Road network state evaluation method based on random forest and natural language model

Abstract

The invention discloses a road network state evaluation method based on a random forest and a natural language model. The method comprises the following steps: collecting data; calculating indexes; constructing a training set; constructing a road network evaluation random forest model; road network evaluation; and the evaluation result is natural and linguistic. The invention provides a scheme for quickly judging the highway accident site position, fully mobilizes the function of highway toll collection flow, plays a role in information identification of highway guardrails, solves the problem of quickly judging the highway accident site position, and accelerates the processing process of highway traffic accidents. According to the method, a complete traffic evaluation index system is formed, various traffic flow evaluation indexes are provided, big data indexes can be directly substituted into the traffic flow evaluation indexes for calculation, the operation state of the expressway network is evaluated through an index classification method, and a data basis is provided for optimization of the operation state of the expressway network.

Description

Technical Field

[0001] The present invention relates to the field of traffic safety, and in particular to a road network status evaluation method based on random forest and natural language models. Background Art

[0002] The research on comprehensive evaluation of highway network traffic flow status is the key to solving highway traffic problems. The three-parameter traffic flow model and the three-phase traffic flow model describe the characteristics of vehicles driving on the road from a microscopic level. However, in many cases, the traffic flow on a section of road will be affected by other sections. If multiple traffic flow models are used to describe it, it will cause redundancy in parameters. The traffic flow macro basic diagram pays more attention to the relationship and properties between the three parameters of traffic flow at the road network level. However, in actual application, it is found that highway traffic flow will have problems such as traffic breakdown, capacity drop, hysteresis, oscillation, relaxation, and boomerang effect, which cannot accurately describe the operation status of the highway network. For this reason, we propose a road network status evaluation method based on random forest and natural language model, aiming to provide an evaluation method for the operation status of the highway network based on actual data, through standardized calculation methods and indicator systems.

[0003] In the prior art, "A Regional Expressway Network Operation Status Evaluation Method" (CN104050803A) improves the traditional expressway network temporal weighted index, using only the congestion and vehicle speed indicators, with fewer parameter options. The expressway network temporal weighted index obtained by weighted average of the results of the sub-segments is used for judgment. The method used is relatively basic and fails to explore the connection between the various sections. It only gives the network status results, and does not combine the researchers' professional knowledge to give optimization measures. This study selected 19 traffic indicators with more specific and comprehensive parameters; used the random forest algorithm to evaluate the road network status, and the model itself can explore the connection between the road networks during the training process; combined with traffic indicators and professional knowledge to train the natural language model, it can give corresponding optimization measures while obtaining the road network status. Summary of the invention

[0004] In order to solve the technical problems existing in the prior art to a certain extent, the purpose of the present invention is to provide a solution for quickly identifying the location of a highway accident scene, fully mobilize the role of highway toll flow, give play to the information identification role of highway guardrails, solve the problem of quickly identifying the location of a highway accident scene, and speed up the handling process of highway traffic accidents.

[0005] The purpose of the present invention is achieved by at least one of the following technical solutions.

[0006] A road network status evaluation method based on random forest and natural language model includes the following steps:

[0007] S1. Data collection: Collect basic highway data in advance;

[0008] S2, index calculation: according to the driving conditions of vehicles on the road within a set time period, multiple data indicators are calculated;

[0009] S3, build training set: collect data indicators of each time period, give corresponding road network evaluation results, and form a training set;

[0010] S4. Constructing a random forest model for road network evaluation: using multiple decision trees to form a random forest model, putting the training set into the random forest model for training to obtain a trained random forest model;

[0011] S5. Road network evaluation: Use real-time data to enter the trained random forest model for calculation to obtain the classification result of the road network status description, that is, the road network status evaluation;

[0012] S6. Natural language conversion of evaluation results: The overall status of the road network is obtained based on the classification results of the road network status description. The basic data of expressways, multiple data indicators and the overall status of the road network are input into the trained language model to obtain specific optimization measures.

[0013] Furthermore, in step S1, the basic highway data refers to the total evaluation indicators set to evaluate the static highway road status, including the total length of the road network, the total lane length of the road network, the total travel time of the road network, the longest shortest path, the total nodes of the road network and the total intersections of the road network.

[0014] Total length of road network:

[0015] L i is the length of road segment i in the road network, and n is the total number of road segments;

[0016] Total lane length of road network:

[0017] m i is the number of lanes in section i of the road network;

[0018] Total travel time of the road network:

[0019] T i is the travel time of road section i in the road network according to the design speed;

[0020] The longest shortest path:

[0021] R max =max(R j ),

[0022] R j is the shortest path between any OD (origin-destination) in the road network, j represents the number of the OD pair, and the weight of the road section considered in calculating the shortest path is the total node P of the travel time road network, and the boundary points and intersection points are all nodes;

[0023] The total intersection of the road network is IS, where two road sections meet is an intersection.

[0024] Further, in step S2, the plurality of data indicators include an average evaluation indicator set for evaluating the dynamic traffic flow state and a variance indicator set for evaluating the stability of the traffic flow state;

[0025] The average evaluation indicators set for evaluating dynamic traffic flow status include average road network density, average road network capacity, average peak hour flow of the road network, average flow deviation of the road network, average design speed of the road network, average travel speed of the road network, average speed deviation of the road network, average design accessibility of road network sections, average travel accessibility of road network sections, average accessibility deviation of road network sections and average congestion time period of the road network;

[0026] Average road network density:

[0027] AREA is the area of ​​the road network;

[0028] Average traffic capacity of road network:

[0029] C i is the traffic capacity of road section i;

[0030] Average peak hour traffic volume of the road network:

[0031] Q i-max is the peak hour flow rate of road section i;

[0032] Average flow deviation of road network:

[0033]

[0034] Q i-max is the peak hour flow rate of road section i, C i is the traffic capacity of road section i, and n is the total number of road sections;

[0035] Average design speed of road network:

[0036] Vi-design is the design speed of section i;

[0037] Average travel speed on the road network:

[0038] V i-mean is the average travel speed of road section i;

[0039] Average speed deviation of the road network:

[0040]

[0041] V i-design is the design speed of section i, V i-mean is the average travel speed of road section i;

[0042] Average design accessibility of road network sections:

[0043] A i-design The accessibility of the design speed section;

[0044] V i-design is the design speed of section i, L i is the length of road section i;

[0045] Average trip accessibility of road network segments:

[0046] A i-mean is the accessibility of the road segment with average travel speed;

[0047] V i-mean is the average travel speed of section i, L i is the length of road section i;

[0048] Average accessibility deviation of road network segments:

[0049]

[0050] A i-mean is the accessibility of the average travel speed section, A i-design The accessibility of the design speed section

[0051] V i-mean is the average travel speed of section i, L i is the length of road section i;

[0052] Average congestion time period of the road network:

[0053] When the hourly traffic volume is less than 60% of the capacity, it is congested.

[0054] The variance indicators set to evaluate the stability of traffic flow state include the variance of the ratio of average speed to speed limit of road network sections and the variance of the ratio of average flow to capacity of road network sections;

[0055] Variance of the ratio of average speed to speed limit of road network section:

[0056] is the average speed of the road section, V i-design is the design speed of section i;

[0057] Variance of the ratio of average flow rate to capacity of road network sections:

[0058] Q i-mean is the average flow rate of the road section, C i The traffic capacity of the road section.

[0059] Furthermore, in step S3, the highway basic data and multiple data indicators within the same time period are combined into one vector data, each vector data corresponds to an evaluation result of a road network, and multiple vector data and road network evaluation results constitute a training set.

[0060] Furthermore, in step S4, the random forest model composed of the multiple decision trees is a known random forest model. The total index, average index and variance index include:

[0061] ① Comprehensive basic static information of the road network, which is used to analyze the road network accessibility, the fastest vehicle travel time and the shortest paths of each OD. As the basic indicators of the road network, it is used to compare with the real-time road network operation results to judge the traffic status, and it is also the optimal result in terms of travel time and driving distance;

[0062] ② Dynamic traffic flow information of the road network, mainly flow rate and speed, is used to analyze the travel time and delay time of vehicles in real-time operation, the congested sections and congestion time they pass through when passing through the road, and provide a data basis for vehicle diversion and optimization of vehicle allocation.

[0063] ③ The stability information of road network flow and speed, namely the variance and deviation results, is used to analyze the changes in vehicle speed in different time periods and on different dates. For example, for sections prone to congestion, sections with large speed changes should be properly diverted, and sections with small speed changes should be expanded.

[0064] The set road network status includes six results: fully meeting travel needs, basically meeting travel needs, barely meeting travel needs, slightly below travel demand expectations, lower than travel demand expectations, and lower than travel demand expectations. The overall situation of the road network is subdivided. The first two situations do not require road network regulation, the middle two situations require appropriate diversion and consolidation, or closure of certain highway ramps, and if the latter two situations occur frequently, more active intervention measures will be needed or consideration will be given to certain modifications and expansions to the road network.

[0065] The input data of the random forest algorithm used are 19 indicators in total, including total indicators, average indicators and variance indicators, which fully considers the dynamic and static information of the road network.

[0066] The API used is an open source natural language model interface, which eliminates the step of manual processing of results by professionals and directly provides natural language that can be directly understood by users.

[0067] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0068] (1) The evaluation index system is more complete. The indicators used in this method include static indicators of the road network and dynamic indicators of traffic flow, which describe the operation status of the highway network from multiple angles and aspects.

[0069] (2) The evaluation indicators are more rigorous and scientific. The evaluation indicators used in this method can all be obtained through mathematical calculations and can be completed without human intervention, which is more rigorous and scientific.

[0070] (3) Evaluating the road network from a macro perspective uses a wider range of data, which improves data stability, reduces the required accuracy of the data, and makes the results more reliable.

[0071] (4) The evaluation process is faster and more automated. Statistical data can be calculated in real time and brought into the model for road network evaluation. Evaluation can be completed in multiple time periods and at multiple granularities, facilitating timely road network optimization.

[0072] (5) The natural language model is used to describe the evaluation results more completely and comprehensively, providing scientific and reasonable suggestions for road network optimization. BRIEF DESCRIPTION OF THE DRAWINGS

[0073] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following is an introduction to the drawings of the embodiments of the present invention or the related technical solutions in the prior art:

[0074] Figure 1 It is a flowchart of a road network status evaluation method based on random forest and natural language model in an embodiment of the present invention;

[0075] Figure 2 This is a schematic diagram of the process of selecting indicators for each layer of the decision tree in an embodiment of the present invention;

[0076] Figure 3 is a schematic diagram of a process for providing optimization suggestions for random forest evaluation results in an embodiment of the present invention;

[0077] Figure 4 It is a flowchart of random forest decision result analysis and output in an embodiment of the present invention. DETAILED DESCRIPTION

[0078] The present invention is further described in detail below in conjunction with embodiments and drawings, examples of which are shown in the drawings.

[0079] Example:

[0080] A road network status evaluation method based on random forest and natural language model, such as Figure 1 As shown, the following steps are included:

[0081] S1. Data collection: Collect basic highway data in advance;

[0082] The basic data of expressways refers to the total evaluation indicators set to evaluate the static road status of expressways, including the total length of the road network, the total lane length of the road network, the total travel time of the road network, the longest shortest path, the total nodes of the road network and the total intersections of the road network.

[0083] Total length of road network:

[0084] L i is the length of road segment i in the road network, and n is the total number of road segments;

[0085] Total lane length of road network:

[0086] m i is the number of lanes in section i of the road network;

[0087] Total travel time of the road network:

[0088] T i is the travel time of road section i in the road network according to the design speed;

[0089] The longest shortest path:

[0090] R max =max(R j ),

[0091] R jis the shortest path between any OD (origin-destination) in the road network, j represents the number of the OD pair, and the weight of the road section considered in calculating the shortest path is the total node P of the travel time road network, and the boundary points and intersection points are all nodes;

[0092] The total intersection of the road network is IS, where two road sections meet is an intersection.

[0093] S2, index calculation: according to the driving conditions of vehicles on the road within a set time period, multiple data indicators are calculated;

[0094] The multiple data indicators include an average evaluation indicator set for evaluating the dynamic traffic flow state and a variance indicator set for evaluating the stability of the traffic flow state;

[0095] The average evaluation indicators set for evaluating dynamic traffic flow status include average road network density, average road network capacity, average peak hour flow of the road network, average flow deviation of the road network, average design speed of the road network, average travel speed of the road network, average speed deviation of the road network, average design accessibility of road network sections, average travel accessibility of road network sections, average accessibility deviation of road network sections and average congestion time period of the road network;

[0096] Average road network density:

[0097] AREA is the area of ​​the road network;

[0098] Average traffic capacity of road network:

[0099] C i is the traffic capacity of road section i;

[0100] Average peak hour traffic volume of the road network:

[0101] Q i-max is the peak hour flow rate of road section i;

[0102] Average flow deviation of road network:

[0103]

[0104] Q i-max is the peak hour flow rate of road section i, C i is the traffic capacity of road section i, and n is the total number of road sections;

[0105] Average design speed of road network:

[0106] V i-design is the design speed of section i;

[0107] Average travel speed on the road network:

[0108] V i-mean is the average travel speed of section i;

[0109] Average speed deviation of the road network:

[0110]

[0111] V i-design is the design speed of section i, V i-mean is the average travel speed of section i;

[0112] Average design accessibility of road network sections:

[0113] A i-design The accessibility of the design speed section;

[0114] V i-design is the design speed of section i, L i is the length of road section i;

[0115] Average trip accessibility of road network segments:

[0116] A i-mean is the accessibility of the road segment with average travel speed;

[0117] V i-mean is the average travel speed of section i, L i is the length of road section i;

[0118] Average accessibility deviation of road network segments:

[0119]

[0120] A i-mean is the accessibility of the average travel speed section, A i-design The accessibility of the design speed section

[0121] V i-mean is the average travel speed of section i, L i is the length of road section i;

[0122] Average congestion time period of the road network:

[0123] When the hourly traffic volume is less than 60% of the capacity, it is congested.

[0124] The variance indicators set to evaluate the stability of traffic flow state include the variance of the ratio of average speed to speed limit of road network sections and the variance of the ratio of average flow to capacity of road network sections;

[0125] Variance of the ratio of average speed to speed limit of road network section:

[0126] V i-mean is the average speed of the road section, V i-design is the design speed of section i;

[0127] Variance of the ratio of average flow rate to capacity of road network sections:

[0128] Q i-mean is the average flow rate of the road section, C i The traffic capacity of the road section.

[0129] S3, build training set: collect data indicators of each time period, give corresponding road network evaluation results, and form a training set;

[0130] The basic data of expressways and multiple data indicators within the same time period are combined into a vector data. Each vector data corresponds to an evaluation result of a road network. Multiple vector data and road network evaluation results constitute a training set.

[0131] In one embodiment, a classification and scoring method for static highway road state evaluation index, dynamic traffic flow state evaluation index and traffic flow state stability variance index is provided, so as to obtain the traffic operation state classification of the entire road network. The evaluation index classification is as follows:

[0132] Table 1 Classification of evaluation indicators

[0133]

[0134] S4. Constructing a random forest model for road network evaluation: using multiple decision trees to form a random forest model, putting the training set into the random forest model for training to obtain a trained random forest model;

[0135] In one embodiment, the random forest model composed of the multiple decision trees is an existing known random forest model Random Forests.

[0136] S5. Road network evaluation: Use real-time data to enter the trained random forest model for calculation to obtain the classification result of the road network status description, that is, the road network status evaluation;

[0137] In one embodiment, the road network status evaluation results are classified as follows:

[0138] Table 2 Classification of road network status evaluation results

[0139]

[0140] The evaluation results of the road network status will have different evaluation standards for different road networks in different regions, so it is impossible to generalize. In one embodiment, the method used is the expert voting method. After obtaining the data indicators, relevant traffic research experts are asked to evaluate the data and road network, and classify the road network status into the status that meets the above six results from a professional perspective through evaluation, which is used as the training data set for this study. Fitting the expert's experience evaluation model into the model to reduce the labor cost of subsequent road network status evaluation is also a goal of this study.

[0141] Among them, the total evaluation index needs to be obtained by investigating the highway network in advance; the average evaluation index is mainly obtained through analysis of historical survey data of the road network, and it is necessary to obtain historical data of the road network in advance (it is recommended to integrate data from the past three months and the same period in history), and obtain specific indicators through calculation. The indicator results will be updated on a monthly basis; the variance indicator is obtained in a similar way to the average indicator.

[0142] If it is necessary to conduct real-time evaluation of the road network in a certain period of time, the statistical range of the average evaluation index and the variance index should be set in this period of time.

[0143] The random forest algorithm needs to be trained in advance based on historical results. When evaluating different road networks, the results of the road network need to be used to adjust parameters. The input data is the above three indicators, and the output result is the status of the road network, including six results: fully meeting travel needs, basically meeting travel needs, barely meeting travel needs, slightly lower than travel demand expectations, lower than travel demand expectations, and lower than travel demand expectations.

[0144] In one embodiment, the calculated basic indicators are put into the random forest classification method, and the judgment criteria are shown in Table 1.

[0145] The random forest model is trained based on the qualitative results of the data indicators, and the initial scores of A, B, C, D, E, and F are 10, 20, 30, 40, and 50, without considering the weight of each of the 19 indicators.

[0146] Divide the index into x1,x2,x3,…,x 19Here, each layer is set to select 5 indicators for classification and judgment. Each layer of the decision tree is guaranteed to have an "average indicator" (an indicator in the average indicator category given in step S2). The node values ​​can be divided into four types: 80, 130, 180, and 230 according to the weighted results. If there is an average indicator among the 5 indicators, one of the average indicators is selected as the significant indicator. The results of the significant indicators are 10, 20, 30, 40, and 50, and 25 and 35 are selected as the node values. Each time a decision tree is established, a node value is selected from the two as the judgment node value of the decision tree. Each layer will only set three judgment results, which are:

[0147] ①The weighted result and average index are both less than the node value;

[0148] ②The weighted result and average index are both less than the node value;

[0149] ③Others.

[0150] The initial setting is 10 layers of forest. All indicators of each layer (indicators used to calculate weighted results) are randomly selected. After the indicator selection is completed, it will be determined whether there are indicators that have not been selected in all previous layers. If so, one of the indicators will be randomly replaced and it will be determined whether the whole has an average indicator. The new indicator judgment and the average indicator judgment are repeated until the selected indicator meets the conditions and the next step of indicator result weighting and classification is performed. Figure 2 The working principle of decision trees in random forests is as follows Figure 3 As shown in Table 2, Value11 represents the weighted result value of the first layer, Value12 represents the average index value of the first layer, Value21 represents the weighted result value of the second layer, Value12 represents the average index value of the second layer, and so on. Each decision tree is a ternary tree with a total of ten layers. The last layer uses the road network status as a node, as shown in Table 2.

[0151] The training process of random forest is to replace the indicators of each layer on the basis of the random forest in which the indicators are initially selected, and observe whether the results can better adapt to the results of the training set. After reaching a certain accuracy, the test set is used for testing. If the accuracy of the test set results also meets the requirements, it means that the indicators selected by each layer of the random forest meet the overall judgment requirements and can be used for real-time judgment.

[0152] S6. Natural language conversion of evaluation results: The overall status of the road network is obtained based on the classification results of the road network status description. The basic data of expressways, multiple data indicators and the overall status of the road network are input into the trained language model to obtain specific optimization measures.

[0153] In one embodiment, the natural language model mainly calls the ChatGPT API for fine-tuning training. The input data is the above three indicators and the output results of the random forest algorithm. The output results are natural language descriptions of the input data, which help people better understand the intention and give optimization suggestions for the current road network. The training process of the natural language model is as follows:

[0154] ① Build a natural language model training database. Different from the traditional machine learning training database, when training a natural language model, only fine-tuning is used (Fine-tuning refers to further training based on a pre-trained large language model using a specific data set to adapt the model to a specific task or field. Fine-tuning is similar to the training process of machine learning, but it is based on the existing natural language model. During the fine-tuning process, there is no need to change too many parameters of the natural language model, so it is more efficient than training a natural language model from scratch). Therefore, there is no need to distinguish between training sets, test sets, and prediction sets. You only need to give a list of training data in a standard format. This study recommends using the JSON format, such as ['in':[],'connect word':'I want','out':[]], where 'in' is a given evaluation indicator, 'out' is a given road network state for fine-tuning, and 'connect word' is a connecting word used to convert the data set into prompt data for "dialogue" with the natural language model during the fine-tuning process in Python.

[0155] ② Fine-tune the Natural Large Language Model API to meet the requirements of natural language of the road network status evaluation results of this study. This study uses OpenAI's ChatGPT API, which is called by python using the https protocol.

[0156] ③Save the fine-tuned model state. After the fine-tuning is completed, save the model state after the user's fine-tuning to facilitate subsequent application calls.

[0157] In one embodiment, the natural language model is fine-tuned in the form of a dialogue, and the content of the dialogue includes the current status of various indicators of the road network and suggested solutions given after analysis and processing by professionals. For example, when the peak hour traffic volume is high, the average travel speed score of the road network is low, and the overall speed is slightly lower than the travel demand, some of the roads are diverted to improve the scores in terms of traffic volume and speed, thereby obtaining a road network evaluation result that basically meets the travel demand.

[0158] The fine-tuning natural language model is an operational model that is fine-tuned using the natural language model of the language example ChatGPT. It is an Internet open source model. The training process is similar to using ChatGPT, except that the communication process uses code to frequently access open interfaces to obtain the corresponding logic of ChatGPT in the current session. For example, sentences such as "I give the vector [80,1500,400,…], and hope you can give the description of 'the average speed of the road section is 80km / h, the average hourly flow rate is 1500pcu / h, the section length is 400m,…, and the road network status fully meets travel needs'", "I give the vector [40,800,300,…], and hope you can give the description of 'the average speed of the road section is 40km / h, the average hourly flow rate is 800pcu / h, the section length is 300m,…, and the road network status is lower than the travel demand expectation'" are used as corpus, and the natural language model interface is frequently accessed to generate such operation logic. When the operator gives the vector [80,1500,400,…], the result of "the average speed of the road section is 80km / h, the average hourly flow rate is 1500pcu / h, the section length is 400m,…, and the road network status fully meets travel needs" can be directly replied, and then displayed directly to the user using code.

[0159] Furthermore, we can make assumptions about some of the static indicators to obtain better adjustment plans. For example, when the road network status is slightly lower than the travel demand, we can increase the average design speed of the road network and bring the re-assigned scores into the random forest model to obtain the final road network status evaluation. If the road network status evaluation results are improved, it means that the method is effective and the natural language model can be used to give corresponding suggestions, such as Figure 4 shown.

[0160] In summary, the method of this embodiment has the following advantages and beneficial effects compared with the prior art:

[0161] (1) The evaluation index system is more complete. The indicators used in this method include static indicators of the road network and dynamic indicators of traffic flow, which describe the operation status of the highway network from multiple angles and aspects.

[0162] (2) The evaluation indicators are more rigorous and scientific. The evaluation indicators used in this method can all be obtained through mathematical calculations and can be completed without human intervention, which is more rigorous and scientific.

[0163] (3) Evaluating the road network from a macro perspective uses a wider range of data, which improves data stability, reduces the required accuracy of the data, and makes the results more reliable.

[0164] (4) The evaluation process is faster and more automated. Statistical data can be calculated in real time and brought into the model for road network evaluation. Evaluation can be completed in multiple time periods and at multiple granularities, facilitating timely road network optimization.

[0165] (5) The natural language model is used to describe the evaluation results more completely and comprehensively, providing scientific and reasonable suggestions for road network optimization.

[0166] The above is only a preferred embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical solution and invention concept of the present invention within the scope disclosed by the present invention, which shall fall within the protection scope of the present invention.

Claims

1. A road network status evaluation method based on random forest and natural language model, characterized in that: The following steps are involved: S1. Data collection: Collect basic highway data in advance; S2, index calculation: according to the driving conditions of vehicles on the road within a set time period, multiple data indicators are calculated; S3, build training set: collect data indicators of each time period, give corresponding road network evaluation results, and form a training set; S4. Constructing a random forest model for road network evaluation: using multiple decision trees to form a random forest model, putting the training set into the random forest model for training to obtain a trained random forest model; S5. Road network evaluation: Use real-time data to substitute into the trained random forest model for calculation to obtain the classification result of the road network status description, i.e., the road network status evaluation; S6. Natural language conversion of evaluation results: The overall status of the road network is obtained based on the classification results of the road network status description. The basic data of expressways, multiple data indicators and the overall status of the road network are input into the trained language model to obtain specific optimization measures.

2. A road network status evaluation method based on random forest and natural language model according to claim 1, characterized in that: In step S1, the basic highway data refers to the total evaluation index set to evaluate the static highway road status, including the total length of the road network, the total lane length of the road network, the total travel time of the road network, the longest shortest path, the total nodes of the road network and the total intersections of the road network.

3. A road network status evaluation method based on random forest and natural language model according to claim 2, characterized in that: The total length of the road network is as follows: L i is the length of road segment i in the road network, and n is the total number of road segments; The total lane length of the road network is as follows: m i is the number of lanes in section i of the road network; The total travel time of the road network is as follows: T i is the travel time of road section i in the road network according to the design speed; The longest shortest path is as follows: R max =max(R j ), R j is the shortest path between any OD (origin-destination) in the road network, j represents the number of the OD pair, and the weight of the road section considered in calculating the shortest path is the total node P of the travel time road network, and the boundary points and intersection points are all nodes; The intersection of two road sections is an intersection, and the total intersection IS of the road network is obtained.

4. A road network status evaluation method based on random forest and natural language model according to claim 1, characterized in that: In step S2, the plurality of data indicators include an average evaluation indicator set for evaluating a dynamic traffic flow state and a variance indicator set for evaluating a traffic flow state stability.

5. A road network status evaluation method based on random forest and natural language model according to claim 4, characterized in that: The average evaluation indicators set to evaluate the dynamic traffic flow status include average road network density, average road network capacity, average peak hour flow rate of the road network, average flow rate deviation of the road network, average design speed of the road network, average travel speed of the road network, average speed deviation of the road network, average design accessibility of road network sections, average travel accessibility of road network sections, average accessibility deviation of road network sections and average congestion time period of the road network.

6. A road network status evaluation method based on random forest and natural language model according to claim 5, characterized in that: The average road network density is as follows: AREA is the area of ​​the road network; The average capacity of the road network is as follows: C i is the traffic capacity of road section i; The average peak hour traffic volume of the road network is as follows: Q i-max is the peak hour flow rate of road section i; The average flow deviation of the road network is as follows: Q i-max is the peak hour flow rate of road section i, C i is the traffic capacity of road section i, and n is the total number of road sections; The average design speed of the road network is as follows: V i-design is the design speed of section i; The average travel speeds for the road network are as follows: V i-mean is the average travel speed of road section i; The average speed deviation of the road network is as follows: V i-design is the design speed of section i, V i-mean is the average travel speed of road section i; The average design accessibility of road network sections is as follows: A i-design The accessibility of the design speed section; V i-design is the design speed of section i, L i is the length of road section i; The average trip accessibility of the road network segments is as follows: A i-mean is the accessibility of the road segment with average travel speed; V i-mean is the average travel speed of section i, L i is the length of road section i; The average accessibility deviation of the road network segments is as follows: A i-mean is the accessibility of the average travel speed section, A i-design The accessibility of the design speed section V i-mean is the average travel speed of section i, L i is the length of road section i; The average congestion time periods of the road network are as follows: When the hourly traffic volume is less than 60% of the capacity, it is congested.

7. A road network status evaluation method based on random forest and natural language model according to claim 4, characterized in that: The variance indicators set to evaluate the stability of traffic flow state include the variance of the ratio of average speed to speed limit of road network sections and the variance of the ratio of average flow to capacity of road network sections.

8. A road network status evaluation method based on random forest and natural language model according to claim 7, characterized in that: The variance of the ratio of average speed to speed limit of road network sections is as follows: V i-mean is the average speed of the road section, V i-design is the design speed of section i; The variance of the ratio of average flow to capacity of road network sections is as follows: Q i-mean is the average flow rate of the road section, C i The traffic capacity of the road section.

9. The method for evaluating road network status based on random forest and natural language model according to claim 1, characterized in that: In step S3, the basic data of the expressway and multiple data indicators in the same time period are combined into a vector data, each vector data corresponds to an evaluation result of a road network, and multiple vector data and road network evaluation results constitute a training set.

10. The road network status evaluation method based on random forest and natural language model according to claim 1, characterized in that: In step S4, the random forest model composed of the multiple decision trees is an existing known random forest model Random Forests.

Images