Smart city monitoring system and method based on AI

Abstract

The invention provides an AI-based smart city monitoring system and method, and relates to the technical field of smart cities, and the system comprises a data analysis module which is used for carrying out the analysis of a city monitoring data flow according to a double-flow network, extracting structured data, and carrying out the correlation analysis, so as to obtain a correlation analysis result; the processing module is used for processing the structured data to obtain an event abstract and a statistical analysis report; and the abnormity monitoring module is used for monitoring and predicting abnormal events including urban pipe network faults, traffic congestion and public safety events in real time according to the analysis result output by the processing module, and sending out an early warning signal. According to the invention, through integration of data analysis, processing, abnormity monitoring and dynamic decision-making functions, intelligent management of city operation states is realized, and city management efficiency and response capability are improved.

Description

Technical Field

[0001] The present invention relates to the field of smart city technology, and in particular to an AI-based smart city monitoring system and method. Background Art

[0002] Traditional monitoring systems rely on manual monitoring and analysis, which is inadequate for large-scale urban monitoring data. Due to the limited energy of manual monitoring, it is difficult to continuously and efficiently process the large amount of data that is pouring in, resulting in low monitoring efficiency.

[0003] For example, in traffic monitoring, it may be difficult for humans to track and analyze traffic flows at all intersections in real time, and some key traffic jams or accidents may be missed, resulting in underreporting.

[0004] Existing technologies lack the ability to accurately identify and intelligently analyze urban monitoring data when processing and analyzing them, which makes it difficult for the system to extract valuable information from massive amounts of data and makes it impossible to achieve a comprehensive and in-depth analysis of the monitoring data.

[0005] For example, in environmental monitoring, traditional systems may only be able to simply record pollutant concentration data but cannot automatically analyze the changing trends, anomalies or potential environmental risks of these data, thereby limiting the early warning and decision support capabilities of the monitoring system. Summary of the invention

[0006] The technical problem to be solved by the present invention is to provide an AI-based smart city monitoring system and method, which improves the level of intelligence in urban management.

[0007] In order to solve the above technical problems, the technical solution of the present invention is as follows: In the first aspect, an AI-based smart city monitoring system includes: The data analysis module is used to analyze the urban monitoring data stream according to the dual-stream network, extract structured data and perform correlation analysis to obtain correlation analysis results; A processing module is used to process structured data to obtain event summaries and statistical analysis reports; The abnormal monitoring module is used to monitor and predict abnormal events including urban pipe network failures, traffic congestion, and public safety incidents in real time based on the analysis results output by the processing module, and issue early warning signals; The dynamic decision-making module is used to dynamically adjust the strategies and configurations of urban operation management based on the output of the anomaly monitoring module and the analysis results of the processing module, combined with the dynamic weight allocation mechanism of reinforcement learning, so as to realize intelligent management of the urban operation status.

[0008] Furthermore, the urban monitoring data stream is analyzed according to the dual-stream network, structured data is extracted and correlation analysis is performed to obtain correlation analysis results, including: Obtain data streams from urban monitoring equipment, including video streams, audio streams, and various physical quantity data in the urban pipe network; The data stream is analyzed using a two-stream network model, where the video stream and audio stream are respectively extracted through their own sub-networks to identify target information elements, including faces, vehicles, voice content, and preliminary behavior patterns, including movement direction and speech speed changes; The physical quantity data in the urban pipe network is analyzed through the physical quantum network to extract the data change trend and abnormal point characteristics, and identify the equipment status and flow change pattern; The video, audio and physical quantity features extracted by the dual-stream network are fused to form fused urban monitoring information. Based on the fused urban monitoring information, correlation analysis is performed to identify the movement trajectories of people, densely distributed areas of crowds, fault conditions of equipment and their impact on urban operations, and obtain correlation analysis results.

[0009] Furthermore, the structured data is processed to obtain event summaries and statistical analysis reports, including: Use event recognition algorithms to scan and analyze structured data to identify various types of urban events, including traffic accidents, densely populated areas of people, and equipment failures; Classify incidents according to their type, nature, and impact, and for each identified incident, extract key information, including the time, location, people or vehicles involved, and description of the incident; Based on key information, an event summary is generated, including time, location, subject and event overview, and a statistical analysis is performed on all event data over a period of time, including the distribution of event types, the frequency of occurrence of various types of events, the scope of impact of the events, and a statistical analysis report is generated.

[0010] Furthermore, according to the analysis results output by the processing module, abnormal events including urban pipe network failures, traffic congestion, and public safety incidents are monitored and predicted in real time, and early warning signals are issued, including: According to the type of abnormal event, including urban pipe network failure, traffic congestion, and public safety incidents, dynamically select relevant features, including timestamp, geographic location, event type, and sensor readings, to generate a feature data set of the abnormal event type; According to the feature data set of the abnormal event type, an isolation forest model is constructed, and the features in the historical data set and the corresponding abnormal event labels are used to train the isolation forest model to obtain a trained isolation forest model; Input the feature data in the real-time data stream into the trained isolation forest model and calculate the abnormality of each data point for real-time monitoring and prediction; The abnormality of each data point is compared with the preset threshold. When the abnormality of each data point is ≥ the preset threshold, the abnormal event detection mechanism is triggered and the abnormal event detection result is generated; Based on the abnormal event detection results, corresponding warning signals are generated, including the event type, location, expected impact range, and recommended response measures.

[0011] Furthermore, the abnormality of each data point includes: For each data point, on each tree in the isolation forest, calculate the path length of the corresponding data point, where the path length refers to the number of edges from the root node of the tree to the leaf node containing the data point; According to the path length of data points on all trees and the total number of trees, the average path length of data points is obtained; A normalization constant is defined, and an abnormality value of the data point is obtained according to the average path length of the data point and the normalization constant, where the abnormality value is between 0 and 1.

[0012] Furthermore, based on the output of the anomaly monitoring module and the analysis results of the processing module, combined with the dynamic weight allocation mechanism of reinforcement learning, the strategy and configuration of urban operation management are dynamically adjusted to achieve intelligent management of the urban operation status, including: Integrate and process the output of the abnormality monitoring module and the analysis results of the processing module to obtain the processed abnormal event analysis results; According to the analysis results of the abnormal events after processing, the strategy weights are dynamically adjusted using the reinforcement learning algorithm to obtain the adjusted strategy weights; According to the adjusted strategy weights, the configuration of urban operation management is dynamically adjusted, including adjusting the control logic of traffic lights and increasing investment in environmental protection facilities, so as to realize intelligent management of the urban operation status.

[0013] Furthermore, according to the analysis results of the abnormal events after processing, the strategy weights are dynamically adjusted using the reinforcement learning algorithm to obtain the adjusted strategy weights, including: Initialize the parameters of the reinforcement learning algorithm, including the Q-value function of the state-action pair, the learning rate, and the discount factor; Define the state space, action space and reward function, and select an action, i.e., an adjustment plan for the strategy weight, based on the current state and the strategy of the reinforcement learning model; Apply the determined strategy weight adjustment scheme to the city operation management strategy. After executing the action, record the new status and reward after executing the action. Based on the new state and reward, update the Q-value function using the update rule of the reinforcement learning algorithm; Repeat the process of selecting a strategy weight adjustment plan, applying the plan, and updating the Q-value function. When the preset number of iterations is reached, the adjusted strategy weight is output.

[0014] In the second aspect, a smart city monitoring method based on AI includes: Analyze the urban monitoring data stream according to the dual-stream network, extract structured data and perform correlation analysis to obtain correlation analysis results; Process structured data to obtain event summaries and statistical analysis reports; Based on event summaries and statistical analysis reports, real-time monitoring and prediction of urban pipe network failures, traffic congestion, and public safety incidents are carried out, and early warning signals are issued when abnormal events are identified; According to the output and analysis results of the abnormal event monitoring process, the strategies and configurations of urban operation management are dynamically adjusted through the dynamic weight allocation mechanism of reinforcement learning to achieve intelligent management of the city's operation status.

[0015] According to a third aspect, a computing device includes: one or more processors; The storage device is used to store one or more programs. When the one or more programs are executed by the one or more processors, the one or more processors implement the system.

[0016] In a fourth aspect, a computer-readable storage medium stores a program, and the program implements the system when executed by a processor.

[0017] The above solution of the present invention includes at least the following beneficial effects: Through the dual-stream network of the data analysis module, the urban monitoring data stream is deeply analyzed. The system can automatically extract structured data and perform correlation analysis, which greatly improves the intelligence level of urban monitoring. This enables urban management departments to obtain various information on urban operations more quickly and accurately, providing strong support for decision-making.

[0018] The processing module processes structured data and generates event summaries and statistical analysis reports, so that urban management departments can quickly understand the overview and occurrence patterns of events, thereby improving the efficiency of event processing. At the same time, the abnormal monitoring module can monitor and predict abnormal events such as urban pipe network failures, traffic congestion, and public safety incidents in real time, and issue early warning signals, enhancing the city's emergency response capabilities. Through statistical analysis reports, urban management departments can fully understand the occurrence and impact of various types of events in the city, providing a scientific basis for optimizing resource allocation. The dynamic decision-making module combines the dynamic weight allocation mechanism of reinforcement learning, which can dynamically adjust the strategies and configurations of urban operation management according to the real-time changes in the city's operation status, and achieve the optimal allocation of resources and the scientific formulation of management strategies.

[0019] Through functions such as correlation analysis, anomaly monitoring and dynamic decision-making, comprehensive, accurate and real-time monitoring of the city's operating status can be achieved, which will help urban management departments to promptly discover and deal with problems in urban operations, improve the accuracy and effectiveness of urban management, and ensure the safe and orderly operation of the city. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 This is a schematic diagram of an AI-based smart city monitoring system provided by an embodiment of the present invention.

[0021] Figure 2 It is a flow chart of an AI-based smart city monitoring method provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0022] The exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although the exemplary embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the present disclosure to those skilled in the art.

[0023] like Figure 1 As shown, an embodiment of the present invention proposes an AI-based smart city monitoring system, comprising: The data analysis module 1 is used to analyze the urban monitoring data stream according to the dual-stream network, extract structured data and perform correlation analysis to obtain correlation analysis results; Processing module 2, used to process structured data to obtain event summaries and statistical analysis reports; The abnormal monitoring module 3 is used to monitor and predict abnormal events including urban pipe network failures, traffic congestion, and public safety incidents in real time according to the analysis results output by the processing module, and issue early warning signals; The dynamic decision-making module 4 is used to dynamically adjust the strategy and configuration of urban operation management according to the output of the abnormal monitoring module and the analysis results of the processing module, combined with the dynamic weight allocation mechanism of reinforcement learning, so as to realize intelligent management of the urban operation status.

[0024] In the embodiment of the present invention, the deep analysis of the urban monitoring data stream through the dual-stream network can efficiently extract structured data. The realization of the association analysis function helps to explore the potential connections and laws in the urban monitoring data and provide a scientific basis for urban management and decision-making. The structured data is systematically processed to generate a concise and clear event summary, which is convenient for managers to quickly understand the operation status of the city. The provision of statistical analysis reports enables managers to fully understand the occurrence of various events in the city and provide data support for the formulation of effective management strategies.

[0025] Real-time monitoring and prediction of abnormal events such as urban pipe network failures, traffic congestion, and public safety incidents improves the emergency response capabilities of urban management. The timely issuance of early warning signals helps managers take quick measures to reduce the impact of abnormal events on urban operations and ensure urban safety and stability. Combined with the dynamic weight allocation mechanism of reinforcement learning, it is possible to dynamically adjust management strategies and configurations according to real-time changes in urban operation status to achieve intelligent management. By dynamically adjusting urban operation management strategies, it is possible to optimize resource allocation, improve urban management efficiency, and enhance the overall level and quality of urban operations.

[0026] In a preferred embodiment of the present invention, the urban monitoring data stream is analyzed according to the dual-stream network, structured data is extracted and correlation analysis is performed to obtain correlation analysis results, which may include: Obtain data streams from urban monitoring equipment, including video streams, audio streams, and various physical quantity data in the urban pipe network; The data stream is analyzed using a two-stream network model, where the video stream and audio stream are respectively extracted through their own sub-networks to identify target information elements, including faces, vehicles, voice content, and preliminary behavior patterns, including movement direction and speech speed changes; The physical quantity data in the urban pipe network is analyzed through the physical quantum network to extract the data change trend and abnormal point characteristics, and identify the equipment status and flow change pattern; The video, audio and physical quantity features extracted by the dual-stream network are fused to form fused urban monitoring information. Based on the fused urban monitoring information, correlation analysis is performed to identify the movement trajectories of people, densely distributed areas of crowds, fault conditions of equipment and their impact on urban operations, and obtain correlation analysis results.

[0027] In an embodiment of the present invention, data streams from various monitoring devices are received in real time through an interface connection with a city monitoring system. The data streams include but are not limited to video streams, audio streams, and various physical quantity data (such as pressure, flow, temperature, etc.) in the city pipe network, and these data streams are preliminarily processed and formatted. The video stream and the audio stream are respectively input into the video subnetwork and the audio subnetwork of the dual-stream network model. The video subnetwork extracts features from the video stream, identifies target information elements such as faces and vehicles, and preliminary behavior patterns such as the moving direction and speed of the target. The audio subnetwork extracts features from the audio stream, identifies audio features such as voice content and speech speed changes, and at the same time, the physical quantity data in the city pipe network is input into the physical quantum network to extract the data change trend, abnormal point features, and equipment status, flow change pattern and other information. The features extracted by the video subnetwork, the audio subnetwork and the physical quantum network are fused to form fused city monitoring information.

[0028] The fusion process may include feature splicing, weighted summation or more complex fusion algorithms to ensure that the fused information can fully and accurately reflect the actual situation of urban monitoring. The fused urban monitoring information is subjected to correlation analysis to identify spatial correlation information such as the movement trajectory of personnel and the dense distribution area of ​​the crowd. At the same time, the fault status of the equipment and its impact on the operation of the city, as well as the time correlation and causal relationship between various events are analyzed. Through correlation analysis, the correlation analysis results are obtained.

[0029] Assume that the city monitoring equipment captures a video stream, an audio stream, and a set of physical quantity data in the city pipe network.

[0030] Video stream: Shows a densely populated area of ​​people, with several people talking.

[0031] Audio stream: Contains a recording of the conversation, where they can be heard discussing a topic.

[0032] Physical quantity data: Shows that the pressure in a nearby water pipe suddenly dropped.

[0033] These data streams are analyzed through a two-stream network model: Video sub-network: Identifies densely populated areas of people and conversation scenes, as well as the facial features of people participating in the conversation.

[0034] Audio sub-network: extracts the voice content of the conversation, as well as audio features such as speaking speed and intonation.

[0035] Physical quantum network: An abnormal drop in water pipe pressure was detected, indicating that there may be equipment failure.

[0036] These features are combined and analyzed: Identify where and when there are dense areas of people.

[0037] Analyze the conversation content and find out about a topic related to the operation of the city.

[0038] It is judged that the abnormal drop in water pipe pressure may be related to nearby construction activities and may affect the city's water supply.

[0039] Finally, the correlation analysis results were obtained, which provided timely warning and decision-making basis for urban management departments.

[0040] Through the dual-stream network model, the machine can automatically identify and understand various information elements and events in urban operation, improve the intelligence level of urban monitoring, and can monitor and predict abnormal events such as urban pipe network failures, traffic congestion, and public safety incidents in real time, and issue early warning signals in time to enhance the emergency response capabilities of urban management. Through correlation analysis, the potential connections and patterns in urban monitoring data can be mined, providing a scientific basis for urban management departments to optimize urban resource allocation and decision-making. Intelligent analysis and early warning capabilities help urban management departments to promptly discover and deal with problems in urban operations and improve the overall efficiency and safety of urban operations.

[0041] In a preferred embodiment of the present invention, the structured data is processed to obtain an event summary and a statistical analysis report, which may include: Use event recognition algorithms to scan and analyze structured data to identify various types of urban events, including traffic accidents, densely populated areas of people, and equipment failures; Classify incidents according to their type, nature, and impact, and for each identified incident, extract key information, including the time, location, people or vehicles involved, and description of the incident; Based on key information, an event summary is generated, including time, location, subject and event overview, and a statistical analysis is performed on all event data over a period of time, including the distribution of event types, the frequency of occurrence of various types of events, the scope of impact of the events, and a statistical analysis report is generated.

[0042] In an embodiment of the present invention, an event recognition algorithm is started, which is designed to automatically scan and analyze specific patterns and features in structured data. The algorithm checks the structured data one by one, and identifies possible urban events such as traffic accidents, densely distributed areas of people, equipment failures, etc. by matching the predefined event type feature library. During the recognition process, the algorithm uses the timestamp, location information, keywords, etc. in the data to determine the event type and preliminarily evaluate the nature and impact of the event. The events are classified according to the identified event type, nature (such as emergency, non-emergency) and impact (such as minor, serious). The classification process involves further analysis of the event description to determine the accurate category and priority of the event. After the classification is completed, a unique identifier is assigned to each event. For each classified event, key information is extracted, including the time and location of the event, the people or vehicles involved, and a detailed description of the event. The process of extracting key information involves further parsing and formatting the original data to ensure the accuracy and readability of the information.

[0043] After the extraction is completed, the key information is stored in a structured database. Based on the extracted key information, a concise and clear event summary is generated for each event, including time, location, subject (such as the people or vehicles involved) and event overview. At the same time, all event data within a period of time are statistically analyzed to calculate indicators such as the distribution of event types, the frequency of occurrence of various types of events, and the scope of impact of events. After the statistical analysis is completed, a detailed statistical analysis report is generated to provide decision-making support for urban management departments.

[0044] Suppose the city surveillance system identifies the following three events in one day: Event 1: A traffic accident occurred at 9 a.m. at an intersection in the city center, involving two cars and causing minor traffic congestion.

[0045] Event 2: An incident occurred in a densely populated area at 3 p.m. in a square in a city park, involving about 500 people, due to a charity performance.

[0046] Event three: Equipment failure occurred at 8 pm at a water supply pump station. The failure type was water pump damage, affecting residents within the water supply area of ​​the pump station.

[0047] Based on the key information of these events, the following event summary and statistical analysis reports were generated: Event Summary: Event 1: At 9 a.m., a traffic accident occurred at an intersection in the city center. Two cars collided, causing minor traffic congestion.

[0048] Event 2: At 3 p.m., there was a densely populated area in the City Park Square, with about 500 people participating in a charity performance.

[0049] Event three: At 8 p.m., a water pump failure occurred at the water supply pump station, affecting residents within the water supply area of ​​the pump station.

[0050] Statistical analysis report: Distribution of event types: 1 traffic accident, 1 dense crowd distribution, and 1 equipment failure.

[0051] Frequency of events: Each type of event occurred once.

[0052] Scope of impact of the incident: Traffic accidents affect traffic flow, dense crowds affect order in public places, and equipment failures affect water supply safety.

[0053] By automatically generating event summaries and statistical analysis reports, the machine can greatly shorten the time for event processing, improve the work efficiency of urban management departments, identify and classify urban events in real time, provide timely early warning and response basis for urban management departments, and enhance the city's emergency response capabilities. Through statistical analysis reports, urban management departments can understand the frequency and impact of various events, thereby optimizing resource allocation and improving the pertinence and effectiveness of urban management. The generated event summaries and statistical analysis reports provide rich data support for urban management departments, which helps to formulate more scientific and reasonable decision-making plans and improve the overall level of urban management.

[0054] In a preferred embodiment of the present invention, according to the analysis results output by the processing module, real-time monitoring and prediction of abnormal events including urban pipe network failures, traffic congestion, and public safety incidents, and issuing early warning signals may include: According to the type of abnormal event, including urban pipe network failure, traffic congestion, and public safety incidents, dynamically select relevant features, including timestamp, geographic location, event type, and sensor readings, to generate a feature data set of the abnormal event type; According to the feature data set of the abnormal event type, an isolation forest model is constructed, and the features in the historical data set and the corresponding abnormal event labels are used to train the isolation forest model to obtain a trained isolation forest model; Input the feature data in the real-time data stream into the trained isolation forest model and calculate the abnormality of each data point for real-time monitoring and prediction; The abnormality of each data point is compared with the preset threshold. When the abnormality of each data point is ≥ the preset threshold, the abnormal event detection mechanism is triggered and the abnormal event detection result is generated; Based on the abnormal event detection results, corresponding warning signals are generated, including the event type, location, expected impact range, and recommended response measures.

[0055] In an embodiment of the present invention, the types of abnormal events that need to be monitored and predicted are first identified, such as urban pipe network failures, traffic congestion, public safety incidents, etc. For each type of abnormal event, the features related to it are dynamically selected. These features include timestamp (specific time of event occurrence), geographic location (location of event occurrence), event type (such as specific type of pipe network failure, severity of traffic congestion, etc.), sensor readings (such as pipe network pressure, flow and other physical quantity data). These features are combined into a feature data set. Isolation Forest is selected as the anomaly monitoring model, which is suitable for processing high-dimensional data and can effectively identify anomalies. The features in the historical data set and the corresponding abnormal event labels are used to construct the isolation forest model. The historical data set contains abnormal events that occurred in the past and their related features.

[0056] Through the training process, the isolation forest model can learn the difference between normal data and abnormal data. The features in the historical data set are input into the isolation forest model, and the parameters of the model are adjusted according to the corresponding abnormal event labels. After multiple iterations of training, a trained isolation forest model is obtained, which can accurately identify abnormal data points. The feature data streams from the city monitoring equipment are received in real time. These feature data streams contain physical quantity data and traffic data that need to be monitored. The feature data in the real-time data stream is input into the trained isolation forest model to calculate the abnormality of each data point. The isolation forest model calculates the abnormality of each data point (that is, the degree of difference between the data point and the normal data point) based on the input feature data. The higher the abnormality, the more likely the data point is an abnormal point.

[0057] Based on the calculated abnormality, abnormal events in the data stream are monitored in real time. When the abnormality of a data point exceeds the preset threshold, the abnormal event detection mechanism is triggered. Based on the triggered abnormal event detection mechanism, the abnormal data points are further analyzed to determine the type of abnormal event, the location of occurrence and other detailed information, and the abnormal event detection results are generated, including the event type, location of occurrence, and expected impact range. Based on the abnormal event detection results, the corresponding early warning signal is generated, which includes the event type, location of occurrence, expected impact range and recommended response measures, so that city managers can respond to and handle abnormal events in a timely manner.

[0058] Assume that the city monitoring equipment collects the pipe network pressure data in real time: Identify the abnormal event type that needs to be monitored as urban pipe network failure, dynamically select features related to timestamp, geographic location, pipe network pressure, etc., and generate a feature data set. Use the features in the historical pipe network failure data set and the corresponding fault labels to build an isolation forest model and train it. Input the real-time pipe network pressure data stream into the trained isolation forest model, calculate the abnormality of each data point, and when the abnormality of a data point exceeds the preset threshold, trigger the abnormal event detection mechanism to further analyze the abnormal data point. Determine the abnormal event as a pipe network failure, and generate abnormal event detection results, including the failure type, location, and expected impact range. Generate an early warning signal based on the detection results to notify city managers to respond to and handle pipe network failures in a timely manner.

[0059] By dynamically selecting relevant features and constructing an isolation forest model, abnormal events can be accurately identified and detection accuracy can be improved, which helps reduce false positives and false negatives and ensures that city managers can obtain accurate abnormal event information in a timely manner. It can monitor abnormal events in data streams in real time and issue warning signals immediately when abnormal events occur, which helps city managers respond to and handle abnormal events in a timely manner and reduce their impact on the lives of urban residents. The generated abnormal event detection results and warning signals can provide detailed and accurate information support for city managers, which helps city managers formulate more refined management strategies, optimize urban operation efficiency, and improve the quality of life of urban residents. By monitoring and predicting abnormal events in real time, it can help city managers improve emergency response speed, which helps reduce the harm of abnormal events to the lives of urban residents and ensure the safety and stability of the city.

[0060] In a preferred embodiment of the present invention, the abnormality of each data point includes: For each data point, on each tree in the isolation forest, calculate the path length of the corresponding data point, where the path length refers to the number of edges from the root node of the tree to the leaf node containing the data point; According to the path length of data points on all trees and the total number of trees, the average path length of data points is obtained; A normalization constant is defined, and an abnormality value of the data point is obtained according to the average path length of the data point and the normalization constant, where the abnormality value is between 0 and 1.

[0061] In the embodiment of the present invention, for each data point , search on each tree in the isolation forest, starting from the root node of the tree, traverse down along the path containing the data point until reaching the leaf node containing the data point, and record the number of edges from the root node to the leaf node. This number of edges is the path length of the data point on the tree, recorded as ,in, Represents the first A tree.

[0062] Add up the path lengths of all the trees obtained. ,in, is the total number of trees in the isolation forest. The total path length after addition is divided by the total number of trees in the isolation forest to obtain the average path length of the data point. According to the distribution of the average path length of all data points in the isolation forest, a normalization constant is defined. This normalization constant is used to standardize the average path length of the data points to a comparable range. In the isolation forest, a normalization constant is derived based on the average path length formula of the binary tree, specifically: ;in, is the first Item and, is the total number of all data points in the isolation forest; is the total number of trees in the isolation forest.

[0063] Divide the average path length of the data points by the normalization constant to get the ratio , take the negative ratio of 2 to the power, that is, get the abnormality value of the data point . Abnormality value The value is between 0 and 1, and the closer the value is to 1, the more likely the data point is an outlier.

[0064] The anomaly value calculated by the formula can quantify the degree of anomaly of each data point, which helps city managers to understand the severity and urgency of abnormal events more accurately, so as to make more reasonable response decisions. The isolation forest model combined with the above anomaly calculation formula can detect anomalies in the data stream more sensitively, which helps to timely discover and warn potential abnormal events and reduce their impact on the lives of urban residents. The anomaly value provides an intuitive indicator for measuring the degree of difference between a data point and a normal data point, which helps city managers to better understand the results of abnormal monitoring and formulate more effective management strategies accordingly. According to the actual situation and the type of abnormal event, city managers can dynamically adjust the threshold of the anomaly value, which helps to adapt to different application scenarios and needs more flexibly and improve the accuracy and efficiency of abnormal monitoring. The anomaly value not only provides the identification results of the anomaly point, but also provides a basis for in-depth analysis of abnormal events. City managers can combine the anomaly value and other relevant information to conduct a more comprehensive analysis and evaluation of abnormal events, so as to formulate more accurate response measures.

[0065] In a preferred embodiment of the present invention, according to the output of the abnormal monitoring module and the analysis result of the processing module, combined with the dynamic weight allocation mechanism of reinforcement learning, the strategy and configuration of urban operation management are dynamically adjusted to realize the intelligent management of the urban operation status, which may include: Integrate and process the output of the abnormality monitoring module and the analysis results of the processing module to obtain the processed abnormal event analysis results; According to the analysis results of the abnormal events after processing, the strategy weights are dynamically adjusted using the reinforcement learning algorithm to obtain the adjusted strategy weights; According to the adjusted strategy weights, the configuration of urban operation management is dynamically adjusted, including adjusting the control logic of traffic lights and increasing investment in environmental protection facilities, so as to realize intelligent management of the urban operation status.

[0066] In an embodiment of the present invention, abnormal event data is received from the abnormal monitoring module, and these data include information such as the type, location, and time of the abnormal event. At the same time, analysis results are received from the processing module, and these results are further processing of the abnormal event data, such as the severity of the abnormal event, the scope of impact, etc. The abnormal event data and the analysis results of the processing module are integrated to remove duplicate information to ensure the accuracy and consistency of the data. The integrated data is processed, such as data cleaning, format conversion, etc. A set of policy weights are defined, and these weights represent the importance of different city operation management strategies.

[0067] Use reinforcement learning algorithms, such as Q-learning or deep reinforcement learning (DRL), to learn and adjust policy weights. In the reinforcement learning process, the processed abnormal event analysis results are used as state inputs and the policy weights are used as action outputs.

[0068] By interacting with the environment (i.e., the city operation status), reward signals are obtained, such as the improvement of city operation efficiency, the reduction of the number of abnormal events, etc. The strategy weights are updated according to the reward signals to find the final strategy combination. Based on the adjusted strategy weights, the execution priority of different city operation management strategies is determined.

[0069] Dynamically adjust the control logic of traffic lights, such as increasing green light time and reducing red light time, to optimize traffic flow; increase investment in environmental protection facilities, such as adding garbage recycling stations and improving sewage treatment efficiency, to improve the urban environment.

[0070] Suppose a serious traffic accident occurs in the city. The abnormal monitoring module detects the incident and receives the analysis results of the incident from the processing module, including the severity of the accident, the scope of impact, etc.

[0071] The abnormal event data and the analysis results of the processing module are integrated and processed to obtain the processed abnormal event analysis results. Then, the reinforcement learning algorithm is used to dynamically adjust the policy weights according to the processed abnormal event analysis results. In this example, the weight of the traffic management policy may be increased to reduce the occurrence of similar traffic accidents. According to the adjusted policy weights, the configuration of urban operation management is dynamically adjusted. For example, the control logic of the traffic lights will be adjusted to increase the green light time of the accident section to reduce traffic congestion.

[0072] By dynamically adjusting the policy weights and city operation management configuration, it is possible to respond to abnormal events more quickly, reduce processing time, and improve the efficiency of city operation management. It is possible to dynamically adjust resource allocation according to the severity and scope of the abnormal event, such as increasing investment in environmental protection facilities, so as to make more effective use of urban resources. By dynamically adjusting measures such as traffic light control logic, it is possible to improve urban safety and reduce the occurrence of traffic accidents. It is possible to increase investment in environmental protection facilities and improve the processing capacity of the urban environment, such as garbage recycling and sewage treatment, thereby improving the quality of the urban environment. Through the application of reinforcement learning algorithms, it is possible to continuously learn and optimize policy weights and city operation management configurations, making urban operation management more intelligent and adaptive.

[0073] In another preferred embodiment of the present invention, according to the processed abnormal event analysis results, the strategy weight is dynamically adjusted using a reinforcement learning algorithm to obtain the adjusted strategy weight, which may include: Initialize the parameters of the reinforcement learning algorithm, including the Q-value function of the state-action pair, the learning rate, and the discount factor; Define the state space, action space and reward function, and select an action, i.e., an adjustment plan for the strategy weight, based on the current state and the strategy of the reinforcement learning model; Apply the determined strategy weight adjustment scheme to the city operation management strategy. After executing the action, record the new status and reward after executing the action. Based on the new state and reward, update the Q-value function using the update rule of the reinforcement learning algorithm; Repeat the process of selecting a strategy weight adjustment plan, applying the plan, and updating the Q-value function. When the preset number of iterations is reached, the adjusted strategy weight is output.

[0074] In an embodiment of the present invention, a Q-value function is initialized, which is a function that maps state-action to expected reward, and is represented by a table or a neural network. A learning rate is set, which determines the proportion of new information each time the Q-value function is updated. A discount factor is set, which determines the importance of future rewards in the current decision. The state space is defined as the set of all possible city operation states, such as traffic flow, air quality, crime rate, etc. The action space is defined as the set of all possible policy weight adjustment schemes, such as increasing or decreasing the weight of a certain policy. The reward function is defined as a quantitative assessment of the degree of improvement of the city's operating state based on the current state and action, such as improved traffic flow, reduced environmental pollution, etc. Observe the current city operation state, that is, the current state. Use a reinforcement learning model (such as the ε-greedy strategy) to select an action, that is, an adjustment plan for the policy weight. This selection may be based on exploration (random selection) or exploitation (selecting the action with the highest current Q value).

[0075] Apply the selected strategy weight adjustment scheme to the city operation management strategy, such as adjusting the control logic of traffic lights, the investment in environmental protection facilities, etc. Execute the adjusted city operation management strategy and observe the city operation status after execution, that is, the new status. According to the new status and the preset reward function, calculate the reward obtained by performing the action. Use the update rule of the reinforcement learning algorithm (such as the update formula of Q-learning) to update the Q-value function, taking the new status and reward information into consideration, repeat the above steps, continuously select new strategy weight adjustment schemes, apply the schemes, and update the Q-value function until the preset number of iterations is reached or other termination conditions are met. After reaching the preset number of iterations, output the final adjusted strategy weights, which represent the optimized city operation management strategy.

[0076] Suppose you are managing a city’s traffic system with the goal of reducing traffic congestion. You have initialized the parameters of the reinforcement learning algorithm, and defined the state space (such as traffic flow), action space (such as adjusting the control logic of traffic lights), and reward function (such as improving traffic smoothness).

[0077] At a certain moment, it is observed that the current traffic flow is large, that is, the current state. According to the strategy of the reinforcement learning model, an action is selected, that is, adjusting the traffic light control logic of a certain section of road to increase the green light time, applying this adjustment plan to the urban operation management strategy, and observing the traffic flow after execution, that is, the new state. According to the new state and the preset reward function, the reward obtained by performing the action is calculated, such as improved traffic flow. Then, the Q-value function is updated using the update rules of the reinforcement learning algorithm, taking into account the new state and reward information. Repeat the above steps, continuously select new policy weight adjustment plans, apply the plans, and update the Q-value function. After multiple iterations, the final adjusted policy weights are output, which represent the optimized traffic management strategy.

[0078] Through continuous learning and optimization of the reinforcement learning algorithm, the strategy weight adjustment scheme can be selected more accurately, thereby improving the efficiency of urban operation management. The reinforcement learning algorithm can adapt to the dynamic changes in the city's operation status, such as fluctuations in traffic flow and deterioration of air quality, thereby maintaining the effectiveness of urban operation management. It can dynamically adjust resource allocation according to the current city operation status and the strategy of the reinforcement learning model, such as increasing or decreasing the control logic adjustment frequency of traffic lights, so as to make more effective use of urban resources. By continuously optimizing the strategy weight adjustment scheme, the quality of urban operation can be improved, such as reducing traffic congestion and improving air quality, thereby improving the quality of life of citizens. The application of reinforcement learning algorithms in urban operation management enhances the practicality of learning and provides new ideas and methods for intelligent urban management.

[0079] like Figure 2 As shown, an embodiment of the present invention further provides an AI-based smart city monitoring method, comprising: Analyze the urban monitoring data stream according to the dual-stream network, extract structured data and perform correlation analysis to obtain correlation analysis results; Process structured data to obtain event summaries and statistical analysis reports; Based on event summaries and statistical analysis reports, real-time monitoring and prediction of urban pipe network failures, traffic congestion, and public safety incidents are carried out, and early warning signals are issued when abnormal events are identified; According to the output and analysis results of the abnormal event monitoring process, the strategies and configurations of urban operation management are dynamically adjusted through the dynamic weight allocation mechanism of reinforcement learning to achieve intelligent management of the city's operation status.

[0080] It should be noted that this method is a method corresponding to the above-mentioned system, and all implementation methods in the above-mentioned system embodiment are applicable to this embodiment and can achieve the same technical effect.

[0081] The embodiment of the present invention further provides a computing device, including: a processor, a memory storing a computer program, and when the computer program is executed by the processor, the system described above is executed. All implementation methods in the above system embodiment are applicable to this embodiment and can achieve the same technical effect.

[0082] The embodiment of the present invention further provides a computer-readable storage medium storing instructions, which, when executed on a computer, enable the computer to execute the system described above. All implementations in the above system embodiment are applicable to this embodiment and can achieve the same technical effect.

[0083] The above is a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. An AI-based smart city monitoring system, characterized in that: include: The data analysis module is used to analyze the urban monitoring data stream according to the dual-stream network, extract structured data and perform correlation analysis to obtain correlation analysis results; A processing module is used to process structured data to obtain event summaries and statistical analysis reports; The abnormality monitoring module is used to monitor and predict abnormal events including urban pipe network failures, traffic congestion, and public safety incidents in real time based on the analysis results output by the processing module, and issue early warning signals; The dynamic decision-making module is used to dynamically adjust the strategies and configurations of urban operation management based on the output of the anomaly monitoring module and the analysis results of the processing module, combined with the dynamic weight allocation mechanism of reinforcement learning, so as to realize intelligent management of the urban operation status.

2. The AI-based smart city monitoring system according to claim 1 is characterized in that: The urban monitoring data stream is analyzed according to the dual-stream network, structured data is extracted and correlation analysis is performed to obtain correlation analysis results, including: Obtain data streams from urban monitoring equipment, including video streams, audio streams, and various physical quantity data in the urban pipe network; The data stream is analyzed using a two-stream network model, where the video stream and audio stream are respectively extracted through their own sub-networks to identify target information elements, including faces, vehicles, voice content, and preliminary behavior patterns, including movement direction and speech speed changes; The physical quantity data in the urban pipe network is analyzed through the physical quantum network to extract the data change trend and abnormal point characteristics, and identify the equipment status and flow change pattern; The video, audio and physical quantity features extracted by the dual-stream network are fused to form fused urban monitoring information. Based on the fused urban monitoring information, correlation analysis is performed to identify the movement trajectories of people, densely distributed areas of crowds, fault conditions of equipment and their impact on urban operations, and obtain correlation analysis results.

3. The AI-based smart city monitoring system according to claim 2 is characterized in that: Process structured data to obtain event summaries and statistical analysis reports, including: Use event recognition algorithms to scan and analyze structured data to identify various types of urban events, including traffic accidents, densely populated areas of people, and equipment failures; Classify incidents according to their type, nature, and impact, and for each identified incident, extract key information, including the time, location, people or vehicles involved, and description of the incident; Based on key information, an event summary is generated, including time, location, subject and event overview, and a statistical analysis is performed on all event data over a period of time, including the distribution of event types, the frequency of occurrence of various types of events, the scope of impact of the events, and a statistical analysis report is generated.

4. The AI-based smart city monitoring system according to claim 3 is characterized in that: Based on the analysis results output by the processing module, abnormal events including urban pipe network failures, traffic congestion, and public safety incidents are monitored and predicted in real time, and early warning signals are issued, including: According to the type of abnormal event, including urban pipe network failure, traffic congestion, and public safety incidents, dynamically select relevant features, including timestamp, geographic location, event type, and sensor readings, to generate a feature data set of the abnormal event type; According to the feature data set of the abnormal event type, an isolation forest model is constructed, and the features in the historical data set and the corresponding abnormal event labels are used to train the isolation forest model to obtain a trained isolation forest model; Input the feature data in the real-time data stream into the trained isolation forest model and calculate the abnormality of each data point for real-time monitoring and prediction; The abnormality of each data point is compared with the preset threshold. When the abnormality of each data point is ≥ the preset threshold, the abnormal event detection mechanism is triggered and the abnormal event detection result is generated; Based on the abnormal event detection results, corresponding warning signals are generated, including the event type, location, expected impact range, and recommended response measures.

5. The AI-based smart city monitoring system according to claim 4 is characterized in that: The degree of abnormality of each data point, including: For each data point, on each tree in the isolation forest, calculate the path length of the corresponding data point, where the path length refers to the number of edges from the root node of the tree to the leaf node containing the data point; According to the path length of data points on all trees and the total number of trees, the average path length of data points is obtained; A normalization constant is defined, and an abnormality value of the data point is obtained according to the average path length of the data point and the normalization constant, where the abnormality value is between 0 and 1.

6. The AI-based smart city monitoring system according to claim 5 is characterized in that: According to the output of the abnormal monitoring module and the analysis results of the processing module, combined with the dynamic weight allocation mechanism of reinforcement learning, the strategy and configuration of urban operation management are dynamically adjusted to achieve intelligent management of the urban operation status, including: Integrate and process the output of the abnormality monitoring module and the analysis results of the processing module to obtain the processed abnormal event analysis results; According to the analysis results of the abnormal events after processing, the strategy weights are dynamically adjusted using the reinforcement learning algorithm to obtain the adjusted strategy weights; According to the adjusted strategy weights, the configuration of urban operation management is dynamically adjusted, including adjusting the control logic of traffic lights and increasing investment in environmental protection facilities, so as to realize intelligent management of the urban operation status.

7. The AI-based smart city monitoring system according to claim 6, characterized in that: According to the analysis results of the processed abnormal events, the reinforcement learning algorithm is used to dynamically adjust the strategy weights to obtain the adjusted strategy weights, including: Initialize the parameters of the reinforcement learning algorithm, including the Q-value function of the state-action pair, the learning rate, and the discount factor; Define the state space, action space and reward function, and select an action, i.e., an adjustment plan for the strategy weight, based on the current state and the strategy of the reinforcement learning model; Apply the determined strategy weight adjustment scheme to the city operation management strategy. After executing the action, record the new status and reward after executing the action. Based on the new state and reward, update the Q-value function using the update rule of the reinforcement learning algorithm; Repeat the process of selecting a strategy weight adjustment plan, applying the plan, and updating the Q-value function. When the preset number of iterations is reached, the adjusted strategy weight is output.

8. A smart city monitoring method based on AI, the method implementing the system as described in any one of claims 1 to 7, characterized in that: include: Analyze the urban monitoring data stream according to the dual-stream network, extract structured data and perform correlation analysis to obtain correlation analysis results; Process structured data to obtain event summaries and statistical analysis reports; Based on event summaries and statistical analysis reports, real-time monitoring and prediction of urban pipe network failures, traffic congestion, and public safety incidents are carried out, and early warning signals are issued when abnormal events are identified; According to the output and analysis results of the abnormal event monitoring process, the strategies and configurations of urban operation management are dynamically adjusted through the dynamic weight allocation mechanism of reinforcement learning to achieve intelligent management of the city's operation status.

9. A computing device, characterized in that include: one or more processors; A storage device for storing one or more programs, when the one or more programs are executed by the one or more processors, the one or more processors implement the system according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a program, and when the program is executed by a processor, the system according to any one of claims 1 to 7 is implemented.

Images