CN106373390B - Traffic state evaluation method based on Adaptive Neuro-fuzzy Inference - Google Patents
Traffic state evaluation method based on Adaptive Neuro-fuzzy Inference Download PDFInfo
- Publication number
- CN106373390B CN106373390B CN201510438629.9A CN201510438629A CN106373390B CN 106373390 B CN106373390 B CN 106373390B CN 201510438629 A CN201510438629 A CN 201510438629A CN 106373390 B CN106373390 B CN 106373390B
- Authority
- CN
- China
- Prior art keywords
- fuzzy inference
- layer
- inference system
- output
- adaptive
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000011156 evaluation Methods 0.000 title claims abstract description 46
- 230000003044 adaptive effect Effects 0.000 title claims abstract description 45
- 238000012549 training Methods 0.000 claims abstract description 73
- 238000012360 testing method Methods 0.000 claims abstract description 43
- 230000001537 neural effect Effects 0.000 claims abstract description 37
- 230000006870 function Effects 0.000 claims description 54
- 238000000034 method Methods 0.000 claims description 37
- 230000005284 excitation Effects 0.000 claims description 13
- 230000002776 aggregation Effects 0.000 claims description 11
- 238000004220 aggregation Methods 0.000 claims description 11
- 238000004364 calculation method Methods 0.000 claims description 8
- 230000010354 integration Effects 0.000 claims description 6
- 238000006116 polymerization reaction Methods 0.000 claims description 6
- 230000008569 process Effects 0.000 description 14
- 238000013528 artificial neural network Methods 0.000 description 12
- 238000013210 evaluation model Methods 0.000 description 12
- 239000011159 matrix material Substances 0.000 description 9
- 238000010586 diagram Methods 0.000 description 6
- 238000013507 mapping Methods 0.000 description 5
- 239000013598 vector Substances 0.000 description 5
- 230000004044 response Effects 0.000 description 4
- 238000002945 steepest descent method Methods 0.000 description 4
- 238000004458 analytical method Methods 0.000 description 3
- 230000008859 change Effects 0.000 description 3
- 238000002474 experimental method Methods 0.000 description 3
- 238000011160 research Methods 0.000 description 3
- 238000007405 data analysis Methods 0.000 description 2
- 230000007547 defect Effects 0.000 description 2
- 238000001514 detection method Methods 0.000 description 2
- 230000003993 interaction Effects 0.000 description 2
- 230000007246 mechanism Effects 0.000 description 2
- 238000012545 processing Methods 0.000 description 2
- 230000007704 transition Effects 0.000 description 2
- 206010039203 Road traffic accident Diseases 0.000 description 1
- 230000004931 aggregating effect Effects 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 230000001149 cognitive effect Effects 0.000 description 1
- 238000004891 communication Methods 0.000 description 1
- 238000010835 comparative analysis Methods 0.000 description 1
- 239000002131 composite material Substances 0.000 description 1
- 230000007423 decrease Effects 0.000 description 1
- 230000003247 decreasing effect Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000018109 developmental process Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000011478 gradient descent method Methods 0.000 description 1
- 238000003384 imaging method Methods 0.000 description 1
- 238000012886 linear function Methods 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 210000002569 neuron Anatomy 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 230000008447 perception Effects 0.000 description 1
- 230000000644 propagated effect Effects 0.000 description 1
- 238000011158 quantitative evaluation Methods 0.000 description 1
- 230000001953 sensory effect Effects 0.000 description 1
- 230000033772 system development Effects 0.000 description 1
Landscapes
- Traffic Control Systems (AREA)
Abstract
The traffic state evaluation method based on Adaptive Neuro-fuzzy Inference that the present invention relates to a kind of, including training stage and test phase;Wherein, the training stage includes:Read the historical data of input variable and output variable;Historical data is input in Adaptive Neuro-fuzzy Inference, the output result of Adaptive Neuro-fuzzy Inference is obtained;Parameter learning rule based on anti-pass thought, the output result of combining adaptive neural fuzzy inference system are adjusted parameter, obtain the Adaptive Neuro-fuzzy Inference after parameter adjustment;The Minimum Mean Square Error of Adaptive Neuro-fuzzy Inference after calculating parameter adjustment, judges whether the value reaches a specified threshold, if reached, it is meant that the training stage terminates;Test phase includes:Read test data;Test data is input in the Adaptive Neuro-fuzzy Inference trained via the training stage, the value of the service level for describing road traffic state is obtained.
Description
Technical Field
The invention relates to the field of road monitoring, in particular to a road traffic state evaluation method based on a self-adaptive neural fuzzy inference system.
Background
With the development of economy and the popularization of automobiles, traffic congestion has become a common problem in large and medium-sized cities. The evaluation of the traffic state of the road is a precondition for relieving traffic congestion and making a quick response to traffic accidents.
In the prior art, most of research results on the evaluation of the running condition of a traffic system are directed to the research of an expressway, and the research results are relatively few in the aspect of urban roads. The road traffic capacity manual is relatively authoritative in the aspect of performance evaluation of a traffic system, the manual discusses the evaluation of urban road traffic conditions (including intersections and main roads), but the acquisition method of performance indexes is estimated by a formula instead of on-site measurement, and meanwhile, the road traffic capacity manual is an estimation model for establishing the performance indexes according to the ideal state of American roads and is not suitable for the national conditions of China. In reference 1, "comparative analysis of intersection traffic capacity calculation method", road traffic technology, 2006(5), 23-29 ", an existing evaluation model is modified for the road condition in China, and the practicability, accuracy, objectivity and applicability of the improved model need to be further proved. Jungkeun Yoon and Brian Noble In reference 2, "surface traffic evaluation", In Proceedings of the 5th Annual ACM/USENIXConreference on Mobile Systems, Applications, and Services (Mobile), San Juan, PR.June 2007, p.220-232, proposed a new evaluation method for urban road traffic conditions, using a vehicle equipped with a GPS device with two-way communication capability to obtain urban roadsThe traffic condition information (mainly speed information) of a road is used for evaluating the traffic condition of a certain road, but the traffic condition of a single intersection and an urban road network cannot be evaluated. Zhao Ming, Hou faithful In reference 3 "data driven signalized intersection traffic assessment", In Proceedings of the 27thIn the Chinese Control Conference, July 16-18,1008, Kunming, China, p.559-563', from the perspective of a traffic system manager and a user, two performance indexes, namely a system average speed and a service level index, are provided for evaluating the traffic condition of an intersection, but the two performance indexes cannot be fully explained and utilized from the perspective of the manager and the user, and only the intersection is evaluated when the service level index is calculated, and an evaluation method for a road network is not clearly provided. In addition, the shenghui performs performance index evaluation of urban road traffic conditions for three aspects of points, lines, planes and the like in the urban road traffic evaluation method and evaluation system development of the master academic thesis of shenghui aiming at a traffic network consisting of a large number of intersections, and provides a new idea for comprehensive evaluation of the traffic network.
The prior art has the following defects: in an urban road traffic system, as a qualitative evaluation method, a service Level (LOS) describes the overall operation condition of a certain road traffic system and the feeling of a driver and passengers on the certain road traffic system, and is an important evaluation index of the urban road traffic system. Many of the existing LOS evaluation methods perform quantitative evaluation based on a single index, and evaluate the service level that can be provided by the traffic infrastructure by comparing the relationship between the index and the LOS level. The methods do not consider the qualitative aspect of the evaluation problem, and cannot fully reflect the subjectivity and integrity of the evaluation of the traffic system, and the establishment of a multi-index comprehensive evaluation model is an important subject of the LOS evaluation of the road traffic.
Disclosure of Invention
The invention aims to overcome the defect that the existing qualitative evaluation method for urban road traffic state is limited in accuracy degree, thereby providing a road traffic state evaluation method with high accuracy and low test error value.
In order to achieve the aim, the invention provides a road traffic state evaluation method based on a self-adaptive neural fuzzy inference system, which comprises a training stage and a testing stage; wherein,
the training phase comprises the steps of:
step 101), reading historical data of input variables and output variables; the input variables include flow, occupancy, and speed, and the output variables include service level values;
step 102), inputting the historical data input in the step 101) into the self-adaptive neural fuzzy inference system to obtain an output result of the self-adaptive neural fuzzy inference system; adjusting parameters of the adaptive neural fuzzy inference system by combining an output result of the adaptive neural fuzzy inference system based on a parameter learning rule of a back propagation thought to obtain the adaptive neural fuzzy inference system after parameter adjustment; wherein,
the parameters of the self-adaptive neural fuzzy inference system comprise a precondition parameter, a conclusion parameter, a weight parameter set for an input variable and a weight parameter set for a rule; the precondition parameters are determined by the shape of a membership function of the inference antecedent; the conclusion parameters are determined by the membership function shape of the inferred piece;
step 103), calculating the minimum mean square error of the adaptive neural fuzzy inference system after the parameters are adjusted, which is obtained in step 102), and judging whether the value reaches a specified threshold value, if so, the training stage is ended; otherwise, executing the next step;
step 104), reading a new set of historical data about input variables and output variables, and then re-executing step 102);
the testing phase comprises the following steps:
step 201), reading test data, wherein the test data comprises: historical data of flow, occupancy and speed;
step 202), inputting the test data into the adaptive neural fuzzy inference system obtained through training in the training phase to obtain a value of the service level for describing the road traffic state.
In the above technical solution, the adaptive neuro-fuzzy inference system includes five layers, which are respectively:
a first layer: a fuzzification layer for fuzzifying the precision input;
O1,i=μAi(x) I is 1,2 or
Wherein, Oj,iI-th node output, μ, representing the j-th layerAi、The membership functions of the front parts are defined by generalized bell-shaped MF:
wherein a isi、bi、ciAs a precondition parameter;
a second layer, inference layer, for computing the excitation strength w of each rulei;
Wherein i ═1、2,μiThe weight of each of the inputs is represented,
the third layer, the hidden layer, is used for calculating the effective back-part MF of each rule;
wherein i is 1 or 2; ciA post-reasoning part;representing an implicit operator;
the fourth layer is a polymerization layer and is used for calculating the sum of all the rule effective back pieces MF;
wherein i is 1 or 2; tau isiA weight representing each rule; aggregation operatorAdopting choquet integration;
the fifth layer is a de-fuzzy layer and is used for calculating the accurate output of the system;
wherein D represents a deblurring operator, and the calculation of the deblurring operator is realized by adopting a central deblurring method.
In the above technical solution, the parameters of the adaptive neuro-fuzzy inference system are adjusted according to the following formula:
Δwji=η(di-xi)·xj·X
wherein, Δ wijRepresenting the increment of any parameter in the adaptive neuro-fuzzy inference system η is the learning step size diIs the desired output, x, of node iiIs the actual output of node i, xjIs an input to node i, which is a node of a layer immediately above node j, j<i; x is a polynomial of formula (X)i×(1-xi) ) is expressed.
In the technical scheme, the adaptive neural fuzzy inference system replaces an AND operator OR an OR operator with an OWA operator at an inference layer, AND calculates the excitation intensity; polymerization is achieved by choquet integration in the polymerization layer; the defuzzification operator adopts a central defuzzification method; and setting the weight of each input by muiRepresents; setting the weight of each rule by tauiAnd (4) showing.
The invention has the advantages that:
compared with the existing method, the road traffic state evaluation method has high accuracy and low test error value.
Drawings
FIG. 1 is a schematic diagram of the inference process of a fuzzy inference system based on Choquet integral-OWA;
FIG. 2 is a schematic diagram of an example of an adaptive neuro-fuzzy inference system;
FIG. 3 is a schematic diagram of an Agg-ANFIS based service level evaluation model established by the present invention;
FIG. 4 is a schematic diagram of a Sugeno-FIS output graph;
FIG. 5 is a schematic diagram of the error between the actual output and the expected output of the Sugeno-FIS fuzzy inference system;
FIG. 6 is a training data error plot for ANFIS;
FIG. 7 is a graph comparing expected output values of test data with actual output values of Sugeno-FIS after training;
FIG. 8(a) shows a membership function for flow in an input variable;
FIG. 8(b) is a membership function representing occupancy in input flow;
FIG. 8(c) is a membership function of velocity in input variables;
FIG. 8(d) shows a membership function for the output variable LOS;
FIG. 9 is a graph of AggFIS output;
FIG. 10 is a graph of error curves for actual and expected output of AggFIS;
FIG. 11 is a graph of the training error for Agg-ANFIS;
FIG. 12 is a graph comparing output values of trained AggFIS to expected output of test data;
FIG. 13 is a graph of the test error for AggFIS;
FIG. 14(a) shows a membership function for flow in an input variable;
FIG. 14(b) is a membership function showing occupancy in input flow;
FIG. 14(c) is a membership function of velocity in input variables;
fig. 14(d) shows a membership function of the output variable LOS.
Detailed Description
Since many concepts are involved in the present invention, a description will be first made of these concepts.
Traffic flow: the number of traffic entities passing through a certain place, a certain section or a certain lane of a road in a selected time period is determined;
road occupancy: the proportion of the current traffic flow on the road to the designed traffic volume is also called road utilization rate.
A fuzzy inference system: fuzzy inference refers to the process of reasoning, also called approximate reasoning, to draw a possible inaccurate conclusion from a set of inaccurate preconditions. The fuzzy inference system is based on fuzzy set theory and fuzzy inference method, and has the capability of processing fuzzy information. The system takes the fuzzy logic theory as a main calculation tool, can realize complex nonlinear mapping relation, and the input and the output of the system are accurate numerical values.
A neural network: a neural network is an operational model, which is formed by connecting a large number of nodes (or neurons). Each node represents a particular output function, called the excitation function. Each connection between two nodes represents a weighted value, called weight, for the signal passing through the connection. The output of the neural network varies depending on the connection mode, weight value, and excitation function of the network.
The self-adaptive neural fuzzy inference system: the adaptive neuro-fuzzy inference system is a combination of a fuzzy inference system and a neural network. By introducing the learning mechanism of the neural network into the fuzzy inference system, an adaptive system with human sensory and cognitive components is formed. The neural network is directly embedded in a whole fuzzy structure, and learns the training data unconsciously, automatically generates, corrects and highly summarizes the optimal membership function and fuzzy rule of the input and output variables; on the other hand, the structures and parameters of the layers of the neural network have definite and easily understood physical meanings.
The foregoing is a description of the concepts involved in the process of the present invention, which is further described below.
The road traffic state evaluation method takes the speed, the flow and the occupancy rate which can be detected on the road as the input values of the self-adaptive neural fuzzy inference system, and outputs a traffic service level value (LOS) after the processing of the self-adaptive neural fuzzy inference system.
For ease of understanding, the adaptive neuro-fuzzy inference system involved in the method of the present invention will first be described in detail.
The adaptive neural fuzzy inference system (Agg-ANFIS for short) related in the method is obtained by combining a Choquet integral-OWA-based fuzzy inference system (aggfs for short) and a feedforward neural network according to a basic topological structure of a fuzzy neural network, and the aggfs and the Agg-ANFIS are sequentially explained below.
Fuzzy inference system based on Choquet integral-OWA
Most fuzzy inference systems can be divided into the following three broad categories, according to the type of fuzzy inference and the expressive form of the fuzzy if-then rule: tsukamoto fuzzy models, Mamdani fuzzy models, and Sugeno fuzzy models.
For the whole operation process of any fuzzy inference system, we must determine a function for each of the following fuzzy inference operators:
1) AND operator (AND operator): for calculating the excitation strength of the and connection predecessor rule, usually in the T-paradigm;
2) OR operator (OR operator): for calculating the excitation strength of the connection predecessor rule by or, usually the T-convention;
3) implicit operator (Implication operator): MF, usually the T-norm, for calculating the valid artifacts of each rule according to a given excitation strength;
4) aggregation operator (aggregation operator): for aggregating all available artifacts MF, resulting in a composite output MF, usually in the T-co-norm;
5) defuzzification operator (Defuzzification operator): for converting the total output MF into an accurate output value.
In order to solve the problems of universality of fuzzy inference operators and expression of important factors in an inference process, deep analysis is performed on various inference operators in the application. The OWA operator can be understood as the universal expression of an AND operator OR an OR operator, the Choquet integral calculation in the fuzzy integral is essentially an aggregation operator with calculation continuity, AND the two operators can solve the universal problem of the fuzzy inference operator; meanwhile, the selection of different weights in the OWA operator expresses the complex interaction relationship between the objects (indexes); the interaction relationship between two objects (indexes) is described by the Choquet integral calculation mechanism. As can be seen, both the OWA operator and the Choquet integral can implement fuzzy inference considering weighting factors.
Based on the analysis, the application provides a fuzzy inference system based on Choquet integral-OWA, wherein the system replaces an AND operator OR an OR operator with an OWA operator at an inference layer (inference layer) to calculate excitation intensity; aggregation is realized by using choquet integration at an aggregation layer (aggregation layer), instead of the traditional T-covariance operators (Max or Sum); the defuzzification operator adopts a central defuzzification method (COA); and setting the weight of each input by muiRepresents; setting the weight of each rule by tauiRepresents; thereby realizing the whole fuzzy reasoning process.
The inference process of the Choquet integral-OWA based fuzzy inference system is depicted in fig. 1.
The fuzzy rule form of the fuzzy inference system is as follows:
Ri:IF V1is A1and V2is B1and V3is C1,THEN U is D1;
V1、V2、V3is an input variable, U is a single output variable; a. the1、B1、C1Respectively representing fuzzy sets of each input variable; d1Is a fuzzy set of output variables.
The parameters of the fuzzy inference system described in fig. 1 are defined as follows:
d: a membership function module (membership neural module) of the rule front-part;
D-1: a membership function module (inverse membership neural module) of the rule back-part;
d (x): the excitation strength of each rule;
Ri: a fuzzy rule module (rule neural module);
m-module: a weight module of the aggregation layer;
τi: the weight of each rule;
ui: a weight for each input;
[ai,bi]: the threshold value of each rule valid output.
Adaptive neural fuzzy inference system based on Choquet integral-OWA
The adaptive neural fuzzy inference system based on Choquet integral-OWA is a product of combination of AggFIS and a feedforward neural network according to a basic topological structure of a fuzzy neural network. The adaptive neural fuzzy inference system integrates all inference processes of AggFIS, so that the system has inference operator universality and can realize index importance factor expression; meanwhile, due to the learning function of the neural network, all parameters in the system can be adjusted according to the learning rule, so that the system has adaptability to data.
The adaptive neuro-fuzzy inference system based on Choquet integral-OWA comprises 5 layers, fig. 2 is an example of the adaptive neuro-fuzzy inference system, the system in the embodiment only comprises two inputs and a single output, and the number of the inputs and the outputs of the system can be expanded in practical use;
wherein the output result of each layer is represented as follows:
layer 1: fuzzy determination Layer
The job to be done in this layer is to blur the exact input:
O1,i=μAi(x),i=1,2or
wherein, Oj,iI-th node output, μ, representing the j-th layerAi,For the membership functions of the front parts, these two membership functions of the front parts can be defined by a generalized bell shape MF:
wherein a isi、bi、ciAs a precondition parameter.
Layer 2 inference Layer or rule Layer
The task to be done in this layer is to calculate the excitation strength w of each rule by using a special case (AND) of the OWA operatori。
Wherein muiRepresenting the weight of each input.
Wherein,
layer 3: imaging Layer (hidden Layer)
The work to be done at this level is to compute the effective back-part MF for each rule.
Conclusion parameter set by inference back-piece (C)i) Determining the shape of the membership function;representing a hidden operator (product).
Layer 4: aggregation Layer
The work to be done at this level is to sum up all rule valid artifacts MF.
Wherein the weight of each rule is defined by tauiAnd (4) defining.
Aggregation operatorChoquet integration was used.
Layer 5 defuzzification Layer
The work to be done in this layer is the accurate output of the computing system, and the result obtained after deblurring is recorded as O5:
Wherein D represents a deblurring operator, and the calculation can be realized by adopting a center deblurring method (COA).
The parameters to be adjusted in the adaptive neural fuzzy inference system are as follows:
1) set of precondition parameters { ai,bi,ci}: the method is determined by the shape of a membership function of a piece before reasoning;
2) conclusion parameter set: the shape of the membership function of the inferred piece is determined;
3) weight of each input: by muiRepresents;
4) weight of each rule: by τiAnd (4) showing.
Parameter learning rule based on back propagation thought and applicable to adaptive neural fuzzy inference system
An adaptive network is a network structure whose overall input-output characteristics are determined by a set of adjustable parameters. Typically, the performance of a network is measured by the difference between the desired output and the network output under the same input conditions. This difference is defined as an error indicator. A particular optimization technique is applied to a given error indicator to obtain a learning rule. The steepest descent method is the most commonly used learning rule, and if gradient vectors are applied in the steepest descent method, the resulting method is called a back propagation idea-based parameter learning rule.
The basic idea of a parameter learning rule (BP) based on a back propagation idea is to define an overall error index of a system and then optimize the index according to the learning rule. The overall error indicator for the system may be defined as follows:
wherein E ispIs the error index of the P-th pair of training data, and E is the overall error index of the system. dkIs the k component, x, of the expected output of the P training data setL,kIs the kth component of the actual output produced when the pth set of training data input vectors is applied to the network. Our goal is to make E (E)p) And minimum.
Defining the error signal of the ith node of the ith layer as follows:
i.e. error index EpThe derivative of the output of the i-th node, i.e. one error signal for each node. We can see that EpAn error index of the last layer output node is pointed, so that only the error signal of the last layer node (output layer node) can be directly calculated out to be-2 (d)i-xi) (ii) a While the error signal of the hidden layer node (such as the middle layer node) can not be directly calculated. The error signal of the hidden layer node can be derived from the error signal of the node of the previous layer. Namely:
the mth node represents a node associated with the previous layer (l + 1).
Minimizing the error index E by applying the steepest descent methodpGradient vectors need to be calculated. The gradient vector is defined as: the derivative of the error indicator with respect to each parameter. The basic concept of computing a gradient vector is: starting from the output layer, the information in derivative form is propagated to the next layer until the input layer is reached.
If α is the parameter of the ith node at the l-th layer, then there are:
the derivative of the overall error indicator E with respect to α is:
the updated formula of the general parameter α is:I.e., learning in the negative gradient direction, η is the learning rate.
Thus, according to classical BP theory, the parameter update formula for Agg-ANFIS in this application is as follows:
wherein, Δ wijA delta representing a certain parameter (the parameters to be updated in the Agg-ANFIS: a set of preconditions, a set of conclusion parameters, a weight of each input, a weight of each rule); node i is a node of a layer immediately above node j, j<i, i.e. xi=fi(∑wij.xj+θ)。fiAnd xiRepresenting the excitation function and the output of node i. Error signal epsiloniThe error signals of each node can be derived from the upper layer error signals. In the formula,(finding a term in the error signalAndcan be solved, namely the parameters can be updated according to a formula.
If the number of the first and second antennas is greater than the predetermined number,namely, it isThe parameter updating formula of each node is known, and the parameters (weight) of the whole network can be updated.
More generally, the Agg-ANFIS parameter update formula is as follows:
Δwji=η(di-xi)·xj·X (12)
where η is the learning step size, diIs the desired output, x, of node iiIs the actual output of node i, xjIs the input to node i, X is a polynomial, generally expressed as (X)i×(1-xi) ) is expressed.
The above is a description of the adaptive neuro-fuzzy inference system involved in the method of the present invention, and the following is a detailed description of how the adaptive neuro-fuzzy inference system is applied to road traffic state evaluation.
As a qualitative measure of urban road traffic conditions, the traffic service Level (LOS) reflects the driver's perception of traffic conditions, and describes the driver's operating conditions in the traffic flow, such as driving speed or travel time, driving freedom, traffic disturbances, comfort or convenience, etc. The nature of the traffic service level evaluation problem is a multi-object (index) decision problem, the nature of the decision (reasoning) problem is nonlinear mapping, and the objective of the application is to establish a model to describe the nonlinear mapping relation.
Firstly, the Agg-ANFIS has the continuous mapping capability, can transparently express the human thinking mode by realizing the process of approximate reasoning, and can realize the decision process of the traffic state LOS; meanwhile, in the process of describing human decision (evaluation), in order to enable the inference result to approach the target decision result, the parameters of the model must be adjusted, and the adaptive capacity of the Agg-ANFIS can just solve the problem.
The following are specifically considered:
1) the service level is a subjective assessment of the quality of service provided. For each level of service level, it is described in natural language, i.e. the rule back-part is expressed in fuzzy sets. For example: very fast, very low traffic, LOS rating of a (very good), etc.
The ANFIS back-piece expression is a linear function, the service level reflects the subjective feeling of the driver on the road condition, and the model cannot really show the essential meaning of the service level.
The Agg-ANFIS model can solve the problem, and in the fuzzy rule back part, the service level grade is represented by a fuzzy set (membership function) instead of a simple linear equation, so that the concept of the service level grade is really reflected, and the method is more close to the requirement of solving the problem.
2) For a certain basic traffic facility, LOS evaluation can be carried out by a plurality of indexes, namely the evaluation indexes are not unique, and the problem of multi-index comprehensive evaluation is to be solved.
3) The problem of transitions between levels. For example: 0.54 belongs to class B and 0.55 belongs to class C, and such a classification is not reasonable. The Agg-ANFIS back-part can carry out fuzzy hierarchical representation on the service Level (LOS), so that the traffic state evaluation is closer to the human thinking mode.
Therefore, the adaptive neural fuzzy inference system (Agg-ANFIS) based on Choquet integral-OWA is selected to solve the LOS evaluation problem, and the problems of service level subjectivity, multi-index evaluation, grade transition and the like can be solved.
Basic data (flow, occupancy and speed) of the experiment are obtained through a detector, and the historical data are real and effective and can be used for LOS (LOS) evaluation of multiple objects (indexes); a traffic status service Level (LOS) is fuzzy and graded based on existing historical data, A-F. All data are divided into training data and test data. According to the mapping relation between the system input and the LOS value, the service level evaluation model based on the Agg-ANFIS is established, the model is trained through historical data, and the trained model is tested.
FIG. 3 is a schematic diagram of the service level evaluation model based on Agg-ANFIS established by the invention, which realizes the road traffic stateEvaluation of (3). In this model, x, y, z represent input variables representing flow, occupancy and velocity, respectively. A. the1-A3Is a fuzzy set of flows; b is1-B3Is a fuzzy set of occupancy rates; c1-C3Is a fuzzy set of velocities; d1-D6Indicating six levels of output LOS. f is the value of the output variable, i.e. the service Level (LOS). The number of the preconditions in this model is 3 × 3 × 3 — 27 (the model includes a)1-A3,B1-B3And C1-C3There are 9 membership functions, each membership function includes 3 premise parameters, so there are 9 × 3 — 27 premise parameters in total, and there are 24 conclusion parameters (including D in the model)1-D6A total of 6 membership functions, each function having 4 conclusion parameters, so that the conclusion parameters have 6 × 4 — 24); setting a weight corresponding to each input variable, wherein the weight is 3 parameters in total; a weight is set for each rule, 27 parameters are set in total, and the initial value is 1. These parameters (including the aforementioned 27 preconditions, 24 conclusions, 3 weights for the input variables, and 27 weights for the rules) are all non-linear parameters, so that the BP algorithm is applied to the entire model for parameter adjustment.
The fuzzy rule of the model is specifically expressed in the following form:
Rule i:IF speed is slow,AND occupancy is high,AND volume is high,THENLOS=F。
on the basis of the service level evaluation model based on the Agg-ANFIS, the method comprises two stages, wherein the first stage is a training stage, and the second stage is a testing stage. The work to be done in the training phase is to train the service level evaluation model based on Agg-ANFIS using the existing historical data, i.e. to adjust the set of parameters (preconditions, conclusions) of the model, and when the model reaches the minimum error index, the training is finished. And the work to be completed in the testing stage is to input test data into the service level evaluation model based on the Agg-ANFIS on the basis of the service level evaluation model based on the Agg-ANFIS obtained in the training stage to obtain a service level evaluation result.
The specific implementation steps of these two phases are described in detail below.
First, training phase
Step 101), reading existing historical data, wherein the historical data comprises: historical data of input variables and output variables; the input variables include flow, occupancy, and speed, and the output variables include service level values.
102) inputting the historical data into a model Agg-ANFIS, referring to a formula (11) or a formula (12), and adjusting parameters of the model Agg-ANFIS by using an output result of the model Agg-ANFIS to obtain the model Agg-ANFIS after parameter adjustment; wherein,
the parameters of the model Agg-ANFIS comprise 27 precondition parameters, 24 conclusion parameters, 3 weight parameters set for input variables and 27 weight parameters set for rules;
step 103), calculating the minimum mean square error of the model Agg-ANFIS after the parameters are adjusted, which is obtained in the step 102), judging whether the value reaches a specified threshold value, if so, indicating that the model training is finished; otherwise, executing the next step;
step 104), read a new set of historical data, and then re-execute step 102).
Second, testing stage
Step 201), reading test data, wherein the test data comprises: historical data including flow, occupancy, and speed;
step 202), inputting the test data into a model Agg-ANFIS to obtain a service Level (LOS) value.
The above is a description of the steps of the method of the present invention. The method of the invention is compared with other methods of the prior art to demonstrate the effectiveness of the method of the invention.
Firstly, establishing a Sugeno-FIS and AggFIS fuzzy inference system for experimental contrast analysis, and establishing a corresponding ANFIS and Agg-ANFIS-based service level evaluation model, and analyzing as follows:
1. in ANFIS and Agg-ANFIS evaluation models, the determination of membership function (fuzzy set) parameters of speed, flow and occupancy in fuzzy rule antecedents can be obtained according to historical data analysis; the determination of the fuzzy rule back-piece service level grade can be obtained according to historical data analysis. However, this classification of the blur level is not the most accurate, and therefore, the level adjustment is performed. Historical data is input, and an adaptive neural fuzzy inference system (Agg-ANFIS) based on Choquet integral-OWA is trained according to learning rules, namely parameter sets (preconditions and conclusions) of a model are adjusted, so that the Agg-ANFIS has the embodiment of learning function (self-adaption). In particular, the inference process of Agg-ANFIS takes into account the weights of the objects (system inputs and each rule), which are also parameters to be learned (trained). When the system reaches the minimum error index, training ends.
2. The fuzzy hierarchical representation of the traffic status service Level (LOS) based on the existing historical data is as follows:
1) obtaining Imin and Imax of an evaluation index I (domain of discourse) of certain LOS according to historical data;
2) for a certain index I (domain), LOS grade division (initialization membership function) is carried out, and ABCDEF6 grades are divided in total.
When taking different types of membership functions, we have corresponding conclusion parameter sets.
3) For the evaluation index I, the historical values I (ti) at times t1 and t2 … … tn are represented by I (t1), I (t2), … …. And also the last column of training data, i.e., the desired output.
And calculating an evaluation index I' (ti) according to the established fuzzy inference system. Corresponding to a certain moment there are:
at a certain time ti, corresponding to a specific fuzzy input, according to a defined fuzzy rule, a value of each row of the matrix can be deduced, and according to a central de-fuzzy method, an accurate value of the evaluation index I at the time ti is obtained, wherein the value is a value calculated by AggFIS, I' (ti).
At time ti, the value of index I is known as I (ti). For example, we use V/C as the evaluation index I and establish ABCDEF6 fuzzy sets at this domain of discourse, then at time ti, V/C is known. And comparing the known value with the value deduced by the model to obtain an error signal I' (ti) -I (ti), and adjusting the established model parameters of the Agg-ANFIS according to a BP algorithm.
When the system input is known, the expected output I (ti) is known, and the error signal is known, a corresponding Agg-ANFIS model can be constructed to realize LOS evaluation, and parameters in the model can be adjusted (learned). All parameters in the Agg-ANFIS model are nonlinear parameters, a gradient descent method (steepest descent method) is used for adjustment, and a formula reference is updated according to specific parameters (formula 11). When the minimum error index of the system is reached, the training is finished.
The specific experimental steps are as follows:
Sugeno-FIS and ANFIS experiments
First, a Sugeno-FIS fuzzy inference system was established. The detection data which can be obtained by the user, namely the flow, the occupancy rate and the speed, are used as system input; and the service level value (grade) is output by the system. From the training (historical) data, the actual output curve of the S-FIS may be obtained.
And secondly, establishing a corresponding ANFIS service level evaluation model, and determining a precondition parameter set and a conclusion parameter set. Inputting historical data (training data), adjusting model parameters by adopting a mixed learning rule according to the error between the actual output of the Sugeno-FIS model and the expected output of the training data, and finishing training when the response times of the system are reached (40). The parameter sets before and after training may be compared.
And finally, inputting test data (smaller than a training data space) to test the trained Sugeno-FIS model, so that a system test error can be obtained and can be used as an important index of the model effectiveness.
The Sugeno-FIS output graph is shown in FIG. 4, wherein the abscissa indicates the training times and the ordinate indicates the actual output of the Sugeno-FIS fuzzy inference system, the graph reflects the actual output of the Sugeno-FIS fuzzy inference system, the error between the actual output and the expected output of the Sugeno-FIS fuzzy inference system is shown in FIG. 5, wherein the abscissa indicates the training times and the ordinate indicates the error between the actual output and the expected output of the Sugeno-FIS fuzzy inference system. The error plot of the training data for ANFIS is shown in fig. 6, where the abscissa indicates the number of responses and the ordinate indicates the training error for ANFIS. And training by adopting a hybrid algorithm, wherein the fault tolerance index is 0, and the response times are 40. The error is gradually decreased along with the increase of the training times, and the training data set selected in the application is proved to be effective. The final training error is error 1.0038 e-005.
Fig. 7 is a graph comparing the expected output value of the test data with the actual output value of the Sugeno-FIS after training, in which the abscissa represents the test data and the ordinate represents the actual output value of the Sugeno-FIS after training (indicated by the star) and the expected output value of the test data (indicated by the cross), and it can be seen from the graph that the expected output value of the test data and the actual output value of the Sugeno-FIS after training match well. The trained S-FIS model can be used for system identification and prediction and is a better nonlinear approximator. The average test error was 0.091368. The experimental time is 8.3110 seconds.
(1) Before and after the change of the membership function (nonlinear parameter) of the front part, parameter comparison is carried out:
(2) comparison before and after change of the parameters of the latter conclusion (linear parameters):
conclusion parameters | Before change | After being changed |
Mf1 | [0.0002 0.006 -0.005 0.4]; | [0.0002906 0.006583 -0.00548 0.4235] |
Mf10 | [0.0002 0.006 -0.005 0.2] | [0.0002911 0.006667 -0.005522 0.4228] |
Mf 24 | [0.0004054 0.01235 -0.003485 0.004147] | [0.0003725 0.01571 -0.004825 0.004552] |
Second, AggFIS and Agg-ANFIS experiments
Firstly, an AggFIS model can be established according to the inference process of AggFIS, the inference of the AggFIS is realized through programming, and the actual output of the AggFIS can be obtained by inputting training data. To ensure that the model parameters have learnability, the computational continuity of the operators is taken care of when selecting each operator. That is, the whole reasoning process is micro-conductive, and the model established by the user has learnability and can adjust the parameters.
Then, we establish a corresponding Agg-ANFIS model for LOS evaluation, and the parameters to be adjusted are:
1) parameters of membership functions of the front piece and the rear piece;
2)ui: the weight of each input is 1 in an initial value;
3)τi: the weight of each rule is initially 1.
All parameters of the Agg-ANFIS model include: the precondition parameter set and the conclusion parameter set are all nonlinear parameters, the weight of each input and the weight of each rule are also objects to be adjusted, the parameter learning rule (formula 11) is adopted for adjustment according to the output of the AggFIS model and the error of the expected output of the training data, and the training is finished when the minimum error index of the system is reached. The parameter sets before and after training may be compared.
Finally, test data (smaller than the training data space) is input, and the model is tested, so that the test error of the model can be obtained.
Before training, the membership function of the input variable and the output variable is shown in fig. 8, wherein fig. 8(a) shows the membership function of the flow rate in the input variable, the abscissa of the graph shows the threshold value of the flow rate, and the ordinate shows the membership function of the flow rate; FIG. 8(b) is a graph showing membership functions for occupancy in an input flow rate, with the abscissa of the graph showing threshold values for occupancy and the ordinate showing membership functions for occupancy; FIG. 8(c) is a membership function of velocity in an input variable, the abscissa of the graph representing a threshold value of velocity and the ordinate representing a membership function of velocity; fig. 8(d) shows the membership function of the output variable LOS, with the abscissa showing the threshold value of LOS and the ordinate showing the membership function of LOS.
Fig. 9 is a graph of aggfs output with 1429 pairs of training data plotted on the abscissa and aggfs output plotted on the ordinate, illustrating the actual output of aggfs prior to training.
FIG. 10 is a graph of error for actual and expected output of AggFIS, with the abscissa representing 1429 versus training data and the ordinate representing error for AggFIS output versus expected output; the graph illustrates that the actual output of the AggFIS model before training is in large error with the expected output, but within the allowable range and evenly distributed up and down the 0 value.
The parameter adjustment and training of the Agg-ANFIS model takes 2.624 seconds. MSE (training error) 0.00022442; as (test error) 0.057391. The experimental results are as follows:
inputting a weight matrix before training:
150 250 400
2.5 6 10
15 9 15
outputting a weight matrix before training:
matrix of mu values of input variables before training
1 1 1
Regular tau value matrix before training
Columns 1through 27
Input weight matrix after training:
149.87 249.93 399.9
2.4567 6.2972 10.297
7.3413 8.3121 11.808
the output weight matrix after training:
input variable mu value matrix after training
0.60937 0.97412 0.59182
Regular tau value matrix after training
Columns 1through 27
FIG. 11 is a graph of the training error for Agg-ANFIS, with the abscissa representing the training sample and the ordinate representing the training error; the figure illustrates that the training error decreases with increasing training samples, proving that the established Agg-ANFIS model is superior.
FIG. 12 is a graph of the output of the trained AggFIS versus expected output of the test data, with 640 on the abscissa and 640 on the ordinate representing the expected output of the test data (indicated by- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -; the figure illustrates that the actual output of the trained AggFIS is well matched with the expected output of the detection data, and the model is trained successfully.
FIG. 13 is a graph of AggFIS test error plotted with 640 pairs of samples plotted on the abscissa and AggFIS test error plotted on the ordinate, showing that AggFIS test error is low.
FIG. 14 is a modified membership function, where FIG. 14(a) is a membership function for flow in the input variable, FIG. 14(b) is a membership function for occupancy in the input flow, and FIG. 14(c) is a membership function for velocity in the input variable; fig. 14(d) shows a membership function of the output variable LOS. In comparison with fig. 8, the membership function parameters of each input variable and output variable are changed.
Finally, it should be noted that the above embodiments are only used for illustrating the technical solutions of the present invention and are not limited. Although the present invention has been described in detail with reference to the embodiments, it will be understood by those skilled in the art that various changes may be made and equivalents may be substituted without departing from the spirit and scope of the invention as defined in the appended claims.
Claims (3)
1. A road traffic state evaluation method based on a self-adaptive neural fuzzy inference system comprises a training stage and a testing stage; wherein,
the training phase comprises the steps of:
step 101), reading historical data of input variables and output variables; the input variables include flow, occupancy, and speed, and the output variables include service level values;
step 102), inputting the historical data input in the step 101) into the self-adaptive neural fuzzy inference system to obtain an output result of the self-adaptive neural fuzzy inference system; adjusting parameters of the adaptive neural fuzzy inference system by combining an output result of the adaptive neural fuzzy inference system based on a parameter learning rule of a back propagation thought to obtain the adaptive neural fuzzy inference system after parameter adjustment; wherein,
the parameters of the self-adaptive neural fuzzy inference system comprise a precondition parameter, a conclusion parameter, a weight parameter set for an input variable and a weight parameter set for a rule; the precondition parameters are determined by the shape of a membership function of the inference antecedent; the conclusion parameters are determined by the membership function shape of the inferred piece;
step 103), calculating the minimum mean square error of the adaptive neural fuzzy inference system after the parameters are adjusted, which is obtained in step 102), and judging whether the value reaches a specified threshold value, if so, the training stage is ended; otherwise, executing the next step;
step 104), reading a new set of historical data about input variables and output variables, and then re-executing step 102);
the testing phase comprises the following steps:
step 201), reading test data, wherein the test data comprises: historical data of flow, occupancy and speed;
step 202), inputting test data into a self-adaptive neural fuzzy inference system obtained through training in a training stage to obtain a value of service level for describing a road traffic state;
the self-adaptive neural fuzzy inference system comprises five layers which are respectively:
a first layer: a fuzzification layer for fuzzifying the precision input;
O1,i=μAi(x) I is 12 or
Wherein, Oj,iI-th node output, μ, representing the j-th layerAi、The membership functions of the front parts are defined by generalized bell-shaped MF:
or
Wherein a isi、bi、ciAs a precondition parameter;
a second layer, inference layer, for computing the excitation strength w of each rulei;
Wherein i is 1,2, muiThe weight of each of the inputs is represented,
the third layer, the hidden layer, is used for calculating the effective back-part MF of each rule;
wherein i is 1 or 2; ciA post-reasoning part;representing an implicit operator;
the fourth layer is a polymerization layer and is used for calculating the sum of all the rule effective back pieces MF;
wherein i is 1 or 2; tau isiA weight representing each rule; aggregation operatorAdopting choquet integration;
the fifth layer is a de-fuzzy layer and is used for calculating the accurate output of the system;
wherein D represents a deblurring operator, and the calculation of the deblurring operator is realized by adopting a central deblurring method.
2. The method for evaluating the road traffic condition based on the adaptive neuro-fuzzy inference system as claimed in claim 1, wherein the parameters of the adaptive neuro-fuzzy inference system are adjusted according to the following formula:
Δwji=η(di-xi)·xj·X
wherein, Δ wijRepresenting the increment of any parameter in the adaptive neuro-fuzzy inference system η is the learning step size diIs the desired output, x, of node iiIs the actual output of node i, xjIs an input to node i, which is a node of a layer immediately above node j, j<i; x is a polynomial of formula (X)i×(1-xi) ) is expressed.
3. The road traffic state evaluation method based on the adaptive neuro-fuzzy inference system according to claim 1, characterized in that the adaptive neuro-fuzzy inference system replaces AND operator OR operator with OWA operator at an inference layer to calculate excitation intensity; polymerization is achieved by choquet integration in the polymerization layer; the defuzzification operator adopts a central defuzzification method; and setting the weight of each input by muiRepresents; setting the weight of each rule by tauiAnd (4) showing.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201510438629.9A CN106373390B (en) | 2015-07-23 | 2015-07-23 | Traffic state evaluation method based on Adaptive Neuro-fuzzy Inference |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201510438629.9A CN106373390B (en) | 2015-07-23 | 2015-07-23 | Traffic state evaluation method based on Adaptive Neuro-fuzzy Inference |
Publications (2)
Publication Number | Publication Date |
---|---|
CN106373390A CN106373390A (en) | 2017-02-01 |
CN106373390B true CN106373390B (en) | 2018-10-26 |
Family
ID=57880011
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201510438629.9A Active CN106373390B (en) | 2015-07-23 | 2015-07-23 | Traffic state evaluation method based on Adaptive Neuro-fuzzy Inference |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN106373390B (en) |
Families Citing this family (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108052793B (en) * | 2017-12-26 | 2023-10-31 | 杭州电子科技大学 | Mobile pollution source emission concentration prediction method based on fuzzy weighted ELM |
CN108346293B (en) * | 2018-04-18 | 2020-12-18 | 湖南大学 | Real-time traffic flow short-time prediction method |
CN109214516B (en) * | 2018-07-20 | 2021-09-03 | 中国航空工业集团公司沈阳飞机设计研究所 | Complete decision generation method of fuzzy inference system |
CN109147319B (en) * | 2018-08-06 | 2020-05-05 | 北京航空航天大学 | Road emergency discrimination method based on multiple traffic data indexes |
CN111539920B (en) * | 2020-04-16 | 2022-03-15 | 泸州老窖酿酒有限责任公司 | Automatic detection method for fermented grain quality in white spirit brewing process |
CN111538232B (en) * | 2020-04-17 | 2023-06-13 | 东南大学 | Unmanned driving anti-rolling positioning method and system based on self-adaptive neural fuzzy control |
CN111583639B (en) * | 2020-04-30 | 2021-05-14 | 山东交通学院 | Road traffic jam early warning method and system |
CN111950203A (en) * | 2020-08-13 | 2020-11-17 | 中核华辰建筑工程有限公司 | Blasting vibration speed prediction method based on adaptive neural fuzzy inference system |
CN113885398B (en) * | 2021-10-25 | 2022-07-01 | 天津大学 | Water circulation intelligent sensing and monitoring system based on micro-reasoning |
CN113793511B (en) * | 2021-11-18 | 2022-02-15 | 西南交通大学 | Highway vehicle driving indication method, device, equipment and readable storage medium |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US6177885B1 (en) * | 1998-11-03 | 2001-01-23 | Esco Electronics, Inc. | System and method for detecting traffic anomalies |
CN101599217A (en) * | 2009-07-17 | 2009-12-09 | 北京交通大学 | A kind of method for rapidly judging traffic state |
CN104155574A (en) * | 2014-07-31 | 2014-11-19 | 国网湖北省电力公司武汉供电公司 | Power distribution network fault classification method based on adaptive neuro-fuzzy inference system |
-
2015
- 2015-07-23 CN CN201510438629.9A patent/CN106373390B/en active Active
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US6177885B1 (en) * | 1998-11-03 | 2001-01-23 | Esco Electronics, Inc. | System and method for detecting traffic anomalies |
CN101599217A (en) * | 2009-07-17 | 2009-12-09 | 北京交通大学 | A kind of method for rapidly judging traffic state |
CN104155574A (en) * | 2014-07-31 | 2014-11-19 | 国网湖北省电力公司武汉供电公司 | Power distribution network fault classification method based on adaptive neuro-fuzzy inference system |
Non-Patent Citations (3)
Title |
---|
一种交通流状态智能推理***;王辉;《***工程》;20071231;第25卷(第12期);第7-13页 * |
基于T-S模型的交通状态自适应神经模糊推理***建模与仿真;朱广宇等;《北京交通大学学报》;20121231;第36卷(第6期);第96-101页 * |
基于自适应神经模糊推理的交通状态判别方法;陈力;《西部交通科技》;20100531(第34期);第98-101页 * |
Also Published As
Publication number | Publication date |
---|---|
CN106373390A (en) | 2017-02-01 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN106373390B (en) | Traffic state evaluation method based on Adaptive Neuro-fuzzy Inference | |
CN106781489B (en) | A kind of road network trend prediction method based on recurrent neural network | |
CN108492555B (en) | Urban road network traffic state evaluation method and device | |
CN105321345B (en) | A kind of road traffic flow prediction method filtered based on ARIMA models and kalman | |
CN104809877B (en) | The highway place traffic state estimation method of feature based parameter weighting GEFCM algorithms | |
CN103606006B (en) | Sludge volume index (SVI) soft measuring method based on self-organized T-S fuzzy nerve network | |
CN112365724A (en) | Continuous intersection signal cooperative control method based on deep reinforcement learning | |
CN111738337B (en) | Driver distraction state detection and identification method in mixed traffic environment | |
CN104318757B (en) | Bus bus or train route section Forecasting Methodology working time on a kind of public transportation lane | |
CN109598381B (en) | Short-term traffic flow prediction method based on state frequency memory neural network | |
CN111027618B (en) | Automobile dynamic property and economic expectation quantification method | |
CN108280998A (en) | Short-time Traffic Flow Forecasting Methods based on historical data dynamic select | |
CN110111573B (en) | Congestion vehicle comprehensive scheduling method based on Internet of things | |
CN109398363A (en) | Pavement grade determination method and device, storage medium and automobile | |
CN112542049A (en) | Comprehensive management and control platform for intelligent traffic | |
CN111553527A (en) | Road passing time prediction method based on PSO and neural network series optimization | |
CN103714382A (en) | Multi-index comprehensive evaluation method for reliability of urban rail train security detection sensor network | |
CN113449905A (en) | Traffic jam early warning method based on gated cyclic unit neural network | |
CN107945534A (en) | A kind of special bus method for predicting based on GMDH neutral nets | |
CN111967308A (en) | Online road surface unevenness identification method and system | |
Göktepe et al. | Comparison of multilayer perceptron and adaptive neuro-fuzzy system on backcalculating the mechanical properties of flexible pavements | |
CN109165800A (en) | A method of realizing abandoned car yield Accurate Prediction | |
CN110704789B (en) | Population dynamic measurement and calculation method and system based on 'urban superconcephalon' computing platform | |
Cirianni et al. | Road traffic noise prediction models in the metropolitan area of the Strait of Messina | |
Maghsoudi et al. | Applying adaptive network-based fuzzy inference system to predict travel time in highways for intelligent transportation systems |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |