Traffic influence autonomous evaluation system based on natural language processing and working method

Abstract

The invention discloses a traffic influence autonomous evaluation system based on natural language processing and a working method, and belongs to the technical field of traffic management. In order to solve the problem of difficulty in urban traffic multi-scene analysis iteration, the system comprises a basic data layer, an NLU engine / intelligent processing layer and a user interaction layer, and the basic data layer, the NLU engine / intelligent processing layer and the user interaction layer are connected in sequence; the basic data layer comprises a road network facility database, a model parameter library, a traffic demand matrix and a traffic partition database; the NLU engine / intelligent processing layer comprises a natural language understanding NLU engine, a traffic analysis modeling evaluation module and a report generation module; and the user interaction layer comprises a Web / App graphical user interface, a voice / text input interface, a parameter confirmation and modification panel and an AR / Web visual report board, and realizes interaction between a user and the traffic influence autonomous evaluation system based on natural language processing. The method has the high efficiency of the whole process.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of traffic management, and particularly relates to a traffic impact autonomous evaluation system based on natural language processing and a working method. BACKGROUND

[0002] The reconstruction, expansion and maintenance of urban traffic infrastructure are normal work in urban development. Accurate prediction of the impact of these traffic interventions (such as road closure, new road construction, intersection reconstruction, etc.) on traffic operation is crucial for urban planners, traffic engineers and decision makers.

[0003] Traditional traffic impact assessment mainly relies on professional macro-simulation software such as TransCAD, VISSIM, AIMSUN, etc. Although these software are powerful, they have significant technical barriers and usage obstacles: high professional requirements: operators need to have deep knowledge of traffic engineering and experience in using software; complex operation: complex road network parameters, demand matrix and simulation conditions need to be manually configured; low efficiency: single scene analysis usually requires several hours of manual operation time; difficult iteration: it is difficult to quickly compare multiple schemes and "hypothesis analysis". SUMMARY

[0004] The problem to be solved by the present application is to solve the problem of difficult iteration of urban traffic multi-scene analysis, inability to quickly respond to decision-making needs, and inability of non-professionals to directly use professional traffic simulation tools, and to propose a traffic impact autonomous evaluation system based on natural language processing and a working method.

[0005] To achieve the above-mentioned purpose, the present application realizes the following technical scheme:

[0006] A traffic impact autonomous evaluation system based on natural language processing, comprising a basic data layer, an NLU engine / intelligent processing layer and a user interaction layer, the basic data layer, the NLU engine / intelligent processing layer and the user interaction layer being connected in sequence to realize full-process autonomous processing from natural language input to report generation; the basic data layer comprises a road network facility database, a model parameter library, a traffic demand matrix and a traffic zoning database, the road network facility database stores road network geographic information and point of interest geographic information, the traffic demand matrix stores OD travel data at different time periods, the model parameter library includes calibrated BPR function parameters, road grade parameter mapping table, vehicle conversion coefficient and distribution algorithm parameters, and the traffic zoning database stores traffic cell and administrative area topological structure including geometric attributes and attribute parameters;

[0007] The NLU engine / smart processing layer includes a natural language understanding NLU engine, a traffic analysis modeling evaluation module, and a report generation module. The natural language understanding NLU engine is used to convert natural language input into executable simulation instructions. The traffic analysis modeling evaluation module is used to modify the basic road network topology based on the simulation instructions to generate an intervention scenario road network and extract traffic evaluation indicators. The report generation module is used to generate a traffic impact evaluation report.

[0008] The user interaction layer includes a Web / App graphical user interface, a voice / text input interface, a parameter confirmation and modification panel, and an AR / Web visualization report board, realizing autonomous analysis of user intent and closed-loop interaction of system feedback.

[0009] Further, the natural language understanding NLU engine includes a traffic intervention type scenario classifier, a parameter extraction and verification unit, and a JSON instruction generator. The traffic analysis modeling evaluation module includes a road network dynamic reconstruction algorithm unit, a traffic assignment algorithm library SUMO / Static UE, and a traffic evaluation indicator processing algorithm unit. The report generation module includes a multi-format exporter, a data filling and rendering engine, and a template selection module.

[0010] In the natural language understanding NLU engine, the traffic intervention type is identified by the traffic intervention type scenario classifier. The road name, technical parameter, and spatio-temporal range key information are extracted from the user input by the parameter extraction and verification unit. The standardized JSON instructions are generated by the JSON instruction generator.

[0011] In the traffic analysis modeling evaluation module, the basic road network topology is modified based on the standardized JSON instructions by the road network dynamic reconstruction algorithm unit. The traffic assignment is autonomously executed by the traffic assignment algorithm library integrated SUMO and static user equilibrium algorithm. The traffic evaluation indicators are extracted from the traffic assignment results by the traffic evaluation indicator processing algorithm unit.

[0012] A working method of a traffic impact autonomous evaluation system based on natural language processing, comprising the following steps:

[0013] S1. The user inputs natural language to the traffic impact autonomous evaluation system based on natural language processing through the user interaction layer. The natural language understanding NLU engine of the NLU engine / smart processing layer performs intelligent analysis to generate initial JSON instructions.

[0014] S2. After the natural language understanding NLU engine verifies the initial JSON instructions, the parameter confirmation and modification panel of the user interaction layer is transmitted. After user review or modification, the instructions are parsed and the confirmed JSON instructions are generated.

[0015] S3. The traffic analysis modeling and evaluation module reconstructs the target scenario road network based on the confirmed JSON instructions in combination with the basic data layer;

[0016] S4. The traffic analysis modeling and evaluation module performs traffic assignment and traffic evaluation index calculation on the generated target scenario road network to obtain simulation results;

[0017] S5. The report generation module generates a traffic impact evaluation report based on the obtained simulation results and transmits the report to the user interaction layer for result display.

[0018] Further, the specific implementation method of the natural language understanding NLU engine of the NLU engine / intelligent processing layer in step S1 includes the following steps:

[0019] S1.1. Perform text cleaning and word segmentation processing on the user input natural language to obtain traffic compound word text;

[0020] S1.2. Construct a traffic domain terminology standardization mapping library, and map the obtained traffic compound word text and the traffic domain terminology standardization mapping library to obtain traffic domain terminology standardized text;

[0021] S1.3. The traffic intervention type scene classifier performs scene classification and intent recognition on the obtained traffic domain terminology standardized text, transmits the text that fails to be recognized to the parameter confirmation and modification panel of the user interaction layer, and performs named entity recognition and relationship extraction on the text that is successfully recognized to obtain a semantic relationship network;

[0022] S1.4. The JSON instruction generator generates initial JSON instructions based on the semantic relationship network obtained in step S1.3, and calls the JSON template of the model parameter library in the basic data layer.

[0023] Further, the traffic domain terminology standardization mapping library in step S1.2 includes a standard terminology set, a synonym mapping set, and a classification system;

[0024] The standard terminology set is used to store the standard terminology defined by the traffic system and its machine-readable code;

[0025] The synonym mapping set is used to establish associated diversified expression variants for each standard terminology;

[0026] The classification system is used to logically classify all terminologies to provide context constraints for semantic analysis.

[0027] Further, the JSON template of the model parameter library in the basic data layer in step S1.4 is classified according to the intervention type, including road closure, new road, and road reconstruction;

[0028] The road closure corresponds to intervention_type=1, and the corresponding JSON parameters include RoadName, Section Description, Evaluation Period, and EvaluationArea. The RoadName is set as a mandatory parameter;

[0029] The new road corresponds to intervention_type=2, and the corresponding JSON parameters include Start Point, End Point, Road Class, Lane Count, Capacity, Evaluation Period, Evaluation Area, and RoadName. The Start Point and End Point are set as mandatory parameters;

[0030] The road reconstruction corresponds to intervention_type=3, and the corresponding JSON parameters include RoadName, Road Class, Lane Count, Capacity, Evaluation Period, Evaluation Area, and Section Description. The RoadName and Road Class are set as mandatory parameters.

[0031] Further, in step S2, the natural language understanding NLU engine verifies the initial JSON instruction to verify the completeness of the mandatory parameters.

[0032] Further, the specific implementation method of step S3 includes the following steps:

[0033] S3.1. The traffic analysis modeling and evaluation module performs road network element searching and matching based on the confirmed JSON instruction;

[0034] The RoadName, Start Point, and End Point in the confirmed JSON instruction are used to query the road network facility database;

[0035] For road closure or road reconstruction, the target road section or road section set is determined;

[0036] For new roads, determine the location of Start Point and End Point in the existing road network, judge whether to connect to the existing intersection or specific coordinate point, if the connection point is a new location, then pre-allocate a new node ID;

[0037] S3.2. Perform road network modification operation according to different intervention types.

[0038] Further, the specific implementation method of step S4 includes the following steps:

[0039] S4.1. The traffic analysis modeling evaluation module simulates the behavior of travelers in the generated target scene road network according to the path impedance self-selected path in the road network by using traffic assignment algorithm, and through iterative calculation until the system reaches a balanced state, finally outputs the stable traffic flow, speed distribution data of each road section, node in the road network;

[0040] S4.2. Perform traffic evaluation index calculation, including regional level index, road section level index and node level index;

[0041] S4.2.1. The calculation formula of regional level index is as follows:

[0042] Calculate the total vehicle kilometers TVK in the region:

[0043]

[0044] Wherein, is the flow of road section a, is the length of road section a;

[0045] Calculate the average speed of the region :

[0046]

[0047] Wherein, is the travel time of road section a;

[0048] Calculate the total congestion mileage CL:

[0049]

[0050] Wherein, I is an indicator function, is the congestion speed threshold;

[0051] Calculate the congestion cost CC:

[0052]

[0053] Wherein, is the free flow time, VOT is the time value, actual travel time for link a;

[0054] S4.2.2. Setting link-level indicators includes link flow, saturation, average speed and delay time;

[0055] S4.2.3. Setting node-level indicators includes node total delay and service level.

[0056] Advantages of the present application:

[0057] The traffic impact autonomous evaluation system based on natural language processing provided by the present application proposes a system architecture supporting full-process autonomous processing. The system includes a basic data layer, an NLU engine / intelligent processing layer and a user interaction layer connected in turn, forming an autonomous closed loop from semantic understanding to simulation evaluation. The basic data layer integrates a road network facility database, a model parameter library, a traffic demand matrix and a traffic zoning database, providing structured data support for autonomous evaluation; the NLU engine / intelligent processing layer covers a natural language understanding NLU engine, a traffic analysis modeling and evaluation module and a report generation module, realizing full-process autonomous processing from user natural language input to simulation instruction generation, road network dynamic reconstruction, traffic assignment calculation and indicator extraction; the user interaction layer provides a Web / App graphical interface, a voice / text input interface, a parameter confirmation and modification panel and an AR / Web visual report board, supporting users to trigger and supervise the autonomous evaluation process in a natural language interaction manner. The present application realizes the intelligentization, automation and low threshold of traffic impact evaluation by integrating natural language processing and traffic simulation technology, significantly improves the efficiency of multi-scenario iterative analysis, and provides efficient decision support for traffic planning and management.

[0058] The traffic impact autonomous evaluation system based on natural language processing provided by the present application realizes that users do not need any professional software operation skills, and can complete the description and confirmation of complex evaluation tasks using daily language, which completely breaks down the professional barriers. The present application realizes the minute-level automatic conversion from "language description" to "simulation-ready road network", shortens the parameter configuration and road network modeling work in the traditional method from hours to seconds, and reduces the overall analysis time from several hours to less than 5 minutes.

[0059] The traffic influence autonomous evaluation system based on natural language processing avoids misjudgment through confidence decision, avoids omissions through parameter reasoning and closed-loop correction, guarantees model accuracy through an accurate topology reconstruction algorithm, eliminates human setting errors from multiple links, and ensures the reliability of simulation results. Through the established multi-level quantitative evaluation index system, the depth of structured analysis and decision support capability that cannot be achieved by traditional methods is provided. The application has high efficiency in the whole process, and decision makers can quickly perform "hypothesis analysis" and comparison of multiple scenarios and multiple schemes, obtain quantitative impact results of different intervention strategies in a short time, and make decisions more timely and scientific. BRIEF DESCRIPTION OF DRAWINGS

[0060] Figure 1 A structural diagram of the traffic influence autonomous evaluation system based on natural language processing;

[0061] Figure 2 A structural diagram of the traffic influence autonomous evaluation system based on natural language processing;

[0062] Figure 3 A flowchart of the working method of the traffic influence autonomous evaluation system based on natural language processing. DETAILED DESCRIPTION

[0063] In order to make the purpose, technical scheme and advantages of the present application clearer, the present application is further described in detail below in combination with the drawings and specific embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application, that is, the described specific embodiments are only a part of the embodiments of the present application, but not all the specific embodiments. The components of the specific embodiments of the present application described and shown in the drawings can be arranged and designed in various different configurations, and the present application can also have other embodiments.

[0064] Therefore, the detailed description of the specific embodiments of the present application provided below in the drawings is not intended to limit the scope of the claimed present application, but only represents selected specific embodiments of the present application. Based on the specific embodiments of the present application, all other specific embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0065] In order to further understand the invention content, characteristics and effects of the present application, the following specific embodiments are exemplified, and the drawings and specific embodiments are combined Figure 1 and the drawings Figure 3 are described in detail as follows:

[0066] Example 1:

[0067] A traffic impact autonomous evaluation system based on natural language processing, comprising a basic data layer, an NLU engine / smart processing layer, and a user interaction layer, which are connected in sequence to realize full-process autonomous processing from natural language input to report generation; the basic data layer comprises a road network facility database, a model parameter library, a traffic demand matrix, and a traffic zoning database, the road network facility database stores road network geographic information and point of interest geographic information, the traffic demand matrix stores OD travel data at different time periods, the model parameter library includes calibrated BPR function parameters, road grade parameter mapping tables, vehicle conversion coefficients, and distribution algorithm parameters, and the traffic zoning database stores traffic cells including geometric attributes and attribute parameters and administrative district topological structures;

[0068] The NLU engine / smart processing layer comprises a natural language understanding NLU engine, a traffic analysis modeling and evaluation module, and a report generation module, the natural language understanding NLU engine is used to convert natural language input into executable simulation instructions, the traffic analysis modeling and evaluation module is used to generate an intervention scenario road network based on the simulation instructions and modify the basic road network topology, and extract traffic evaluation indicators, and the report generation module is used to generate a traffic impact evaluation report;

[0069] The user interaction layer comprises a Web / App graphical user interface, a voice / text input interface, a parameter confirmation and modification panel, and an AR / Web visual report board, and realizes closed-loop interaction of autonomous analysis of user intent and system feedback.

[0070] Further, the natural language understanding NLU engine comprises a traffic intervention type scenario classifier, a parameter extraction and verification unit, and a JSON instruction generator, the traffic analysis modeling and evaluation module comprises a road network dynamic reconstruction algorithm unit, a traffic distribution algorithm library SUMO / Static UE, and a traffic evaluation indicator processing algorithm unit, and the report generation module comprises a multi-format exporter, a data filling and rendering engine, and a template selection module;

[0071] In the natural language understanding NLU engine, the intervention type is identified by the traffic intervention type scenario classifier, road names, technical parameters, and time-space range key information are extracted from user input by the parameter extraction and verification unit, and standardized JSON instructions are generated by the JSON instruction generator;

[0072] In the traffic analysis modeling and evaluation module, the basic road network topology is modified based on the standardized JSON instructions by the road network dynamic reconstruction algorithm unit, traffic distribution is autonomously executed by the traffic distribution algorithm library integrated with SUMO and static user equilibrium algorithm, and traffic evaluation indicators are extracted from the traffic distribution results by the traffic evaluation indicator processing algorithm unit.

[0073] Embodiment 2:

[0074] A working method of a natural language processing-based traffic impact autonomous evaluation system according to embodiment 1, comprising the following steps:

[0075] S1. A user inputs natural language to the natural language processing-based traffic impact autonomous evaluation system through a user interaction layer, and a natural language understanding NLU engine of an NLU engine / smart processing layer intelligently analyzes and generates an initial JSON instruction;

[0076] Further, the specific implementation method of the intelligent analysis of the natural language understanding NLU engine of the NLU engine / smart processing layer in step S1 comprises the following steps:

[0077] S1.1. Text cleaning and word segmentation processing are performed on the natural language input by the user to obtain a traffic compound word text;

[0078] S1.2. A traffic domain term standardization mapping library is constructed, and the obtained traffic compound word text and the traffic domain term standardization mapping library are mapped to obtain a traffic domain term standardized text;

[0079] Further, the traffic domain term standardization mapping library in step S1.2 comprises a standard term set, a synonym mapping set, and a classification system;

[0080] The standard term set is used to store the standard terms defined by the traffic system and the machine-readable codes thereof, as shown in Table 1:

[0081] Table 1

[0082]

[0083] The synonym mapping set is used to establish a corresponding diversified expression variant for each standard term, and a diversified expression variant (such as “closed” and “closed road” corresponding to “closed”) associated with each standard term is established. This table is the core, which stores the mapping relationship between all user input variants and their corresponding standard terms. As shown in Table 2:

[0084] Table 2

[0085]

[0086] The classification system is used to logically classify all terms to provide context constraints for semantic analysis, as shown in Table 3:

[0087] Table 3

[0088]

[0089] Further, the intelligent matching and mapping process is:

[0090] Input: Tokenized sequence of user's raw text.

[0091] Library query and matching: The system scans each tokenized unit in the synonym mapping set, and finds the corresponding standard term according to the preset matching rule (such as exact match, prefix match).

[0092] Ambiguity resolution: When the same variant may be mapped to multiple standard terms, the system makes a comprehensive decision according to the mapping confidence, usage frequency and context scenario, and selects the most likely mapping item.

[0093] Output: Replace the colloquial and diversified expressions in the original text with a unified and system-recognizable standard term sequence. This method efficiently and accurately converts scattered and ambiguous natural language expressions into structured and explicit machine instructions through table lookup and rule-based methods, thereby reducing the analysis difficulty of subsequent modules.

[0094] S1.3. The traffic intervention type scene classifier classifies the text after the traffic domain term standardization, and performs intent recognition. The text that fails to be recognized is transmitted to the parameter confirmation and modification panel of the user interaction layer, and the text that is successfully recognized is subjected to named entity recognition and relationship extraction to obtain a semantic relationship network.

[0095] Further, the detailed process is as follows:

[0096] Feature extraction and classification: The standardized text is input into a multi-classification model based on deep learning (such as a text classifier using the BERT architecture). The model outputs a probability vector representing the confidence that the input text belongs to the pre-defined three types of intervention measures (road closure, new road, road reconstruction).

[0097] Given the input text x, the model outputs a three-dimensional vector z=(z1,z2,z3), representing the original scores of the three categories (road closure, new road, road reconstruction). Then, the Softmax function is used to convert it into a probability distribution:

[0098]

[0099] Where: is the predicted probability that the text x belongs to the i-th category.

[0100] The sum of the probabilities of all three categories is 1, i.e.

[0101] Confidence decision and interaction trigger: the highest predicted probability in the three categories is defined as the "confidence" of this classification by the system. The calculation formula is as follows:

[0102] ;

[0103] The system sets a confidence threshold T, which is a hyperparameter that needs to be determined experimentally during system development. It represents a trade-off between precision and recall. Here, it is set to 0.85. The decision rules are as follows:

[0104] if Then: Classification successful. System adoption. The corresponding category is taken as the final result, and the process continues.

[0105] Otherwise: The classification is deemed unsuccessful or the confidence level is insufficient. The system will interrupt the automated process and trigger interactive logic. Classification is only considered successful and the process continues if the highest confidence level exceeds this threshold. Otherwise, the system will interrupt the process and send a friendly message to the user: "Sorry, we did not understand your specific plan. Please clarify whether you want to close the road, build a new road, or renovate and expand an existing road?" This mechanism ensures that the system will not force execution when it "does not understand," avoiding the accumulation of errors in subsequent operations.

[0106] Further details of the named entity recognition and relation extraction process:

[0107] Domain Entity Recognition: The system uses a sequence labeling-based neural network model (such as BERT-BiLSTM-CRF), which has been trained on a large amount of traffic text and can accurately identify and label entities in the text.

[0108] For example, given the input "closed Shennan Avenue from Hongli Road to the ring road section", the model will label it as: `[ACTION: closed]`, `[ROAD_NAME: Shennan Avenue]`, `[START_POINT: Hongli Road]`, `[END_POINT: ring road section]`.

[0109] Semantic relation network construction: Entities are isolated points, and relations are the lines connecting them. The system constructs semantic relations between entities by analyzing the dependency syntax structure of sentences.

[0110] Action-target relationship: Determine which `ROAD_NAME` (Shennan Avenue) is applied to by `ACTION` (closed).

[0111] Spatial Relationship: Associate `START_POINT` (Hongli Road) and `END_POINT` (Ring Road Section) with `ROAD_NAME` (Shennan Avenue) to form a complete `ROAD_SECTION` (Road Section Description).

[0112] Attribute relationship: Bind `TECH_PARAM` (e.g. "6 lanes") to the entity it modifies.

[0113] S1.4. The JSON instruction generator generates the initial JSON instruction based on the semantic relationship network obtained in step S1.3, calling the JSON template of the model parameter library in the underlying data layer.

[0114] Further, the JSON template of the model parameter library in the underlying data layer in step S1.4 is classified according to the intervention type, including road closure, new road, road reconstruction;

[0115] Road closure corresponds to intervention_type=1, and the corresponding JSON parameters include road name RoadName, section description Section Description, evaluation period Evaluation Period, and evaluation area EvaluationArea. Set the road name RoadName as a required parameter;

[0116] New road corresponds to intervention_type=2, and the corresponding JSON parameters include start point Start Point, end point End Point, road class Road Class, lane count Lane Count, capacity Capacity, evaluation period Evaluation Period, evaluation area Evaluation Area, and road name Road Name. Set the start point Start Point and end point End Point as required parameters;

[0117] Road reconstruction corresponds to intervention_type=3, and the corresponding JSON parameters include road name RoadName, road class Road Class, lane count Lane Count, capacity Capacity, evaluation period Evaluation Period, evaluation area Evaluation Area, and section description Section Description. Set the road name RoadName and road class Road Class as required parameters.

[0118] Further, the parameter mapping and structured JSON generation process is as follows:

[0119] Template selection: According to the intervention type determined in S1.2, the system calls the corresponding JSON template from the parameter template library. For example, the "road closure" template predefines fields such as `road_name`, `section`, `capacity_after` (fixed as 0).

[0120] Field Filling & Intelligent Inference:

[0121] Direct Mapping: Recognized entities are directly filled into corresponding fields (e.g., `ROAD_NAME` -> `"road_name"`).

[0122] Logical Inference & Default Value Completion: For optional parameters not explicitly specified by the user, the system intelligently completes them based on an internal traffic knowledge base. For example, if the user does not specify the evaluation period, the system defaults to `"AM_PEAK"`; if the user specifies `"road_class": "expressway"` in "New Road" but does not provide the number of lanes, the system infers `"lane_count": 6` as the default value from the knowledge base.

[0123] Generate Preliminary JSON: After filling and inferring all fields, the system generates a preliminary, complete JSON object.

[0124] S2. After the natural language understanding NLU engine verifies the initial JSON instructions, it transmits them to the parameter confirmation and modification panel of the user interaction layer for user review or modification, and then performs instruction analysis and generates confirmed JSON instructions;

[0125] Further, in step S2, the natural language understanding NLU engine verifies the initial JSON instructions, verifying the completeness of mandatory parameters.

[0126] Further, the detailed process is as follows:

[0127] S2.1 System background verification: The system first verifies the mandatory parameters of the preliminary generated JSON. Check if all parameters that must be provided by the user or explicitly inferred by the system are present and valid under this intervention type.

[0128] S2.2 Front-end interaction and correction;

[0129] Case A: Mandatory parameter missing: If the verification finds that the mandatory parameter is missing (such as "New Road" missing `"end_point"`), the system will explicitly list all missing items and prompt the user to supplement (for example: "Please provide the end point of the new road.").

[0130] Case B: Parameters are complete: The system displays all parameters (including system-completed default values) to the user in a clear and readable form (such as a form or highlighted JSON).

[0131] S2.3 User-led final confirmation;

[0132] Review: User checks all parameters to make sure they are as intended.

[0133] Modify: User can modify any parameter value, including the system-completed default values.

[0134] Confirm and Submit: User confirms that there are no errors and clicks submit. Once the user modifies the parameters, the system receives the modified data and immediately jumps back to S2.1, re-executes the parameter mapping and JSON generation process based on the user's latest modified data, forming a partial closed loop, ensuring that the latest and most accurate instructions are submitted to the simulation engine.

[0135] This detailed process describes the internal operating mechanism and interaction logic from the initial JSON parameter generation to the final confirmation. It highlights the balance the system has achieved between automated processing and human supervision, through key proactive interactions (requesting supplements when parameters are incomplete) and a correctable confirmation closed loop, forming a robust, reliable, and user-friendly natural language understanding and parameter mapping system.

[0136] S3. Traffic analysis modeling and evaluation module based on confirmed JSON instructions, combined with the basic data layer to dynamically reconstruct the road network, generate the target scenario road network;

[0137] Further, the specific implementation method of step S3 includes the following steps:

[0138] S3.1. Traffic analysis modeling and evaluation module based on confirmed JSON instructions, road network element search and matching;

[0139] Using the Road Name, Start Point, and End Point in the confirmed JSON instructions to query in the road network facility database;

[0140] For road closures or road reconstruction, determine the target road segment or road segment set;

[0141] For newly built roads, determine the location of the Start Point and End Point in the existing road network, judge whether to connect to the existing intersection or specific coordinate point, if the connection point is a new location, then pre-allocate a new node ID;

[0142] S3.2. Perform road network modification operations according to different intervention types;

[0143] S3.2.1. The road network modification method for road closure is to first copy the original road network G, generate the target scenario road network G ’ ; based on the road segment identifier in the confirmed JSON instructions, locate the target road segment a in G ’ , set the free-flow travel time of the target road segment a to , the travel time of the target link a is obtained according to the BPR function is expressed as:

[0144]

[0145] wherein flow a is the distribution flow of the link a, capacity a is the passing capacity of the link a; and alpha and beta are respectively the first function parameter and the second function parameter;

[0146] S3.2.2. The road network modification method for the newly-built road is:

[0147] When the end point (the starting point or the terminal point) of the newly-built road is close to the existing node, the spatial threshold is adopted for association, and the nearest associated point is found within the specified distance threshold range;

[0148] Supposing that the end point coordinate of the newly-built road is P new =(x new ,y new ), the existing node coordinate is P existing =(x i ,y i ), and the associated distance threshold parameter is defined as d threshold , the association condition is calculated as follows:

[0149]

[0150] wherein x new is the x-axis coordinate of the end point of the newly-built road, y new is the y-axis coordinate of the end point of the newly-built road, x i is the x-axis coordinate of the existing node, y i is the y-axis coordinate of the existing node, and d is the associated distance. The nodes satisfying the above threshold condition are constituted into a candidate node set N candidate ;

[0151] In the candidate node set, the node selection strategy calculation formula is:

[0152]

[0153] wherein P selected is the existing node finally selected by the system and to be associated with the end point of the new road, argmin is the selection function, and returns the coordinate result of a candidate node with the minimum distance value. P i =(x i ,y i ) is the node belonging to the candidate node set N candidate ;

[0154] When the endpoint (start or end) of a newly constructed road is not within the threshold range of any existing node, a new node is inserted on the nearest road segment. The nearest road segment selection method is as follows: Let the set of existing road segments be A, and for each road segment a... j ∈A, calculate the distance from the endpoint (starting or ending point) of the newly built road to the nearest road segment:

[0155]

[0156] Where (x1,y1) and (x2,y2) are the coordinates of the first and second endpoints of the road segment;

[0157] Calculate the parameter t of the perpendicular point on the road segment, and calculate the coordinates of the perpendicular point;

[0158]

[0159]

[0160] in, Let x be the x-coordinate of the foot of the perpendicular. The y-coordinate of the perpendicular foot point;

[0161] Set the constraint condition as 0 ≤ t ≤ 1;

[0162] Then, the relevant attributes of the segmented road are updated. The original road segment 'a' is divided into a first new road segment 'a1' and a second new road segment 'a2'. Attributes are inherited proportionally to their length, calculated using the following formula:

[0163]

[0164] Among them, L a L is the length of the original road segment a. a1 L represents the length of the first new road segment a1 after the division. a2 This is the length of the second new road segment a2 after the division;

[0165] For capacity and lane number attributes, inherit the original values; free-flow travel time is adjusted proportionally to length.

[0166]

[0167] in, L represents the free-flow travel time of the original road segment a. a1 L represents the free-flow travel time of the first new road segment a1 after the split. a2 The free-flow travel time for the second new road segment a2 after the split;

[0168] To assign values ​​to the attributes of newly created roads, the user-specified values ​​are first retrieved from the confirmed JSON instructions; if parameters are missing, they are derived from the road level parameter mapping table based on the road level.

[0169]

[0170] wherein lane_count n is the number of lanes of road n, Capacity n is the capacity of road n, K[] is a mapping function that calls the road class parameter mapping table, road_class is the road class parameter, default_lanes, capacity_pre_lane are the lane number and lane capacity parameter values related to the road class in the road class parameter mapping table.

[0171] The link length geometry attribute is calculated as follows:

[0172]

[0173] The free flow travel time is calculated as follows:

[0174]

[0175] wherein, is the design speed, from user input or knowledge base default value.

[0176] S3.2.3. The road network modification method for road reconstruction is direct parameter update and associated parameter derivation;

[0177] The direct parameter update is to cover the original attributes of the target link with the new parameters in the confirmed JSON instruction;

[0178] The associated parameter derivation is to estimate according to the new lane number and road class when the lane number changes, and the capacity update calculation formula is as follows:

[0179]

[0180] The free flow travel time update calculation formula is as follows:

[0181]

[0182] wherein, is the design speed.

[0183] Further, global topological consistency check and output are performed;

[0184] After all modifications are completed, final quality control is performed to ensure the generated road network is topologically correct and consistent. This includes: 1. Connectivity check: Checking for isolated nodes or invalid connections resulting from "road closure" operations. Ensuring all valid road segments are connected by some path. 2. Logical verification: For example, verifying that the start and end points of newly built roads have been correctly integrated into adjacency relationships. 3. Generating the target scenario road network: Packaging the checked and modified road network data into a new, independent road network data file or memory object, labeled as the "target scenario road network". 4. Output: Outputting this target scenario road network to the traffic simulation engine as input for this impact assessment.

[0185] This detailed process reveals how dynamic road network reconstruction transforms abstract intervention commands into concrete, computable road network changes. Through categorized atomic operations and a final global consistency check, the system achieves a high degree of automation while rigorously ensuring the data integrity and topological correctness required for traffic simulation.

[0186] S4. The traffic analysis, modeling, and evaluation module performs traffic assignment and traffic evaluation index calculations on the generated target scenario road network to obtain simulation results;

[0187] Furthermore, the specific implementation method of step S4 includes the following steps:

[0188] S4.1. The traffic analysis modeling and evaluation module uses a traffic assignment algorithm to simulate the behavior of travelers autonomously choosing paths based on path impedance in the generated target scenario road network. Through iterative calculation until the system reaches a balanced state, it finally outputs stable traffic flow and speed distribution data for each road segment and node in the road network.

[0189] S4.2. Calculate traffic assessment indicators, including regional, road segment, and node-level indicators;

[0190] S4.2.1. The calculation formula for regional indicators is set as follows:

[0191] Calculate the total vehicle mileage (TVK) for the area:

[0192]

[0193] in, Let the traffic flow be that of road segment a. Let a be the length of road segment a;

[0194] Calculate the average velocity of the region :

[0195]

[0196] in, Travel time for link a;

[0197] Calculate total congestion mileage CL:

[0198]

[0199] where I is an indicator function, is the congestion speed threshold;

[0200] Calculate congestion cost CC:

[0201]

[0202] where, is the free flow time, VOT is the time value, is the actual travel time for link a;

[0203] S4.2.2. Set link-level indicators including link flow, saturation, average speed, and delay time;

[0204] Further, the calculation formula of saturation is VOC a = Flow a / Capacity a ;

[0205] The calculation formula of delay time is ;

[0206] S4.2.3. Set node-level indicators including node total delay and service level;

[0207] Further, the calculation formula of node total delay is:

[0208]

[0209] where I n is all the inlet roads of node n.

[0210] The calculation formula of service level is:

[0211]

[0212] According to the Road Capacity Manual and other standards, delay is mapped to A-F levels.

[0213] S5. The report generation module generates a traffic impact assessment report based on the simulation results obtained, and transmits it to the user interaction layer for result display.

[0214] Further, the numerical simulation results are automatically converted into easy-to-understand, industry-standard professional reports.

[0215] Result delivery and template selection: the simulation results of S4 are delivered to the report generation layer. The system automatically selects the most appropriate report template according to the intervention type. Automatic report generation: based on the template-driven architecture, the data filling and rendering engine integrates the simulation indicators, analysis conclusions, comparison charts, etc. into the template organically. Multi-format output: the report generator generates PDF, Word and interactive Web reports at the same time, meeting different needs such as archiving, editing and online viewing.

[0216] The embodiment has been verified on the actual road network of Shenzhen City (including 19902 nodes and 32061 road segments). The following typical examples illustrate the actual operation effect of the system according to the S1-S5 process, proving the practicality and effectiveness of the technology.

[0217] Typical example 1, as shown in Table 4, is an example of road closure.

[0218] Table 4

[0219]

[0220] Typical example 2, as shown in Table 5, is an example of new road construction.

[0221] Table 5

[0222]

[0223] Typical example 3, as shown in Table 6, is an example of road reconstruction.

[0224] Table 6

[0225]

[0226] It should be noted that the relational terms, such as "first" and "second", and the like, are used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any such actual relationship or order between such entities or actions. Moreover, the terms "comprising", "including", or any other variant thereof are intended to cover non-exclusive inclusion, so that a process, method, article, or apparatus that includes a list of elements does not necessarily include only those elements in the list, but can include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without more limitations, an element defined by the phrase "comprising a" does not exclude the existence of additional identical elements in the process, method, article, or apparatus including the element.

[0227] Although the present application has been described with reference to the specific embodiments thereof, it should be understood by those skilled in the art that various changes can be made and equivalents can be substituted for elements thereof without departing from the scope of the present application. In particular, various features and aspects of the present application can be used individually or in any combination depending on the specific application and implementation. Therefore, it is expressly intended that the specific embodiments of the present application both as set forth and including any equivalents thereof should not limit the present application or scope of the claims herein, but rather the overall scope of pertaining solely to the methods and the articles of manufacture specifically recited in the following claims.

Claims

1. A natural language processing based traffic impact autonomous assessment system, characterized in that, The system comprises a basic data layer, an NLU engine / smart processing layer and a user interaction layer, which are sequentially connected to realize autonomous processing of the whole process from natural language input to report generation. The basic data layer comprises a road network facility database, a model parameter library, a traffic demand matrix and a traffic zoning database, the road network facility database stores road network geographic information and point of interest geographic information, the traffic demand matrix stores OD travel data in different time periods, the model parameter library comprises calibrated BPR function parameters, road grade parameter mapping table, vehicle conversion coefficient and distribution algorithm parameters, and the traffic zoning database stores traffic cells including geometric attributes and attribute parameters and administrative area topology structure. The NLU engine / smart processing layer comprises a natural language understanding NLU engine, a traffic analysis modeling and evaluation module and a report generation module, the natural language understanding NLU engine is used to convert natural language input into executable simulation instructions, the traffic analysis modeling and evaluation module is used to generate an intervention scenario road network based on the simulation instructions and modify the basic road network topology and extract traffic evaluation indexes, and the report generation module is used to generate a traffic impact evaluation report. The user interaction layer comprises a Web / App graphical user interface, a voice / text input interface, a parameter confirmation and modification panel and an AR / Web visual report board, and realizes closed-loop interaction of autonomous analysis of user intention and system feedback.

2. The traffic impact autonomous evaluation system based on natural language processing according to claim 1, wherein, The natural language understanding NLU engine comprises a traffic intervention type scenario classifier, a parameter extraction and verification unit and a JSON instruction generator, the traffic analysis modeling and evaluation module comprises a road network dynamic reconstruction algorithm unit, a traffic distribution algorithm library SUMO / Static UE and a traffic evaluation index processing algorithm unit, and the report generation module comprises a multi-format exporter, a data filling and rendering engine and a template selection module. In the natural language understanding NLU engine, the intervention type is identified by the traffic intervention type scenario classifier, the road name, technical parameter and time-space range key information are extracted from the user input by the parameter extraction and verification unit, and the standardized JSON instruction is generated by the JSON instruction generator. In the traffic analysis modeling and evaluation module, the basic road network topology is modified based on the standardized JSON instruction by the road network dynamic reconstruction algorithm unit, the traffic distribution is autonomously executed by the traffic distribution algorithm library integrated with SUMO and static user equilibrium algorithm, and the traffic evaluation indexes are extracted from the traffic distribution result by the traffic evaluation index processing algorithm unit.

3. The working method of the traffic impact autonomous evaluation system based on natural language processing according to any one of claims 1-2, characterized in that, The method comprises the following steps: S1. The user inputs natural language to the natural language processing-based traffic impact autonomous evaluation system through the user interaction layer, the natural language understanding NLU engine of the NLU engine / smart processing layer performs intelligent analysis, and generates an initial JSON instruction; S2. After the natural language understanding NLU engine verifies the initial JSON instruction, the verified JSON instruction is transmitted to the parameter confirmation and modification panel of the user interaction layer, and after the user audits or modifies the JSON instruction, the instruction is parsed and a confirmed JSON instruction is generated. S3. The traffic analysis modeling and evaluation module reconstructs the target scenario road network based on the confirmed JSON instructions in combination with the basic data layer; S4. The traffic analysis modeling and evaluation module performs traffic assignment and traffic evaluation index calculation on the generated target scenario road network to obtain simulation results; S5. The report generation module generates a traffic impact evaluation report based on the obtained simulation results and transmits the report to the user interaction layer for result display.

4. The working method of the traffic impact autonomous evaluation system based on natural language processing according to claim 3, characterized in that, The specific implementation method of the natural language understanding NLU engine of the NLU engine / smart processing layer in step S1 includes the following steps: S1.

1. Perform text cleaning and word segmentation processing on the user input natural language to obtain traffic compound word text; S1.

2. Construct a traffic domain terminology standardization mapping library, and map the obtained traffic compound word text and the traffic domain terminology standardization mapping library to obtain traffic domain terminology standardized text; S1.

3. The traffic intervention type scenario classifier classifies and identifies the intent of the obtained traffic domain terminology standardized text, transmits the text that fails to be identified to the parameter confirmation and modification panel of the user interaction layer, and performs named entity recognition and relationship extraction on the text that is successfully identified to obtain a semantic relationship network; S1.

4. The JSON instruction generator generates an initial JSON instruction based on the semantic relationship network obtained in step S1.3, and calls the JSON template of the model parameter library in the basic data layer.

5. The working method of the traffic impact autonomous evaluation system based on natural language processing according to claim 4, characterized in that, The traffic domain terminology standardization mapping library in step S1.2 includes a standard terminology set, a synonym mapping set, and a classification system; The standard terminology set is used to store the standard terminology defined by the traffic system and its machine-readable code; The synonym mapping set is used to establish a corresponding diversified expression variant for each standard terminology; The classification system is used to logically classify all terminologies to provide context constraints for semantic analysis.

6. The working method of the traffic impact autonomous evaluation system based on natural language processing according to claim 5, characterized in that, The JSON template of the model parameter library in the basic data layer in step S1.4 is classified according to the intervention type, including road closure, new road, and road reconstruction; Road closure corresponds to intervention_type=1, and the corresponding JSON parameters include Road Name, Section Description, Evaluation Period, and Evaluation Area. The Road Name is set as a required parameter; New road corresponds to intervention_type=2, and the corresponding JSON parameters include Start Point, End Point, Road Class, Lane Count, Capacity, Evaluation Period, Evaluation Area, and Road Name. The Start Point and End Point are set as required parameters; The road reconstruction corresponds to intervention_type=3, and the corresponding JSON parameters include Road Name, Road Class, Lane Count, Capacity, Evaluation Period, Evaluation Area, and Section Description. The Road Name and Road Class are mandatory parameters.

7. The working method of the traffic impact autonomous evaluation system based on natural language processing according to claim 6, characterized in that, In step S2, the natural language understanding (NLU) engine verifies the initial JSON instructions and checks the completeness of the mandatory parameters. 8.The working method of the traffic impact autonomous evaluation system based on natural language processing according to claim 7, wherein, The specific implementation method of step S3 includes the following steps: S3.

1. The traffic analysis modeling and evaluation module performs road network element searching and matching based on the confirmed JSON instructions. The Road Name, Start Point, and End Point in the confirmed JSON instructions are used to query the road network facility database. For road closure or road reconstruction, the target road segment or road segment set is determined. For a new road, the positions of the Start Point and End Point in the existing road network are determined, and it is judged whether they are connected to an existing intersection or a specific coordinate point. If the connection point is a new location, a new node ID is assigned in advance. S3.

2. Perform road network modification operations according to different intervention types. S3.2.

1. The road network modification method for road closure is to first copy the original road network G, generate the target scene road network G ’ ; based on the road segment identifier in the confirmed JSON instruction, locate the target road segment a in G ’ , set the free flow travel time of the target road segment a to , and obtain the travel time of the target road segment a according to the BPR function , which is expressed as: ； wherein flow a is the assigned flow for link a, capacity a is the capacity of link a; and a and β are first and second function parameters, respectively. S3.2.

2. The road network modification method for a new road is as follows: When the endpoint of the new road is close to an existing node, a spatial threshold is used for association. Within a specified distance threshold, the closest associated point is found. The endpoint is the starting point or the ending point. Let the end point coordinate of the new road be P new =(x new ,y new ), the existing node coordinate P existing =(x i ,y i ), define the associated distance threshold parameter as d threshold , and the associated condition is calculated as follows: ； wherein x new is the x-axis coordinate of the endpoint of the new road, y new is the y-axis coordinate of the endpoint of the new road, x i is the x-axis coordinate of the existing node, y i is the y-axis coordinate of the existing node, and d is the association distance candidate ; In the candidate node set, the node selection strategy calculation formula is as follows: ； where P selected The system finally selects an existing node to be associated with the new road endpoint, argmin is a selection function that returns the coordinates of the candidate node with the smallest distance value, P i = (x i ,y i ) is a node belonging to the set of candidate nodes N candidate . When the end point of the new road is not within the threshold range of any existing node, a new node is inserted on the nearest road segment; the nearest road segment selection method is to set the existing road segment set as A, for each road segment a j ∈A, calculate the distance from the end point of the new road to the nearest road segment: ； where (x1, y1) and (x2, y2) are the first endpoint coordinates and the second endpoint coordinates of the road segment. Calculate the parameter t of the foot point on the road segment, and calculate the foot point coordinates. ； ； wherein is the x-axis coordinate of the foot point, is the y-axis coordinate of the foot point; The constraint condition is set as 0 ≤ t ≤ 1. Then update the relevant attributes of the segmented road segment. The original road segment a is divided into the first new road segment a1 and the second new road segment a2. The attributes are inherited according to the length ratio, and the calculation formula is as follows: ； wherein, L a is the length of the original road segment a, L a1 is the length of the first new road segment a1 after the split, L a2 is the length of the second new road segment a2 after the split; For the capacity and lane count attributes, the original values are inherited. The free-flow travel time is adjusted according to the length ratio: ； wherein, is the free flow travel time of the original link a, L a1 is the free flow travel time of the first new link a1 after splitting, L a2 is the free flow travel time of the second new link a2 after splitting; Assign values to the attributes of the new road. First, obtain the user-specified values from the confirmed JSON instructions. When the parameters are missing, derive them from the road class parameter mapping table based on the road class, and obtain: ； wherein lane_count n is the number of lanes of road n, Capacity n is the capacity of road n, K[] is a mapping function for calling a road class parameter mapping table, road_class is a road class parameter, and default_lanes and capacity_pre_lane are the number of lanes and the lane capacity parameter values related to the road class in the road class parameter mapping table. S3.2.

3. The road network modification method for road reconstruction is direct parameter updating and associated parameter derivation. Direct parameter updating is to cover the original attributes of the target road segment with the new parameters in the confirmed JSON instructions. When the lane count changes, the capacity is updated according to the new lane count and the road class. The calculation formula is as follows: ； The free-flow travel time is updated according to the following calculation formula: ； wherein, is the design speed for road n.

9. The working method of the traffic impact autonomous evaluation system based on natural language processing according to claim 8, characterized in that, The specific implementation method of step S4 includes the following steps: S4.

1. The traffic analysis modeling evaluation module simulates the behavior of travelers in the generated target scene road network selecting paths according to path impedance by using traffic assignment algorithm, and iteratively calculates until the system reaches a balanced state, finally outputs the stable traffic flow, speed distribution data of each road section and node in the road network; S4.

2. Perform traffic evaluation index calculation, including regional level index, road section level index and node level index; S4.2.

1. The calculation formula of the regional level index is as follows: Calculate the total vehicle kilometers TVK: ； wherein, is the flow for link a, is the length of link a; Computing zone average speed : ； wherein, is the travel time for link a; Calculate the total congestion mileage CL: ； wherein I is an indicator function, is a congestion speed threshold value; Calculate the congestion cost CC: ； wherein, VOT is the value of time, and is the actual travel time for link a. S4.2.

2. The road section level index includes road section flow, saturation, average speed and delay time; S4.2.

3. The node level index includes node total delay and service level.

Images