Method for remote management of traffic equipment based on a multi-level regional hierarchy

Abstract

This invention relates to the field of intelligent traffic management technology and discloses a remote management method for traffic equipment based on a multi-level regional hierarchy. The method divides the traffic network into multiple logically related hierarchical management areas. Then, it collects operational status data and environmental parameters of traffic equipment within each area. By extracting equipment operational feature vectors and combining them with preset management rules for feature fusion, a comprehensive equipment status map is generated. A dynamic decision engine generates a basic management strategy based on this map and environmental parameters. Furthermore, a time series prediction algorithm is used to perform evolutionary analysis on the equipment status and environmental parameters, obtaining a prediction map and sequence. The prediction results are used to calibrate the basic management strategy, forming an optimized management strategy. Finally, the optimized strategy is converted into a specific sequence of equipment control commands, achieving refined and forward-looking remote management of multi-level regional traffic equipment throughout the entire traffic network.

Description

Technical Field

[0001] This invention relates to the field of intelligent traffic management technology, specifically a method for remote management of traffic equipment based on multi-level regional hierarchy. Background Technology

[0002] Modern urban transportation networks are becoming increasingly complex, and their efficient and stable operation relies heavily on ubiquitous transportation equipment, such as traffic lights, electronic police systems, variable message signs, and road surveillance cameras. These devices are numerous, widely distributed, and diverse in function, forming the physical foundation of urban traffic management. Traditional methods of managing transportation equipment typically rely on manual inspections or decentralized, independent systems, which have significant limitations. On the one hand, manual inspections are inefficient and slow to respond, failing to meet the demands for real-time monitoring and rapid response. Especially when dealing with sudden traffic incidents, this lag can exacerbate traffic congestion or even trigger secondary accidents. On the other hand, the lack of effective information exchange and coordination mechanisms between decentralized systems creates "information silos," making it difficult for managers to grasp the overall operational status of the entire transportation network from a macro perspective.

[0003] While existing remote monitoring technologies can enable remote viewing and basic control of equipment status, their functionality is relatively limited. These technologies are mostly confined to independent monitoring of single devices or small localized areas, lacking a perspective for comprehensive management at the regional or even the entire transportation network level. When coordinated control of multiple related devices within a region is required, such as coordinating green waves for traffic lights, existing systems often rely on human experience for strategy formulation and parameter adjustments, resulting in insufficient intelligence. This control method is highly subjective, difficult to adapt to the dynamic changes in traffic flow, and has poor timeliness in strategy adjustments.

[0004] Current management methods largely focus on passively responding to the current state of equipment, lacking the ability to predict future trends. The operation of transportation systems is influenced by various factors, including traffic flow changing over time, weather conditions, and special events. The lack of analysis into the evolving trends of these factors results in a lack of foresight in management strategy formulation, often leading to reactive measures after problems arise—a "remedial" management approach that fails to achieve proactive intervention and preventative maintenance. This management model is ill-suited to the high demands of modern transportation systems for real-time performance, adaptability, and robustness. Therefore, there is an urgent need for a systematic approach that integrates real-time and future information from a multi-level regional perspective to achieve collaborative, intelligent, and remote management of transportation equipment. Summary of the Invention

[0005] The purpose of this invention is to provide a remote management method for traffic equipment based on a multi-level regional hierarchy, so as to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides a remote management method for traffic equipment based on a multi-level regional hierarchy, the method comprising:

[0007] The transportation network is divided into multiple levels of management areas, each of which contains a set of logically related transportation facilities;

[0008] Collect operational status data sequences and environmental parameter sequences of traffic equipment within each management area;

[0009] The system extracts equipment operation feature vectors from the operation status data sequence and combines them with a preset management rule base to perform feature fusion and generate a comprehensive equipment status map.

[0010] Based on the comprehensive equipment status map and environmental parameter sequence, a basic management strategy is generated through a dynamic decision engine.

[0011] Time series prediction algorithms are used to perform evolutionary analysis on the comprehensive equipment status map and environmental parameter sequence to obtain the equipment status prediction map and environmental parameter prediction sequence;

[0012] The basic management strategy is calibrated using equipment status prediction maps and environmental parameter prediction sequences to generate an optimized management strategy.

[0013] The optimized management strategy is converted into a sequence of equipment control commands to enable remote management of multi-level regional traffic equipment.

[0014] Preferably, the division of the traffic network into multiple levels of management areas includes:

[0015] Acquire traffic network topology data, including device geographic location information and the connection relationships between devices;

[0016] The K-means clustering algorithm is used to divide spatial regions based on the geographical location information of the equipment, forming basic management areas;

[0017] Based on the strength of the connection relationships between devices, a community detection algorithm is used to aggregate basic management areas into higher-level integrated management areas;

[0018] Establish a regional hierarchical index table to record the hierarchical relationship of each management region and the device identifiers it contains.

[0019] Preferably, the collection of operational status data sequences and environmental parameter sequences of traffic equipment within each management area includes:

[0020] The system collects equipment operating parameters, including current, voltage, temperature, and vibration frequency, by deploying sensor arrays on transportation equipment.

[0021] Environmental parameters, including ambient temperature, humidity, wind speed, and visibility, are collected using regional environmental monitoring equipment.

[0022] Set a data acquisition time window to sample equipment operating parameters and environmental parameters at fixed time intervals;

[0023] The sampled data is timestamped and cleaned to form a regular sequence of operational status data and environmental parameters.

[0024] Preferably, the step of extracting the device operation feature vector from the operation status data sequence includes:

[0025] Wavelet transform analysis was performed on the operating status data sequence to extract time-frequency domain feature components;

[0026] Calculate the statistical characteristics of the operating parameters of the computing device over time, including mean, variance, and skewness;

[0027] Construct a device operation feature vector, which includes a combination of time-frequency domain feature components and statistical features;

[0028] The equipment operation feature vector is input into a preset management rule base for matching to obtain the equipment status assessment result;

[0029] By combining the equipment condition assessment results with the spatial relationships between equipment, a comprehensive equipment condition map is generated.

[0030] Preferably, the generation of basic management strategies based on the comprehensive equipment status map and environmental parameter sequence through a dynamic decision engine includes:

[0031] Establish a decision-making rule knowledge base to store management measures corresponding to various combinations of equipment status and environmental parameters;

[0032] Graph neural networks are used to learn features from the comprehensive status map of the equipment and extract key decision features.

[0033] Key decision features are fused with environmental parameter sequences in a multimodal manner to generate a decision input feature vector;

[0034] The decision tree classifier processes the input feature vector and outputs a basic management strategy.

[0035] Preferably, the evolutionary analysis of the equipment integrated state map and environmental parameter sequence using a time series prediction algorithm includes:

[0036] Long Short-Term Memory (LSTM) networks are used to model the feature change trends in the overall device state map;

[0037] An autoregressive ensemble moving average model was used to predict environmental parameter sequences.

[0038] The prediction results of equipment characteristic change trends and environmental parameter prediction results are fused to generate equipment status prediction maps and environmental parameter prediction sequences.

[0039] Preferably, the step of using equipment status prediction maps and environmental parameter prediction sequences to calibrate the basic management strategy includes:

[0040] Construct a strategy evaluation model, which includes multiple evaluation indicators and corresponding weight coefficients;

[0041] The basic management strategy was applied to the equipment status prediction map and the environmental parameter prediction sequence to simulate the effect of the strategy execution.

[0042] Calculate the deviation between the strategy execution effect and the expected goal, and generate a strategy evaluation score;

[0043] Based on the strategy evaluation score, the parameter settings of the basic management strategy are adjusted using the gradient descent algorithm;

[0044] Repeat the strategy evaluation and parameter adjustment process until the strategy evaluation score reaches a preset threshold, and then generate an optimized management strategy.

[0045] Preferably, the evaluation metrics included in the strategy evaluation model are:

[0046] Equipment operation stability indicators reflect the fluctuation range of equipment operating parameters after the strategy is implemented;

[0047] Energy efficiency indicators measure the level of energy consumption during strategy execution.

[0048] Equipment lifespan impact indicators, and the degree of impact of assessment strategies on equipment lifespan;

[0049] Environmental adaptability indicators characterize the applicability of strategies under different environmental conditions.

[0050] Preferably, the step of converting the optimization management strategy into a sequence of device control instructions includes:

[0051] Analyze and optimize the control parameters and requirements in the management strategy;

[0052] Based on the specific model and communication protocol of the transportation equipment, generate corresponding equipment control commands;

[0053] Set the execution time and order of instructions to form a sequence of device control instructions;

[0054] Perform syntax checking and logic verification on the sequence of equipment control instructions.

[0055] Preferably, the remote management of multi-level regional traffic equipment includes:

[0056] The device control command sequence is transmitted to the target management area through a dedicated communication network. The area gateway receives the device control command sequence and performs command parsing and routing forwarding.

[0057] The target transportation equipment receives and executes control commands to complete equipment status adjustments.

[0058] The system monitors the execution results of commands in real time and transmits the execution feedback data back to the management center. The device status database is then updated based on the execution feedback data, forming a closed-loop management system.

[0059] Compared with the prior art, the beneficial effects of the present invention are:

[0060] This invention constructs a multi-level regional hierarchical management framework, organically integrating previously isolated traffic equipment according to spatial and logical connections, thus elevating the management perspective from single-point equipment to the regional network level. This hierarchical management structure enables the control system to understand the interrelationships between equipment, allowing for comprehensive consideration of overall regional traffic benefits when formulating strategies, avoiding the problem of local optimization leading to a decline in global performance. For example, when controlling a traffic light at an intersection, its impact on upstream and downstream related intersections can be considered simultaneously, achieving regional collaborative control and effectively improving road network traffic efficiency.

[0061] The method generates a comprehensive equipment status map through feature fusion, transforming scattered, multi-source status information into a unified and easily understood comprehensive representation for decision-making systems. This map-based processing transforms equipment status from isolated data points into an intuitive view reflecting its internal connections and overall health, greatly enhancing the depth and breadth of managers' understanding of equipment operation and providing a more reliable basis for accurately judging equipment status.

[0062] By introducing a dynamic decision engine combined with a pre-defined rule base, the strategy generation process not only relies on real-time data but also incorporates domain knowledge and management experience. This ensures that the generated basic management strategies are both data-driven and objective, while also conforming to actual management logic, thus improving the rationality and executability of the strategies. The decision-making process is no longer a simple "if-then" rule but possesses greater adaptability and a certain degree of intelligence.

[0063] Of particular note is the introduction of a time-series forecasting algorithm to analyze the evolutionary trends of equipment and environmental conditions. This step transforms management from a reactive response to a proactive anticipation approach. The system can predict the degradation trend of equipment performance or potential failure risks, thereby implementing maintenance or adjustment control strategies before problems occur. Simultaneously, it can predict changes in external conditions such as traffic flow and the environment, optimizing management strategies in advance to cope with upcoming traffic peaks or severe weather, significantly enhancing the predictability and robustness of traffic management.

[0064] The strategy calibration process based on prediction results ensures that the final optimized management strategy not only adapts to the current situation but also aligns with the system's future development trends. This strategy possesses time-dimensional adaptability, reducing the need for frequent adjustments due to changing circumstances and improving management stability and efficiency. The long-term effectiveness of the strategy is thus enhanced.

[0065] The optimization strategy is automatically converted into a sequence of equipment control commands, achieving an automated closed loop from decision-making to execution. This significantly reduces manual intervention, lowers labor costs and administrator workload, while improving the accuracy and timeliness of command issuance and execution. The entire method forms a complete technology chain from state perception, intelligent analysis, predictive judgment to precise control, enabling refined, intelligent, and forward-looking remote management of large-scale, distributed transportation equipment, and improving the operational efficiency and service level of the entire transportation system. Attached Figure Description

[0066] Figure 1 This is a schematic diagram illustrating the working principle of the remote management method for traffic equipment based on multi-level regional hierarchy as described in this invention.

[0067] Figure 2 A flowchart for dividing the transportation network into multiple levels of management areas;

[0068] Figure 3 This is a flowchart for extracting device operation feature vectors from a sequence of operating status data.

[0069] Figure 4 A time series prediction analysis chart of equipment status and environmental parameters;

[0070] Figure 5 A diagram illustrating the process of evaluating and optimizing management strategies. Detailed Implementation

[0071] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0072] Please see Figure 1 This invention provides a remote management method for traffic equipment based on a multi-level regional hierarchy, the method comprising:

[0073] Intelligent remote control of traffic equipment is achieved through steps such as hierarchical management, data collection, feature extraction, decision generation, predictive analysis, strategy calibration, and command execution. The specific implementation method is described in detail below with reference to the accompanying drawings, but the scope of protection of this invention is not limited thereto. The overall implementation scheme of the multi-level regional hierarchical remote management method for traffic equipment includes several key steps: dividing the traffic network into multiple levels of management areas, each management area containing a group of logically related traffic equipment. This division is based on the geographical location and connection relationships of the equipment, forming basic management areas and higher-level comprehensive management areas. The operating status data sequence and environmental parameter sequence of traffic equipment in each management area are collected. Data is acquired in real time through sensor groups and environmental monitoring equipment, and timestamp alignment and data cleaning are performed to ensure data regularity. Equipment operating feature vectors are extracted from the operating status data sequence and fused with a preset management rule base to generate a comprehensive equipment status map, which reflects the real-time status and interrelationships of the equipment. Based on the comprehensive equipment status map and environmental parameter sequence, a basic management strategy is generated through a dynamic decision engine. The dynamic decision engine uses graph neural networks and a decision rule knowledge base to output preliminary management measures. A time-series forecasting algorithm is employed to perform evolutionary analysis on the comprehensive equipment status map and environmental parameter sequences, obtaining the equipment status prediction map and environmental parameter prediction sequences. The prediction algorithm includes a Long Short-Term Memory (LSTM) network and an autoregressive ensemble moving average (ASM) model. The equipment status prediction map and environmental parameter prediction sequences are used to calibrate the basic management strategy. Strategy parameters are adjusted using a strategy evaluation model and a gradient descent algorithm to generate an optimized management strategy. This optimized management strategy is then converted into a sequence of equipment control commands and transmitted to the target equipment via a dedicated communication network, enabling remote management. Simultaneously, execution feedback is monitored to form a closed-loop management system.

[0074] Example 1: See Figure 2In practical implementation, when dividing the transportation network into multiple levels of management areas, it is necessary to obtain transportation network topology data. This data includes equipment geographic location information and the connections between equipment. The equipment geographic location information comes from a geographic information system database, and the connections between equipment are defined based on physical links or communication protocols. A K-means clustering algorithm is used to divide spatial regions based on equipment geographic location information, forming basic management areas. The K-means clustering algorithm uses the latitude and longitude coordinates of the equipment as input features and determines the cluster centers by iteratively calculating Euclidean distance, grouping equipment within a distance threshold into the same cluster. The size of the basic management area can be adjusted by configurable parameters to accommodate different network densities. Based on the strength of connections between equipment, a community detection algorithm is used to aggregate the basic management areas into higher-level integrated management areas. The community detection algorithm analyzes the frequency of interaction or data flow weights between equipment, uses the Louvain method to optimize modularity, identifies tightly connected sub-networks, and thus merges the basic areas into integrated areas. A regional hierarchical index table is established to record the hierarchical relationship of each management region and the device identifiers contained therein. The regional hierarchical index table is stored in a tree data structure, with the root node representing the top-level management region and the leaf nodes corresponding to the basic management regions. Each entry contains a region ID, a parent region ID, and a list of devices, supporting fast querying and traversal.

[0075] In some embodiments, when collecting operational status data sequences and environmental parameter sequences of traffic equipment within each management area, the operating parameters of the equipment are collected through sensor groups deployed on the traffic equipment. These sensor groups include current sensors, voltage sensors, temperature sensors, and vibration sensors, which monitor the equipment status in real time at a fixed sampling rate. Environmental parameters are collected using regional environmental monitoring equipment deployed at key points within the management area, such as traffic intersections or the center of equipment clusters, to collect ambient temperature, humidity, wind speed, and visibility values. A data acquisition time window is set to sample the equipment operating parameters and environmental parameters at fixed time intervals. The length of the time window can be dynamically adjusted according to management needs; for example, a sampling interval of 5 minutes can be set to ensure data continuity and real-time performance. The sampled data undergoes timestamp alignment and data cleaning to form a regular operational status data sequence and environmental parameter sequence. Timestamp alignment uses linear interpolation to handle missing data points, and data cleaning removes outliers and noise. The sequences are stored in a time-sorted array format.

[0076] In practical implementation, the K-means clustering algorithm's execution process includes initializing cluster centers, assigning devices to the nearest centers, recalculating center positions, and iterating until convergence. The number of clusters is determined based on the elbow rule or business requirements to ensure the geographical continuity of the basic management area. When processing device connectivity, the community discovery algorithm constructs a weighted graph structure, where nodes represent devices and edge weights represent connection strength, such as communication frequency or the reciprocal of physical distance. Community boundaries are identified by maximizing the modularity function. The maintenance of the regional hierarchical index table involves a periodic update mechanism. When devices are added or removed, the regional division is recalculated and the index table is refreshed to maintain data consistency. Optionally, the data collection time window configuration supports multiple modes, such as fixed interval mode or event-triggered mode. The event-triggered mode initiates sampling when a sudden change in device status is detected, improving data collection efficiency. During timestamp alignment, GPS time is used as the benchmark to ensure multi-source data synchronization. Data cleaning algorithms include Z-score anomaly detection and sliding window filtering to improve data quality.

[0077] The purpose of dividing management areas is understandable: to optimize resource allocation and fault isolation. Basic management areas focus on local equipment management, while comprehensive management areas coordinate cross-regional operations. Collected data sequences provide raw input for subsequent analysis; operational status data sequences reflect equipment health; and environmental parameter sequences affect equipment performance evaluation. In practice, acquiring equipment geographic location information involves multi-source data fusion, such as importing coordinate data from a geographic information system and correcting errors using a real-time positioning system. Connection relationships between devices are exported from a network configuration database, including wired and wireless links. The K-means clustering algorithm is implemented using a distributed computing framework, such as Apache Spark, to process large-scale equipment data. The clustering results are visualized to verify the rationality of the area division. The modularity calculation of the community detection algorithm is accelerated through parallel optimization to ensure stable community structure. The comprehensive management area can have multiple levels, such as regions, sub-regions, and global regions.

[0078] In some embodiments, the data acquisition protocol of the sensor group follows industry standards, such as IEC 61850, and sensor data is transmitted to a regional gateway cache via a local area network or cellular network. Calibration of the environmental monitoring equipment is performed periodically, using standard instruments for comparison to ensure parameter accuracy. Sampling points within the data acquisition time window are stored in a circular buffer to prevent data overflow. A timestamp alignment algorithm handles time offsets caused by network latency, and data cleaning rules are defined based on domain knowledge, such as temperature range checks. Optionally, the query interface for the regional hierarchical index table provides a RESTful API to support external system integration, and index table update transactions guarantee atomicity to avoid concurrent conflicts. During data acquisition, encrypted transmission protects data security, and sampled data compression reduces storage overhead. It can be understood that the division of management areas and data acquisition are the foundation of remote management; the hierarchical structure supports scalable management, and the normalization of data sequences facilitates subsequent feature extraction and decision generation.

[0079] In practical implementation, quantifying the strength of connectivity between devices is a crucial step. Connectivity strength is a comprehensive indicator, and its calculation relies on data from multiple dimensions. The calculation process requires collecting raw interaction data between devices, including but not limited to the number of communication packets, communication duration, physical port connection status, and historical records of collaborative work relying on the same control command within a specific time period. Each dimension is assigned an initial value; for example, the number of communication packets is recorded as an integer, and the communication duration as seconds. These dimensions with different units and numerical ranges are normalized, mapping them to dimensionless values ​​between 0 and 1. The normalization process uses a min-max scaling method to ensure that all dimensions are within a comparable range. After normalization, each dimension is assigned a specific weight coefficient based on its contribution to assessing connectivity importance. These weight coefficients can be set through domain expert experience or derived from analysis of historical operational data. The strength of the connection between device A and device B is calculated using a weighted summation formula. This formula multiplies the normalized values ​​of each dimension by their corresponding weight coefficients and then sums the results. The result is a continuous value between 0 and 1, with higher values ​​indicating a stronger connection and greater dependency between the two devices. This quantified connection strength value will be directly used as the edge weights in the weighted graph constructed in the community detection algorithm.

[0080] In practice, timestamp alignment involves a multi-step data preprocessing process. The purpose of timestamp alignment is to resolve inconsistencies in the time dimension between device operating parameters and environmental parameters from different data sources. Achieving timestamp alignment requires establishing a unified time reference system, typically using a network time protocol to obtain high-precision atomic clock time from a time synchronization server as the benchmark. All data acquisition devices, including sensor arrays on transportation equipment and regional environmental monitoring equipment, need to periodically synchronize with this benchmark time to reduce the inherent drift error of each device's local clock. Once the timestamped sampled data is aggregated at the data processing center, the alignment algorithm begins working. The algorithm sets a standard time axis grid, with intervals consistent with the fixed time intervals of the data acquisition window. For each data source, the algorithm checks the timestamps of its data points. If a data point does not have a corresponding sampled value at the standard time grid point, linear interpolation is used to calculate and fill the gap. Linear interpolation uses the values ​​of the two nearest actual sampled points before and after the data source, calculating an estimated value at the standard time point according to the time ratio. For multiple sampled points existing within a very small time tolerance range near the standard time point, an arithmetic mean is used to merge them into a representative value. This process is performed independently on all data channels, generating a set of regular operational status data sequences and environmental parameter sequences with complete data records at strictly identical time points, laying an accurate temporal foundation for subsequent feature extraction and fusion analysis.

[0081] Example 2: See Figure 3 In practical implementation, extracting equipment operation feature vectors from the operational status data sequence requires wavelet transform analysis. Wavelet transform analysis uses a multi-resolution analysis method to decompose the time-series signal, selecting appropriate wavelet basis functions such as the Daubechies wavelet to perform multi-scale transformations on the operational parameter sequences such as current and voltage values. Through wavelet decomposition, time-frequency domain feature components on different frequency sub-bands are obtained. These time-frequency domain feature components can capture the transient characteristics and periodic patterns of equipment operation. Calculating the statistical characteristics of equipment operation parameters in the time dimension includes calculating the arithmetic mean, variance, and skewness within a sliding window. The arithmetic mean reflects the average level of the parameter, the variance characterizes the dispersion of the data, and the skewness describes the asymmetry of the distribution shape. The statistical characteristics are continuously calculated based on a preset time window to form a feature sequence. Constructing the equipment operation feature vector involves combining and splicing the extracted time-frequency domain feature components and statistical characteristics to form a high-dimensional numerical vector. The features in each dimension of the vector are normalized to eliminate dimensional differences. The equipment operation feature vector serves as a digital representation of equipment status assessment.

[0082] When the equipment operation feature vector is input into a preset management rule base for matching, the management rule base uses a production rule system to store equipment status assessment knowledge. The rule form is a logical statement of "IF condition THEN conclusion". The condition part judges the numerical range of the equipment operation feature vector, and the conclusion part outputs the equipment status level or abnormality type. The matching process uses a forward chain reasoning mechanism to traverse all rules in the rule base. When the equipment operation feature vector meets the condition part of a rule, the rule is triggered, and the equipment status assessment result is obtained. The equipment status assessment result includes discrete status indicators such as normal, warning, and abnormal. Combining the equipment status assessment result and the spatial relationships between equipment, a comprehensive equipment status map is generated. The comprehensive equipment status map is represented by a graph data structure. In the graph, nodes represent traffic equipment and are accompanied by status assessment result attributes. Edges represent spatial adjacency relationships or functional association relationships between equipment. The topological structure of the graph shows the overall operating status of the equipment group.

[0083] In some embodiments, generating basic management strategies based on a dynamic decision engine using a comprehensive device state map and environmental parameter sequences requires establishing a decision rule knowledge base. This knowledge base uses a frame representation to store management measures corresponding to various combinations of device states and environmental parameters. Each knowledge frame contains slots for storing information such as state conditions, environmental thresholds, and recommended strategies. When using a graph neural network to learn features from the comprehensive device state map, the network employs a message-passing mechanism to propagate state information between nodes. It aggregates features from neighboring nodes through multi-layer graph convolution operations, extracting key decision features that reflect the overall state of the device cluster. These key decision features are low-dimensional dense vectors. Multimodal fusion of the key decision features and environmental parameter sequences is performed using feature concatenation. The key decision features output by the graph neural network are connected to the statistical features of the environmental parameter sequences to form a comprehensive decision input feature vector. This vector contains integrated information about device states and environmental factors.

[0084] When processing the decision input feature vector using a decision tree classifier, the classifier is constructed using the CART algorithm. The decision tree model is trained and generated on a historical decision dataset. Each internal node corresponds to a feature judgment condition, and the leaf nodes correspond to specific basic management strategy types. The decision tree classifier recursively judges the decision input feature vector, and based on the feature values, it follows the decision tree path to reach the leaf nodes, outputting the corresponding basic management strategy. The basic management strategy includes specific operational instructions such as equipment parameter adjustments and maintenance scheduling suggestions.

[0085] Optionally, the wavelet basis function selection in wavelet transform analysis can be adaptively adjusted according to the equipment type. Morlet wavelets are used for vibration signal analysis, while Haar wavelets are used for power parameter analysis. The rules in the management rule base support a dynamic update mechanism; when new equipment failure modes appear, new evaluation rules can be added through the rule editor. The training process of the graph neural network adopts a semi-supervised learning approach, utilizing partially labeled equipment state data to optimize model parameters. It can be understood that the extraction of equipment operating feature vectors transforms the original time-series data into a feature representation more suitable for state assessment, and the matching process of the management rule base realizes knowledge-based equipment state diagnosis. The comprehensive equipment state map visually displays the state distribution of the equipment group through a graph structure, and the dynamic decision engine generates management strategies by integrating equipment state and environmental factors.

[0086] In practical implementation, the extraction of time-frequency domain feature components includes the calculation of detail coefficients and approximation coefficients. Detail coefficients reflect the high-frequency components of the signal, while approximation coefficients reflect the low-frequency trends. After feature selection, the most discriminative components are retained. Statistical feature calculations employ a rolling window approach, with the window size set according to the equipment's operating characteristics and the window sliding step consistent with the data acquisition interval. The dimensionality of the equipment operating feature vector is reduced through principal component analysis, retaining the main feature components while reducing computational complexity. The rule conditions in the management rule base support fuzzy logic judgments, allowing rules to be triggered even when feature values ​​fluctuate within a specific range. In addition to discrete state identifiers, the equipment status assessment results also include confidence scores indicating the reliability of the assessment results. The edge weights of the comprehensive equipment status map can be adjusted based on the actual connection strength between devices, assigning higher weights to important connections.

[0087] In some embodiments, the knowledge framework of the decision rule knowledge base supports an inheritance mechanism, allowing state processing strategies for general device types to be inherited and rewritten by specific device types. The number of layers in the graph neural network is determined based on the scale of the comprehensive device state map; large maps use deep network structures, while small maps use shallow network structures. During multimodal fusion, environmental parameter sequences undergo feature engineering, extracting statistical quantities such as mean, maximum, and minimum values ​​before concatenating them with key decision features. The decision tree classifier employs a pruning strategy to prevent overfitting during training, and selects the optimal tree depth parameter through cross-validation. The output format of the basic management strategy uses a standardized template, containing structured fields such as strategy type, target device, and execution parameters. Optionally, the normalization of device running feature vectors can employ Z-score normalization or min-max scaling, selecting the appropriate method based on the feature distribution characteristics. The matching efficiency of the management rule base is optimized by establishing a rule index for rapid rule location. The message passing function of the graph neural network can employ an attention mechanism to dynamically adjust the importance weights of neighbor node information.

[0088] It is understandable that the quality of equipment operation feature vectors directly affects the accuracy of condition assessment, and the combination of wavelet transform and statistical features provides a multi-faceted data analysis perspective. The graph structure characteristics of the comprehensive equipment condition map can effectively capture the mutual influence relationships between equipment, and the multimodal fusion of the dynamic decision engine ensures that management strategies take into account both equipment condition and environmental factors.

[0089] In practical implementation, the process of extracting time-frequency domain feature components using wavelet transform analysis involves a systematic signal decomposition and feature selection process. The core of wavelet transform analysis lies in multi-scale decomposition of the operating state data sequence using selected wavelet basis functions. This process requires selecting wavelet basis functions suitable for the characteristics of the equipment's operating signals. For example, Morlet wavelets are typically chosen for vibration signals containing transient impact characteristics, while Haar wavelets can be selected for electrical signals with step changes. After selecting the wavelet basis functions, multi-level wavelet decomposition is performed on the input operating state data sequence. Each level of decomposition decomposes the signal into low-frequency approximation coefficients and high-frequency detail coefficients. The number of decomposition levels is determined based on the main frequency components of the signal and the analysis requirements, typically ranging from 4 to 8 levels. Time-frequency domain feature components are extracted from the decomposed wavelet coefficients. These feature components include, but are not limited to, statistical quantities such as the energy values ​​of each level of detail coefficients, the maximum values ​​of each level of detail coefficients, and the variances of each level of approximation coefficients. These feature components can simultaneously reflect the local characteristics of the signal in both the time and frequency domains. After being standardized, the extracted time-frequency domain feature components are combined in a predetermined order to form a time-frequency domain feature vector, which serves as an important component of the equipment operation feature vector.

[0090] In practical implementation, the message passing mechanism used by graph neural networks (GNNs) for feature learning in the comprehensive state graph of equipment comprises three core computational steps. This mechanism forms the basis for information propagation in GNNs when processing graph-structured data. The first step is defining the message content each node sends to its neighbors. In the context of the comprehensive state graph, each node represents a transportation device, and its initial feature vector consists of the device's state assessment results and attributes. The message function processes the node's feature vector through a linear transformation or a non-linear activation function to generate the message to be sent. The second step is message aggregation. Each node receives messages from all its directly connected neighbors and merges these messages into a single aggregated message using a defined aggregation function. Common aggregation functions include summation, averaging, and maximization. The choice of aggregation function influences how neighbor information is integrated. The third step is node updating. Each node combines the aggregated message with its current feature vector and generates a new feature vector through a learnable update function. This update function is typically implemented by a neural network, such as a multilayer perceptron. The message passing mechanism is iterated in the multi-layer graph neural network. Each layer corresponds to one message passing process. Through multi-layer iteration, each node can capture information from a wider neighborhood in the graph and extract node features containing rich contextual relationships as key decision features.

[0091] Example 3: In specific implementation, when using time series prediction algorithms to perform evolutionary analysis on the comprehensive equipment state map and environmental parameter sequences, a Long Short-Term Memory (LSTM) network is used to model the feature change trends in the comprehensive equipment state map. LSTM is a special type of recurrent neural network structure that can effectively handle long-term dependencies in sequence data. The input to the LSTM is the time series data of each node's features in the comprehensive equipment state map. These feature sequences are arranged in chronological order, and the feature vector at each time step contains the equipment state assessment result and associated attributes. The LSTM controls the flow of information through gating mechanisms including input gates, forget gates, and output gates. The input gate determines which parts of the current input information need to be updated to the cell state; the forget gate controls which information from the previous cell state needs to be retained or discarded; and the output gate determines which information from the current cell state needs to be output to the hidden state. The training process of the LSTM uses historical comprehensive equipment state map data as training samples. Input-output pairs are constructed through a time series sliding window, and the backpropagation algorithm is used to optimize the network parameters, enabling the LSTM network to accurately predict the changing trends of equipment features at future time points. The output of the Long Short-Term Memory Network is a sequence of predicted values ​​for the features of each node in the overall state map of the device. These predicted values ​​reflect the future evolution of the device state.

[0092] In some embodiments, an autoregressive ensemble moving average (AEM) model is used to predict environmental parameter sequences. The AEM model is a classic time series forecasting method suitable for modeling and predicting stationary time series. The modeling process of the AEM model includes three stages: model identification, parameter estimation, and model validation. In the model identification stage, the autoregressive order and moving average order are determined by analyzing the autocorrelation function and partial autocorrelation function of the environmental parameter sequences. In the parameter estimation stage, the specific values ​​of the autoregressive coefficients and moving average coefficients are calculated using maximum likelihood estimation or least squares methods. In the model validation stage, the applicability of the model is verified through residual analysis to ensure that the residual sequence is white noise. The prediction process of the AEM model is based on historical environmental parameter sequence data. Through iterative calculation, predicted values ​​of environmental parameters for multiple future time points are generated. These predicted values ​​include numerical sequences of parameters such as environmental temperature, humidity, and wind speed. The AEM model can effectively capture the linear dependencies in environmental parameter sequences and provide accurate predictions of environmental change trends.

[0093] When fusing the predicted results of equipment characteristic change trends and environmental parameter predictions, a feature-level fusion strategy is employed to integrate the two types of predictions into a unified representation. The predicted results of equipment characteristic change trends are derived from the output of a long short-term memory network and are multi-dimensional time series data. The predicted results of environmental parameters are derived from the output of an autoregressive ensemble moving average model and are single-dimensional or multi-dimensional time series data. The fusion process normalizes the two types of predictions, converting predicted values ​​of different dimensions into dimensionless ratios, and performs time alignment to ensure consistency at the prediction time points. The generation of the equipment state prediction map is based on the fused prediction results. By reconstructing the graph structure of the comprehensive equipment state map, the predicted feature values ​​are remapped onto the map nodes to form the equipment state prediction map for future time points. The environmental parameter prediction sequence is directly composed of the fused environmental parameter prediction values, arranged in chronological order to form a complete prediction sequence.

[0094] In practical implementation, the modeling of the changing trends of the comprehensive state map features of devices using the Long Short-Term Memory (LSTM) network involves detailed configuration of the network structure. The number of neurons in the hidden layer of the LSM network is determined based on the dimension of the input features, typically set to 2-4 times the feature dimension. Training the LSM network uses the stochastic gradient descent optimization algorithm, with a learning rate set between 0.001 and 0.01, and the number of training iterations is selected from 100 to 500 depending on the data size. Input data preprocessing for the LSM network includes normalization, scaling the feature values ​​to the range of 0 to 1 to avoid the impact of numerical range differences on the training process. The output layer activation function of the LSM network is a linear activation function, directly outputting the predicted feature values. The loss function uses the mean squared error function to measure the deviation between the predicted and true values.

[0095] In some embodiments, the prediction of environmental parameter sequences by the autoregressive ensemble moving average model requires sequence stationarization processing. For non-stationary environmental parameter sequences, a differencing method is used to convert them into stationary sequences. The order of the autoregressive ensemble moving average model is optimized using information criteria such as the Akaike information criterion or the Bayesian information criterion, selecting the combination of model orders that minimizes the information criterion values. Parameter estimation of the autoregressive ensemble moving average model uses conditional least squares, obtaining coefficient estimates through iterative calculation. The model uses a rolling prediction method, gradually updating the historical data window. The prediction results of the autoregressive ensemble moving average model are inversely differencing to restore the predicted values ​​of the original sequence, ensuring the practical significance of the predicted values. Optionally, the fusion process of the equipment characteristic change trend prediction results and the environmental parameter prediction results can adopt an adaptive weighting method, dynamically adjusting the weight coefficients according to the confidence level of the prediction results. The fusion weight coefficients are determined by the reciprocal of the prediction error; models with smaller prediction errors are assigned higher weights to improve the accuracy of the fusion results. The fused prediction results undergo post-processing operations, including smoothing and outlier filtering, to improve the stability of the prediction results.

[0096] It is understandable that Long Short-Term Memory (LSTM) networks excel at capturing nonlinear patterns and long-term dependencies in equipment state sequences, while autoregressive ensemble moving average (ATM) models are suitable for predicting linear trends in environmental parameter sequences. Combining these two predictive models fully leverages their respective advantages, providing more comprehensive evolutionary analysis results. Equipment state prediction maps and environmental parameter prediction sequences provide future state information for subsequent policy calibration, supporting forward-looking management decisions. In specific implementation, the modeling of the changing trends of the comprehensive equipment state map features includes map serialization, converting graph-structured data into a serialized representation for LTM network processing. Map serialization generates node sequences using graph traversal algorithms such as breadth-first search or depth-first search, preserving the spatial relationships between nodes. The batch size during LTM network training is set between 32 and 128, and early stopping is used to prevent overfitting; training terminates when the validation set loss function does not decrease for several consecutive cycles. The prediction results of the LTM network are denormalized, converting them back to the original feature value range for easier subsequent analysis and use.

[0097] In some embodiments, the application of the autoregressive ensemble moving average model needs to consider the seasonal characteristics of the environmental parameter series. For parameter series with obvious seasonality, a seasonal autoregressive ensemble moving average model is adopted. The seasonal autoregressive ensemble moving average model adds a seasonal autoregressive term and a seasonal moving average term to the basic model. The model order includes both non-seasonal and seasonal orders. The identification of the seasonal autoregressive ensemble moving average model eliminates seasonal effects through seasonal differencing. Parameter estimation uses the maximum likelihood estimation method, and the prediction results include seasonal variation patterns.

[0098] The fusion process of the predicted results of equipment characteristic change trends and environmental parameter predictions can be expressed by the following formula:

[0099]

[0100] in: This represents the dimensionless fusion prediction index at time t. This represents the weight coefficient of the i-th device feature. Let represent the predicted value of the i-th device feature at time t. This represents the reference value for the i-th device feature. Indicates the number of device features. This represents the weight coefficient of the j-th environmental parameter. This represents the predicted value of the j-th environmental parameter at time t. This represents the baseline value of the j-th environmental parameter. This represents the number of environmental parameters. The baseline value is typically the historical average or rated value of the corresponding parameter, making the ratio dimensionless. Weighting coefficients are calculated using historical prediction errors, assigning higher weights to predictions with smaller errors. The fusion results are used to generate equipment condition prediction maps and environmental parameter prediction sequences, supporting subsequent management strategy optimization.

[0101] It is understandable that the choice of time series forecasting algorithm is based on data characteristics. The complex nonlinear characteristics of equipment state sequences are suitable for long short-term memory networks, while the linear characteristics of environmental parameter sequences are suitable for autoregressive ensemble moving average models. The fusion of forecast results ensures the synergistic consideration of equipment state and environmental factors, improving the comprehensiveness and accuracy of evolutionary analysis. The equipment state prediction map continues the graph structure characteristics of the comprehensive equipment state map, providing a visual representation of the future state distribution of the equipment group.

[0102] In practical implementation, the process of constructing input-output pairs using a time-series sliding window is a core step in preparing training data for Long Short-Term Memory (LSTM) networks. The time-series sliding window method is used to generate supervised learning samples from continuous historical device state profile feature data. The construction process requires determining three key parameters of the sliding window: window size, sliding step, and prediction step. The window size defines the number of historical data points used for prediction, the sliding step controls the distance the window moves each time, and the prediction step specifies the future time points to be predicted. The window size is typically determined based on the periodicity of device state changes and device response time. For example, a larger window size can be set for slowly changing device states to capture long-term trends, while a smaller window size is needed for rapidly changing states to focus on short-term dynamics. The sliding step is generally set to 1 to ensure the continuity of the sequence and that each time point is fully utilized. The prediction step is set according to the forward-looking needs of management decisions, such as predicting the state at several future time points. Specifically, starting from the beginning of a complete historical feature time series, feature data from consecutive time points of the window size are extracted as the input vector, and feature data from the next closest prediction step time point is used as the target output vector, forming a complete input-output pair. The window moves backward by a sliding step size unit, and the above truncation operation is repeated to generate the next input-output pair, until the entire history sequence is traversed, generating a large sample dataset for training the Long Short-Term Memory network.

[0103] In practice, verifying the applicability of an autoregressive ensemble moving average model through residual analysis is a systematic statistical testing process. Residual analysis aims to test whether the residual sequence remaining after model fitting is a white noise sequence. The first step in residual analysis is to calculate the residual sequence. The residual is the difference between the actual observed values ​​of the environmental parameter sequence and the fitted values ​​of the autoregressive ensemble moving average model, with one residual value corresponding to each time point. After calculating the residuals, it is necessary to test the autocorrelation of the residual sequence. The Ljung-Box Q test is typically used to verify whether there is a significant autocorrelation structure in the residual sequence, with the null hypothesis being that the residuals are a white noise sequence. The test process requires calculating the Q statistic, which follows a chi-square distribution. By comparing the value of the Q statistic with the critical value of the chi-square distribution for the corresponding degrees of freedom, it is determined whether to reject the null hypothesis. If the p-value of the Q statistic is greater than the significance level, the null hypothesis cannot be rejected, indicating that the residual sequence has no significant autocorrelation. The second step in residual analysis is to test the normality of the residual sequence. A normal probability plot or the Shapiro-Wilk test is used to assess whether the residuals approximately follow a normal distribution. Although the autoregressive ensemble moving average model does not strictly require the residuals to follow a normal distribution, normality helps improve the efficiency of parameter estimation and the accuracy of prediction interval calculation. Only when the residual sequence passes the white noise test and shows no obvious regularity can it be considered that the autoregressive ensemble moving average model has fully extracted the information from the original sequence, and the model's applicability is validated.

[0104] See Figure 4 This paper presents the evolutionary analysis results of a comprehensive equipment state map and environmental parameter sequences based on a time series prediction algorithm. The figures show the trends of key state parameters of multiple devices over time, as well as the synchronous changes in environmental temperature parameters. Modeling the changing trends of equipment characteristics using a Long Short-Term Memory (LSTM) network effectively captures the nonlinear patterns and long-term dependencies of equipment states. Simultaneously, combining an autoregressive ensemble moving average model with the prediction of environmental parameter sequences enables synergistic analysis of equipment states and environmental factors. The curves in the figures clearly show the periodic patterns and trend changes of different equipment operating states, providing forward-looking predictive information for equipment management. This fusion prediction method fully utilizes the advantages of each algorithm, providing more comprehensive evolutionary analysis results and supporting intelligent traffic equipment management decisions. The accuracy and stability of the prediction results provide a reliable foundation of future state information for subsequent strategy calibration.

[0105] Example 4: In practical implementation, constructing a strategy evaluation model requires defining multiple evaluation indicators and assigning corresponding weight coefficients to each indicator. The strategy evaluation model is a multi-objective decision-making framework used to quantitatively evaluate the expected effects of basic management strategies. The strategy evaluation model includes four core evaluation dimensions: equipment operation stability indicators, energy efficiency indicators, equipment lifespan impact indicators, and environmental adaptability indicators. Each indicator requires the design of specific calculation methods and quantitative standards. The equipment operation stability indicator reflects the fluctuation range of equipment operating parameters after strategy implementation. Stability is quantified by calculating the variance or coefficient of variation of key operating parameters under the influence of the strategy; a smaller fluctuation range indicates higher stability. The energy efficiency indicator measures the energy consumption level during strategy implementation. Efficiency is assessed by statistically analyzing the ratio of total energy consumption to output or service volume within the strategy implementation period; a lower ratio indicates higher energy efficiency. The equipment lifespan impact indicator assesses the degree of influence of the strategy on equipment lifespan. It predicts changes in remaining lifespan after strategy implementation based on equipment degradation models; smaller lifespan decay indicates a more positive impact. The environmental adaptability indicator characterizes the applicability of the strategy under different environmental conditions. Adaptability is evaluated by testing the strategy's performance in various typical environmental scenarios; broader adaptability results in higher indicator scores.

[0106] Referring to Table 1, the weight coefficients in the strategy evaluation model are determined using the analytic hierarchy process (AHP). A judgment matrix is ​​constructed to compare the relative importance of each indicator, and the weight coefficients are obtained by calculating the largest eigenvalue and eigenvector. These weight coefficients need to be adjusted based on specific management objectives and equipment types. For example, energy efficiency indicators are given higher weights in energy-sensitive scenarios, while equipment operational stability indicators are given higher weights in scenarios with high reliability requirements. The output of the strategy evaluation model is a comprehensive score, calculated by weighting the quantified values ​​of each indicator with their corresponding weight coefficients. A higher comprehensive score indicates a better overall strategy performance.

[0107] Table 1: Weighting Coefficients of Indicators in the Strategy Evaluation Model

[0108] Evaluation indicators Indicator Description Weighting coefficient range Typical values Equipment operational stability indicators Operating parameter fluctuation range 0.2-0.4 0.35 Energy efficiency indicators Energy consumption level 0.1-0.3 0.25 Equipment lifespan impact indicators Degree of impact on service life 0.2-0.35 0.25 Environmental adaptability indicators Environmental suitability 0.15-0.3 0.15

[0109] When applying basic management strategies to simulate the execution effects of equipment status prediction maps and environmental parameter prediction sequences, a strategy execution simulation environment needs to be established. This simulation environment is built using digital twin technology, taking equipment status data from the equipment status prediction map and environmental data from the environmental parameter prediction sequence as input, and the basic management strategy as the control logic. The strategy execution simulation process progresses according to time steps. Within each time step, the control actions specified by the strategy are executed based on the current equipment status and environmental parameters, updating the equipment status and recording energy consumption data. During the simulation, the raw data corresponding to various evaluation indicators are tracked and recorded, such as real-time values ​​of equipment operating parameters, cumulative energy consumption values, and equipment wear coefficients, providing a data foundation for subsequent indicator calculations.

[0110] The deviation between the strategy's execution effect and the expected goal requires defining a deviation calculation method for each evaluation indicator. The deviation is quantified by comparing the simulation results with the preset target values. The deviation of the equipment operation stability indicator is calculated as the absolute difference between the actual parameter fluctuation range and the ideal fluctuation range. The deviation of the energy efficiency indicator is calculated as the relative error between the actual energy efficiency ratio and the target energy efficiency ratio. The deviation of the equipment lifespan impact indicator is calculated by the difference between the predicted lifespan decay and the expected lifespan decay. The deviation of the environmental adaptability indicator assesses the consistency of the strategy's performance under different environmental scenarios. When generating the strategy evaluation score, the deviation of each indicator is normalized to a score within the range of 0-1, with smaller deviations resulting in higher scores. A weighted average score is calculated based on the weighting coefficients to obtain the strategy evaluation score.

[0111] The gradient descent algorithm is used to adjust the parameters of the basic management strategy based on the strategy evaluation score. Gradient descent iteratively optimizes to find the parameter combination that maximizes the strategy evaluation score. It calculates the gradient of the strategy evaluation score with respect to the strategy parameters, and adjusts the parameter values ​​in the opposite direction of the gradient to improve the score. The learning rate parameter controls the step size of each adjustment. The strategy parameter adjustment process requires setting convergence conditions; optimization stops when the increase in the strategy evaluation score is less than a threshold or when the maximum number of iterations is reached, and the optimized parameter settings are output.

[0112] The process of repeatedly evaluating strategies and adjusting parameters forms an optimization loop. Each iteration generates a new strategy evaluation score and updates the parameters until the strategy evaluation score reaches a preset threshold. The preset threshold is set according to management requirements and represents the minimum acceptable level of strategy effectiveness. Once the threshold is reached, the optimization process stops generating the optimized management strategy. The optimized management strategy includes adjusted control parameters and execution logic, resulting in better overall performance. In some embodiments, the simulation of strategy execution effectiveness needs to consider the interactions between devices. The connections in the device state prediction graph affect the propagation of strategy execution effectiveness. The simulation environment needs to implement a propagation mechanism for device state changes. When the operating state of one device changes, the states of related devices will also adjust accordingly. This interaction is modeled through edge relationships in the graph; state changes propagate along the edges and decay, more realistically reflecting the strategy execution effectiveness.

[0113] Optionally, the gradient descent algorithm can employ a stochastic gradient descent variant to improve optimization efficiency, randomly sampling gradients from the policy parameter space. A momentum term can accelerate the convergence process and avoid getting trapped in local optima, while adaptive learning rate adjustment can dynamically adjust the step size based on gradient changes. In essence, the policy evaluation model quantifies the multi-dimensional policy effects into a single score, facilitating comparison and optimization of different policies. The gradient descent algorithm improves policy performance through systematic parameter search, and the simulated environment allows for safe testing of policy effects without affecting the actual system. In specific implementation, calculating equipment operational stability indicators requires determining the set of key operating parameters and the monitoring time window. Key operating parameters include core parameters such as current, voltage, and temperature. The monitoring time window covers the entire policy execution cycle, and parameter fluctuation ranges are quantified by calculating the standard deviation or range of parameter values ​​within the window. Calculating energy efficiency indicators requires clearly defining energy consumption metering points and output metering units. Energy consumption data is read from the equipment's electrical energy metering device, and output is defined according to equipment function, such as traffic flow or service duration.

[0114] Assessing equipment lifespan impact indicators requires establishing an equipment degradation model. This model, built upon historical operational data, describes the relationship between operating parameters and equipment health status. Testing environmental adaptability indicators necessitates defining a set of typical environmental scenarios covering various combinations of temperature, humidity, and wind speed conditions to evaluate the strategy's performance stability under each scenario. Setting the time step for the strategy execution simulation environment requires a balance between accuracy and efficiency; an excessively large time step may miss details, while an excessively small one increases computational burden. The time step is typically determined based on equipment response time and parameter change frequency, commonly set to minutes or hours. The state update logic during the simulation process needs to be based on the equipment's physical model or a data-driven model to ensure that state changes conform to actual patterns.

[0115] The parameter initialization of the gradient descent algorithm affects the optimization effect. Initial parameters are usually set to empirical values ​​or multiple randomly generated initial points for parallel optimization. The learning rate setting needs to be careful; too large a rate can lead to oscillations, while too small a rate will result in slow convergence. A learning rate decay strategy can be used to improve the accuracy of later optimization stages. Optionally, a penalty term can be introduced into the policy evaluation score to handle constraint violations, such as reducing the score when policy parameters exceed the safe range. Multi-objective optimization methods can simultaneously optimize multiple evaluation indicators, generating a Pareto optimal policy set for decision-makers to choose from. It can be understood that the policy evaluation and calibration process transforms predictive information into optimization decisions, improving the foresight and adaptability of management strategies. Optimizing management strategies considers both the current state and future evolution, achieving smarter traffic equipment management. In specific implementation, the process of determining the weight coefficients of the policy evaluation model using the analytic hierarchy process (AHP) requires constructing the system's judgment matrix and completing consistency checks. The AHP quantifies the relative importance between indicators by decomposing the decision problem into a hierarchical structure and performing pairwise comparisons. The specific implementation process establishes a hierarchical model, placing the overall strategy evaluation goal at the top layer. The middle layer comprises four evaluation dimensions: equipment operation stability indicators, energy efficiency indicators, equipment lifespan impact indicators, and environmental adaptability indicators. The bottom layer consists of specific quantitative sub-indicators for each dimension. A judgment matrix is ​​constructed, and domain experts are invited to use a 1-9 scale to compare the pairwise importance of indicators within the same level. A scale value of 1 indicates that two indicators are equally important, 3 indicates that one indicator is slightly more important than the other, 5 indicates significant importance, 7 indicates strong importance, 9 indicates extreme importance, 2, 4, 6, and 8 are intermediate values, and the reciprocal indicates the importance of the inverse comparison. After forming a positive and negative judgment matrix through expert scoring, the largest eigenvalue and its corresponding eigenvector of the matrix are calculated. The eigenvector is solved using the sum-product method or the power method, and the components of the eigenvector are normalized to obtain the initial weight coefficients of each indicator. A consistency test is conducted, and the consistency index (CI) and random consistency ratio (CR) are calculated. When the CR value is less than 0.1, the judgment matrix is ​​considered to meet the consistency requirements, and the weight coefficient allocation is reasonable and usable. If the CR value is greater than or equal to 0.1, experts need to readjust the scale values ​​in the judgment matrix until the consistency test is passed, thereby obtaining a scientific and reasonable weight coefficient allocation scheme.

[0116] In practical implementation, predicting changes in remaining service life based on equipment degradation models requires establishing an accurate equipment health status assessment system. The equipment degradation model constructs a mathematical relationship between equipment performance parameters and service life decay by analyzing historical operating data. The specific implementation process involves identifying key performance parameters affecting equipment lifespan. These parameters vary across different types of traffic equipment. For example, traffic light equipment focuses on luminous intensity decay rate and color coordinate drift; barrier gate equipment focuses on motor torque change rate and wear coefficient of mechanical transmission components; and traffic signal controllers focus on component aging parameters and heat dissipation performance indicators. Data is collected throughout the equipment's entire lifecycle, including baseline parameter test records at the time of manufacture, initial operating data after installation and commissioning, parameter change records from regular maintenance and inspection, and actual service life data at the time of scrapping, constructing a historical database covering multiple equipment batches. Regression analysis is used to establish the mapping relationship between performance parameter changes and remaining service life. For equipment with obvious linear degradation characteristics, linear regression is used to fit the slope curve of performance parameters changing over time. For equipment with non-linear degradation characteristics, machine learning algorithms such as support vector regression or Gaussian process regression are used to construct degradation models. During the model validation phase, historical data is split chronologically. The first 70% of the data is used as the training set for model parameter training, and the last 30% is used as the test set to evaluate the model's prediction accuracy. Indicators such as root mean square error (RMSE) and mean absolute percentage error (MASE) are calculated to ensure that the prediction error remains within acceptable limits. In the strategy evaluation phase, the predicted equipment performance parameters after strategy implementation are input into the trained degradation model to calculate the corresponding remaining useful life estimate. This estimate is then compared with the baseline useful life prediction without the strategy implementation to determine the degree of impact on useful life.

[0117] See Figure 5 This paper demonstrates the changing trends of the four core evaluation metrics and the improvement process of the overall strategy evaluation score during the gradient descent algorithm optimization process. The operational stability metric reflects the effectiveness of controlling the fluctuation range of equipment operating parameters after strategy implementation, while the energy efficiency metric measures the degree of optimization in energy consumption levels. The equipment lifespan impact metric assesses the positive impact of the strategy on equipment lifespan, and the environmental adaptability metric characterizes the applicability of the strategy under different environmental conditions. Using weight coefficients determined by the analytic hierarchy process (AHP), the multi-dimensional metrics are integrated into a unified strategy evaluation score, providing a clear objective function for parameter optimization. The synergistic improvement of various metrics during the optimization process reflects the comprehensive effect of strategy calibration, and the final optimized management strategy exhibits better performance across multiple dimensions. This systematic strategy evaluation and optimization method ensures the scientific and effective nature of management decisions, achieving intelligent and refined management of traffic equipment.

[0118] Example 5: In practical implementation, parsing the control parameters and requirements in the optimization management strategy requires establishing a strategy parsing specification. The optimization management strategy is stored in a structured data format, including fields such as strategy type, target device identifier, control parameter list, and execution conditions. The parsing process identifies the strategy type, such as device parameter adjustment, operation mode switching, or maintenance scheduling strategies. Different types correspond to different parsing rules. For device parameter adjustment strategies, it is necessary to extract control parameters such as the target parameter name, target value, and adjustment method, for example, adjusting the brightness of a traffic light from the current 80% to 60%. For operation mode switching strategies, it is necessary to parse the target mode identifier and switching timing requirements, such as switching a traffic signal controller from daily mode to peak mode. After the control parameters are parsed, it is necessary to verify the legality and rationality of the parameters and check whether the parameter values ​​are within the allowable range of the equipment to avoid equipment malfunctions due to parameter errors.

[0119] Generating corresponding equipment control commands based on the specific model and communication protocol of the transportation equipment requires establishing an equipment command library. This library stores the command sets and communication protocols supported by different equipment models. The command generation process involves querying the target equipment's model information and retrieving the equipment's communication protocol type from the equipment registration database, such as MODBUS-RTU, TCP / IP, or a custom binary protocol. Control parameters are converted into command data frames according to the protocol specifications. For MODBUS-RTU, this requires constructing a data frame consisting of the slave address, function code, register address, register data, and CRC checksum. For TCP / IP, command data needs to be encapsulated according to the application layer protocol format, including a header, command body, and checksum fields. Data encoding formats must be considered during command generation, such as endianness for integer parameters and IEEE 754 encoding for floating-point parameters, to ensure the equipment can correctly parse the command content.

[0120] Setting the execution time and order of instructions requires establishing an instruction scheduling mechanism. The instruction execution time is determined based on policy requirements and management needs. For control instructions with high real-time requirements, an immediate execution flag is set, and the instruction is sent immediately after generation. For planned control instructions, a specific execution timestamp is set, such as daily mode switching at 8:00 AM. The instruction execution order is achieved by setting priority flags; higher-priority instructions are executed first, and instructions with dependencies are arranged in chronological order. Setting the instruction execution time needs to consider network transmission latency and device processing time, reserving sufficient time margin for instruction execution. After the instruction sequence is generated, the metadata of each instruction needs to be recorded, including instruction number, target device, execution time, priority, and other information, to facilitate execution process tracking and management.

[0121] Performing syntax checks and logical verifications on device control command sequences requires establishing a set of verification rules. Syntax checks verify whether the command format conforms to communication protocol specifications. Syntax checks include command length verification, field type verification, and checksum calculation verification to ensure the command format is correct. Logical verification checks the rationality and security of the command content, including parameter range verification, device status verification, and command conflict detection. Parameter range verification checks whether command parameters are within the device's allowed value range, such as whether the parameter value for a brightness adjustment command is between 0-100%. Device status verification checks whether the target device is in a state that can receive commands, such as not sending control commands when the device is offline. Command conflict detection analyzes whether there are contradictory commands in the command sequence, such as sending both on and off commands simultaneously.

[0122] In practical implementation, transmitting device control command sequences to the target management area via a dedicated communication network requires configuring network transmission parameters. The dedicated communication network is constructed using wired Ethernet or wireless 4G / 5G networks. During transmission, a TCP connection or UDP datagram channel is established. The command sequence is packaged according to a preset encapsulation format, and sequence numbers and timestamps are added for data integrity verification. Large-scale command transmission employs a batch transmission strategy, dividing the command sequence into multiple data packets and setting up a retransmission mechanism to ensure transmission reliability. Encrypted channels are enabled for network transmission, using TLS / SSL protocols to encrypt command data, preventing commands from being stolen or tampered with.

[0123] After receiving the sequence of control commands from the devices, the area gateway performs command parsing and routing forwarding. Deployed at the boundary of the management area, the area gateway has protocol conversion and routing decision-making functions. The command parsing process decapsulates the transport layer protocol header, extracts the command data payload, and verifies the legitimacy of the command source. For routing forwarding, the routing table is consulted based on the target device identifier to determine the next-hop address, and the command data is recapsulated into a format suitable for transmission within the area's network. The area gateway maintains a device connection status table, records the online status of devices in real time, and temporarily stores commands for offline devices and resends them periodically. The target traffic device receives and executes the control commands to complete the device status adjustment. The device communication interface receives data frames, parses the command content, and verifies the command's legitimacy. Before command execution, a security check is performed, checking the command signature and execution permissions. Once confirmed, the command operation is executed. During execution, changes in device status are monitored, and command execution logs are recorded, including the execution start time, execution result, and exception information. After command execution, an execution result code is generated, and the execution result is returned to the area gateway via a response message.

[0124] Real-time monitoring of command execution results and the transmission of execution feedback data back to the management center require a feedback mechanism. The feedback data includes information such as command execution status, current device parameter values, and exception codes. A timestamp and command number are added to the feedback data to establish a correspondence with the sent command, and it is transmitted to the management center via a dedicated communication network. Upon receiving the feedback data, the management center updates the command execution status table, recording the execution result of each command.

[0125] The device status database is updated based on execution feedback data to form a management closed loop. The device status database stores the latest parameter values ​​and operating status of the devices. Update operations modify the corresponding device record fields based on feedback data; for example, updating device parameter values ​​upon successful instruction execution and marking an abnormal device state upon failure. Database updates trigger status change events, notifying relevant system components for subsequent processing, completing the full management closed loop from policy generation to execution feedback. In some embodiments, the generation of device control instructions can consider differences in device firmware versions, as different firmware versions support different instruction sets. Before instruction generation, the device firmware version number is queried, and the corresponding instruction template is selected to ensure instruction compatibility. For parameters that do not support direct control, indirect control methods are used, such as combining multiple basic instructions to achieve complex control functions. Optionally, the instruction execution time point setting can support conditional triggering modes. In addition to absolute time points, relative time points or event trigger conditions can also be set. The relative time point is based on the completion time of the previous instruction, and the event trigger condition is associated with changes in device status or environmental conditions, improving the flexibility of instruction execution. It can be understood that the generation and verification of device control instruction sequences ensure that control intentions are accurately translated into executable device operations, and the instruction transmission and execution mechanism enables reliable remote control. The execution feedback and status update form a closed management loop, providing data support for subsequent strategy optimization.

[0126] In practical implementation, syntax checking and logical verification can be achieved using automated verification tools integrated into the instruction generation pipeline. The syntax checker automatically verifies the instruction format based on the protocol specification file, while the logical verifier performs a validity check based on the device rule base. When errors are detected during verification, error reports are automatically generated with suggested modifications, improving the accuracy and efficiency of instruction generation. Instruction transmission supports QoS (Quality of Service) level settings, with critical instructions set to high priority to ensure timely transmission. A heartbeat detection mechanism is implemented during transmission, periodically checking network connection status and automatically switching to an alternative transmission path when a disconnection is detected. Digital signatures are added to transmitted data, and the recipient verifies the signature's legitimacy to ensure the instruction's origin is trustworthy.

[0127] The regional gateway implements a command caching mechanism, temporarily storing commands to be sent during network interruptions and automatically resuming transmission after network recovery. The gateway supports command priority scheduling, forwarding high-priority commands first while preventing low-priority commands from being starved. The gateway implements flow control, dynamically adjusting the transmission rate based on network conditions to avoid network congestion. Timeout control is implemented during device command execution, setting an execution timeout period; if no response is received within the timeout period, the command execution is marked as failed. A secondary confirmation is performed before executing important commands, sending a pre-command to inquire about the device status, and only sending the formal execution command after confirming the device is ready. Detailed logs are recorded during execution, including timestamps for command reception time, parsing time, execution start time, and execution end time. Optionally, execution feedback data can include device environmental parameters, such as real-time data like temperature and voltage during command execution, providing more comprehensive contextual information for subsequent analysis. Feedback transmission uses a differential synchronization strategy, transmitting only changed status data to reduce network bandwidth consumption. It can be understood that the accurate generation and reliable execution of device control command sequences are the foundation for remote management, and the execution feedback mechanism forms a closed-loop control. The closed-loop management system enables the system to continuously optimize management strategies based on actual performance, achieving intelligent management that is self-adjusting and self-improving.

[0128] In practical implementation, instruction conflict detection requires the establishment of an instruction relationship model and a conflict rule base. Instruction conflict detection is used to discover contradictory or interfering instruction combinations in the device control instruction sequence. The instruction relationship model represents the dependencies and mutual exclusions between instructions using a directed graph structure. Nodes represent individual control instructions, and edges represent timing constraints or resource competition relationships between instructions. The conflict rule base includes device operation constraints and business logic restrictions, such as the same device not being able to receive both open and close instructions simultaneously, and adjacent devices needing to meet collaborative work requirements in their state changes. The detection process performs static analysis on the instruction sequence, parsing the operation object, action type, and parameter settings of each instruction to construct an instruction operation relationship graph. A graph traversal algorithm is used to detect violations of conflict rules, such as detecting mutually exclusive operations on the same resource or circular dependency paths in the graph. For detected instruction conflicts, the system generates a conflict report, indicating the conflicting instruction pairs and conflict type, and provides resolution suggestions, such as adjusting the instruction execution order or modifying instruction parameters. Instruction conflict detection is completed before the instruction sequence is issued, ensuring the logical consistency and execution safety of the instruction sequence.

[0129] In practical implementation, the differential synchronization strategy requires the design of data change detection and incremental transmission mechanisms. This strategy aims to reduce network bandwidth consumption by sending only changed status information when transmitting execution feedback data. The core of the differential synchronization strategy lies in comparing the current device status with the previously reported status to identify the changed data fields. The implementation involves maintaining historical copies of the device status at both the management center and the device, storing the previously reported values ​​of each parameter. When a device needs to report execution feedback data, the system compares the currently collected real-time device status value with the corresponding value in the historical copy, calculates the difference between each numerical parameter, and compares the changes in enumerated parameters or status identifiers. For changed data fields, the system records the field identifier, the value before the change, the value after the change, and the change timestamp, constructing a differential dataset. This differential dataset is encapsulated according to a predefined compression format, containing a data header, a list of changed fields, and a checksum. The data header indicates that this transmission is a differential data packet rather than a full data packet, and the list of changed fields records the identifier and current value of each changed field in sequence. After the transmission is complete, the management center receives the difference data packet, locates the corresponding field in the device status record based on the field identifier, updates the old value with the new value, and simultaneously updates the local historical copy to maintain synchronization. The difference synchronization strategy also includes an integrity verification mechanism, periodically performing full data synchronization to correct accumulated synchronization errors and ensure the consistency of status data at both ends.

[0130] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A remote management method for traffic equipment based on multi-level regional hierarchy, characterized in that, The method includes the following steps: The transportation network is divided into multiple levels of management areas, each of which contains a set of logically related transportation facilities; Collect operational status data sequences and environmental parameter sequences of traffic equipment within each management area; The system extracts equipment operation feature vectors from the operation status data sequence and performs feature fusion with a preset management rule base to generate a comprehensive equipment status map, including: performing wavelet transform analysis on the operation status data sequence to extract time-frequency domain feature components; Calculate the statistical characteristics of the operating parameters of the computing device over time, including mean, variance, and skewness; Construct a device operation feature vector, which includes a combination of time-frequency domain feature components and statistical features; The equipment operation feature vector is input into a preset management rule base for matching to obtain the equipment status assessment result; By combining the equipment condition assessment results with the spatial relationships between equipment, a comprehensive equipment condition map is generated; Based on the comprehensive equipment status map and environmental parameter sequence, a basic management strategy is generated through a dynamic decision engine, including: Establish a decision-making rule knowledge base to store management measures corresponding to various combinations of equipment status and environmental parameters; Graph neural networks are used to learn features from the comprehensive status map of the equipment and extract key decision features. Key decision features are fused with environmental parameter sequences in a multimodal manner to generate a decision input feature vector; The decision input feature vector is processed by a decision tree classifier to output a basic management strategy. Time series prediction algorithms are used to perform evolutionary analysis on the comprehensive equipment status map and environmental parameter sequence to obtain the equipment status prediction map and environmental parameter prediction sequence; The basic management strategy is calibrated using equipment status prediction maps and environmental parameter prediction sequences to generate an optimized management strategy. The optimized management strategy is converted into a sequence of equipment control commands to enable remote management of multi-level regional traffic equipment.

2. The remote management method for traffic equipment based on a multi-level regional hierarchy as described in claim 1, characterized in that, The division of the traffic network into multiple levels of management areas includes: Acquire traffic network topology data, including device geographic location information and the connection relationships between devices; The K-means clustering algorithm is used to divide spatial regions based on the geographical location information of the equipment, forming basic management areas; Based on the strength of the connection relationships between devices, a community detection algorithm is used to aggregate basic management areas into higher-level integrated management areas; Establish a regional hierarchical index table to record the hierarchical relationship of each management region and the device identifiers it contains.

3. The remote management method for traffic equipment based on a multi-level regional hierarchy as described in claim 1, characterized in that, The collected operational status data sequences and environmental parameter sequences of traffic equipment within each management area include: The system collects equipment operating parameters, including current, voltage, temperature, and vibration frequency, by deploying sensor arrays on transportation equipment. Environmental parameters, including ambient temperature, humidity, wind speed, and visibility, are collected using regional environmental monitoring equipment. Set a data acquisition time window to sample equipment operating parameters and environmental parameters at fixed time intervals; The sampled data is timestamped and cleaned to form a regular sequence of operational status data and environmental parameters.

4. The remote management method for traffic equipment based on a multi-level regional hierarchy as described in claim 1, characterized in that, The evolutionary analysis of the equipment integrated state map and environmental parameter sequence using time series prediction algorithms includes: Long Short-Term Memory (LSTM) networks are used to model the feature change trends in the overall device state map; An autoregressive ensemble moving average model was used to predict environmental parameter sequences. The prediction results of equipment characteristic change trends and environmental parameter prediction results are fused to generate equipment status prediction maps and environmental parameter prediction sequences.

5. The remote management method for traffic equipment based on a multi-level regional hierarchy as described in claim 1, characterized in that, The process of calibrating the basic management strategy using equipment status prediction maps and environmental parameter prediction sequences includes: Construct a strategy evaluation model, which includes multiple evaluation indicators and corresponding weight coefficients; The basic management strategy was applied to the equipment status prediction map and the environmental parameter prediction sequence to simulate the effect of the strategy execution. Calculate the deviation between the strategy execution effect and the expected goal, and generate a strategy evaluation score; Based on the strategy evaluation score, the parameter settings of the basic management strategy are adjusted using the gradient descent algorithm; Repeat the strategy evaluation and parameter adjustment process until the strategy evaluation score reaches a preset threshold, and then generate an optimized management strategy.

6. The remote management method for traffic equipment based on a multi-level regional hierarchy as described in claim 5, characterized in that, The evaluation metrics included in the strategy evaluation model are: Equipment operation stability indicators reflect the fluctuation range of equipment operating parameters after the strategy is implemented; Energy efficiency indicators measure the level of energy consumption during strategy execution. Equipment lifespan impact indicators, and the degree of impact of assessment strategies on equipment lifespan; Environmental adaptability indicators characterize the applicability of strategies under different environmental conditions.

7. The remote management method for traffic equipment based on a multi-level regional hierarchy as described in claim 1, characterized in that, The step of converting the optimization management strategy into a sequence of device control commands includes: Analyze and optimize the control parameters and requirements in the management strategy; Based on the specific model and communication protocol of the transportation equipment, generate corresponding equipment control commands; Set the execution time and order of instructions to form a sequence of device control instructions; Perform syntax checking and logic verification on the sequence of equipment control instructions.

8. The remote management method for traffic equipment based on a multi-level regional hierarchy as described in claim 1, characterized in that, The implementation of remote management of multi-level regional transportation equipment includes: The device control command sequence is transmitted to the target management area through a dedicated communication network. The area gateway receives the device control command sequence and performs command parsing and routing forwarding. The target transportation equipment receives and executes control commands to complete equipment status adjustments. The system monitors the execution results of commands in real time and transmits the execution feedback data back to the management center. The device status database is then updated based on the execution feedback data, forming a closed-loop management system.

Images