recommended measures and distributed to relevant decision makers and managers. The generated solutions and strategy recommendations include dynamic adjustments based on predictions and abnormal response measures. The dynamic adjustment based on predictions is to recommend adding temporary parking spaces or adjusting parking fees to manage demand when high traffic volume is predicted, and to provide discounts or promotions to attract more customers when low traffic volume is predicted; abnormal response measures are to recommend immediate repair or replacement of detected equipment failures; for illegal parking, security personnel are automatically notified or warnings are sent to users through the application; the generated report uses charts, trend lines and heat map visualization methods to display the analysis results and predictions, and provides a text-descriptive analysis summary, key indicator explanations and recommended action plans; the report is distributed to relevant decision makers and managers via email, mobile application notifications or integration into the management dashboard.

[0019] The structures and algorithms not described in detail in this article are general techniques in the field.

[0020] The above are only preferred embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

[0021] Although the above describes the specific implementation mode of the present invention in conjunction with the accompanying drawings, it is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art on the basis of the technical solution of the present invention without creative work are still within the scope of protection of the present invention.

Claims

1. A parking lot intelligent operation analysis method based on the Mamba model, characterized in that: The specific steps include: S1. Data collection and processing: collect multi-source data and process the data; S2, feature extraction and preprocessing: feature extraction and preprocessing are performed on the data processed in step S1 by a feature extraction and preprocessing unit to obtain multi-dimensional tensor data; S3, Mamba model construction: construct the Mamba model based on the long short-term memory network and the self-attention mechanism; input the multi-dimensional tensor data obtained in step S2 into the Mamba model, and output the predicted value of future traffic flow, abnormal and normal behavior, and predicted value of future parking demand; S4. Training and optimization of Mamba model: training and optimization of Mamba model; S5, solution and report generation: Based on the output results of step S4, automatically generate solutions and strategy recommendations, and generate reports with analysis results, predictions and recommended measures and distribute them to relevant decision makers and managers.

2. The parking lot intelligent operation analysis method based on the Mamba model according to claim 1 is characterized in that: The multi-source data described in step S1 is obtained through a sensor data acquisition unit, a video surveillance system acquisition unit, a user terminal device acquisition unit and an environment perception device acquisition unit, wherein the sensor data acquisition unit uses sensors installed at various entrances, lanes and parking spaces in the parking lot to obtain real-time vehicle entry and exit and parking space usage conditions, the video surveillance system acquisition unit captures traffic flow and vehicle activities in the parking lot through cameras, the user terminal device acquisition unit collects user behavior data and transaction information through mobile phone applications, license plate recognition systems and payment terminals, and the environment perception device acquisition unit collects environmental data, which includes weather, temperature, and light, for auxiliary analysis and prediction.

3. The parking lot intelligent operation analysis method based on the Mamba model according to claim 2 is characterized in that: The data processing process in step S1 includes: S11. Data cleaning: De-noise and clean the collected data to remove invalid, duplicate and abnormal data; S12. Data conversion: Format and convert data, standardize numerical data to a unified range, eliminate dimensional differences, and convert raw data into CSV or JSON standard formats for easy storage and analysis. At the same time, reorganize the data according to time series to facilitate time series analysis.

4. The parking lot intelligent operation analysis method based on the Mamba model according to claim 3 is characterized in that: The multidimensional tensor data in step S2 includes order information, total number of orders, normal orders, invalid orders, abnormal data, business orders, long-term rental data, length of stay, daily average parking rate in each period, daily average traffic flow in each period, equipment data and transaction source data, expressed as .

5. The parking lot intelligent operation analysis method based on the Mamba model according to claim 4 is characterized in that: The Mamba model constructed in step S3 includes an input layer, a time series modeling layer, a state space module, an attention calculation module, a fully connected layer, and an output layer. The specific working process is as follows: Input layer: receives tensor data , and normalize and standardize them to eliminate the differences between different feature scales and ensure the training stability of the model. The normalization operation scales the feature value to the range of [0, 1], and the standardization operation adjusts the feature value to the mean. is 0, standard deviation is 1, and the formulas for normalization and standardization are as follows: , ,in is the normalized result, To standardize the results; Time series modeling layer: The time series modeling layer uses a long short-term memory network to model the input time series data to capture long-term dependencies and time patterns. Specifically, the number of hidden units in the time series modeling layer is 128, the number of layers is 2, and the random inactivation is set to 0.

2. The update formula of the long short-term memory network is as follows: ， ， ， ， ， in, They are input gate, forget gate, output gate and unit state. are network parameters, is the sigmoid function, For the moment Input, , are the outputs at the previous moment and this moment respectively; State space module: The features output by the time series modeling layer are weighted through the self-attention mechanism to identify important time periods and features and improve the prediction performance of the model. The multi-head attention mechanism uses multiple attention heads to capture different time and feature dependencies. The number of attention heads is 8, and the input dimension is consistent with the output dimension of the time series modeling layer. Attention calculation module: The input sequence is transformed into query, key and value vectors through linear transformation, the similarity between query and key is calculated, attention weight is generated, and the value is weighted summed using attention weight to generate a new sequence representation. The attention calculation formula is expressed as follows: ,in, The output of LSTM Converted from is the vector K The dimensions are as follows: , , ; as well as They are the weight parameters and bias parameters in the attention calculation module respectively; Fully connected layer: The fully connected layer further nonlinearly transforms the features processed by the self-attention mechanism calculation module to generate the final prediction results; Output layer: The fully connected output layer outputs the prediction result after a linear activation function.

6. The parking lot intelligent operation analysis method based on the Mamba model according to claim 5 is characterized in that: The process of training and optimizing the Mamba model in step S4 is as follows: S41, data preparation: collect historical parking data and perform data preprocessing, including data cleaning, feature extraction and standardization, and generate training sets and test sets; S42, Model training: Use the Adam optimizer and cross entropy loss function to train the model, optimize the training efficiency by setting the learning rate and batch size, and adopt the early stopping strategy to prevent overfitting; S43, Model Verification: Use the test set to verify the trained model, evaluate the model's performance indicators through accuracy, recall rate, and F1-score, and adjust the model parameters according to the verification results to obtain a trained model; S44, Model deployment: Deploy the trained model to the cloud or edge computing device, and update the model regularly; S45. Optimization strategy hyperparameter optimization: Use grid search to adjust hyperparameters to find the best model configuration, and use L2 regularization or batch normalization techniques to increase the generalization ability of the model.

7. The parking lot intelligent operation analysis method based on the Mamba model according to claim 6 is characterized in that: The solution and strategy recommendations automatically generated in step S5 include dynamic adjustments based on predictions and abnormal response measures, wherein the dynamic adjustments based on predictions are to recommend adding temporary parking spaces or adjusting parking fees to manage demand when high traffic volume is predicted, and to provide discounts or preferential activities to attract more customers when low traffic volume is predicted; Abnormal response measures include recommending immediate repair or replacement for detected equipment failures; automatically notifying security personnel or sending warnings to users through the app for illegal parking; generated reports use charts, trend lines, and heat map visualizations to display analysis results and predictions, and provide a text-descriptive analysis summary, key indicator explanations, and recommended action plans; reports are distributed to relevant decision makers and managers via email, mobile app notifications, or integration into management dashboards.

Images