CN102722989A - Expressway microclimate traffic early warning method based on fuzzy neural network - Google Patents

Abstract

The invention discloses a fuzzy neural network-based expressway micro-meteorological traffic early warning method, comprising the following steps: laying out traffic flow and micro-meteorological detection points; defining a fuzzy neural network traffic controller; training the fuzzy neural network traffic controller; The optimal fuzzy neural network traffic controller is used to generate traffic safety driving parameters; traffic control information is released. This method uses the method of fuzzy neural network to comprehensively detect meteorological parameters such as rainfall, snowfall, temperature, and visibility along the expressway. Publish the vehicle speed limit value, vehicle distance limit value, overtaking limit and lane change limit measures. Applying the method on the expressway can improve driving safety under severe weather conditions.

Description

Micro-meteorological traffic early warning method for expressway based on fuzzy neural network

technical field

The invention relates to a traffic safety technology, in particular to a fuzzy neural network-based micro-meteorological traffic early warning method for expressways.

Background technique

At present, with the increase of expressway mileage, the impact of disastrous weather on expressway traffic safety has become increasingly prominent. In severe weather conditions, it is even more difficult for drivers to obtain warning information in time, resulting in major accidents involving hundreds of cars rear-end. Therefore, highways can only be closed in severe weather conditions. Expressway early warning and intelligent Humanized management has attracted more and more people's attention and attention.

Found through searching papers: Cheng Conglan, Li Xun, Zheng Zuofang, Wang Zaiwen, Liang Xudong. Beijing Road Meteorological Early Warning Index Construction and Preliminary Application, Proceedings of the 27th Annual Conference of the Chinese Meteorological Society, 2010, October. Feng Minxue, Expressway Traffic Meteorological Intelligent Detection Research on Early Warning System, Doctoral Dissertation of Nanjing University of Information Engineering, 2005. Zhang Changjun, Zou Kaiqi. Research on NN Control System of Expressway under Severe Weather Conditions, Computer Engineering and Application, 2007, 43(4): 210-212. Wang Shaofei, Guan Ke. Expressway Meteorological Information Service System. China Transportation Information Industry, 2007 (1): 116-119. The above research is based on the analysis of the impact of different meteorological conditions on traffic capacity, and provides meteorological warning information for different meteorological conditions. However, it does not give traffic warning information directly from the perspective of expressway management. Tang Junjun, Gao Hailong, Zhang Weihan. Research on the Structural Scheme of the Monitoring System in the Fog Area of Expressway, Highway, 2005,8. Wang Weiya. Research on the Bad Weather Monitoring and Traffic Control Model of Expressway Based on Wireless Sensor Network, Chang'an University, 2008. Liu Benmin, Research on Safety Control Technology of Expressway Operation under Disastrous Weather, Doctoral Dissertation of Tongji University, 2008. On the premise of analyzing the influence of visibility and road surface adhesion coefficient on expressway driving safety, the above research adopted fuzzy control theory or meteorological department, The expressway management department and the actual experience and tests of front-line drivers stipulate the speed limit and safety distance in fog, snow and rain. There are three problems: first, the early warning of vehicle driving safety is not based on comprehensive weather monitoring, and the considerations are not comprehensive enough; second, the membership degree of fuzzy reasoning is directly given by experience, which is highly subjective. Third, the influence of road traffic flow conditions is not considered.

Chinese patent application 200910061448.3 discloses a highway weather monitoring system. And the application number: 200910060562.4 discloses a kind of expressway anti-rear-collision warning system. Chinese patent application 200710077671.8 discloses a meteorological information prompt system for expressway sections. The above-mentioned inventions are all designed for the hardware facilities of the expressway early warning system, and do not involve the method of obtaining safe driving parameters and traffic early warning on the basis of data collection.

Contents of the invention

The purpose of the present invention is to overcome above-mentioned deficiencies in the prior art, provide a kind of expressway micro-meteorological traffic early-warning method based on fuzzy neural network, this method is through the synthesizing of meteorological parameters such as rainfall, snowfall, temperature, visibility along the expressway Detection, using the method of fuzzy neural network, publishes measures such as vehicle speed limit value, vehicle distance limit value, overtaking limit and lane change limit. Applying the method on the expressway can improve driving safety under severe weather conditions.

To achieve the above object, the present invention adopts the following technical solutions:

A fuzzy neural network-based expressway micro-meteorological traffic early warning method, comprising the following steps: layout of traffic flow and micro-meteorological detection points; definition of fuzzy neural network traffic controller; training of fuzzy neural network traffic controller; using optimal The fuzzy neural network traffic controller generates traffic safety driving parameters; the traffic control information is released, and the specific operation steps are as follows:

Step 1: Set up several traffic flow and micro-meteorological monitoring points at certain intervals on the roadside of the expressway to detect the traffic flow and micro-meteorological parameter data of the road section, and obtain the traffic flow and meteorological information of the road section through multi-sensor data fusion technology. Acquisition of flow and micro-meteorological data prepares for the next step of fuzzy neural network control;

Step 2: Use the fuzzy neural network based on Takagi-Sugeno (Takagi-Sugeno) inference to construct the highway micro-meteorological traffic controller, define the traffic flow information and meteorological information collected in step 1 as state variables, as the controller The input value defines the expressway control mode, the speed limit value and the safety distance as control variables, as the output value of the controller;

Step 3: Use the weather, traffic flow, control measures and their implementation effect historical database of the meteorological department and the traffic management department, combine the expert experience to construct the training sample of the fuzzy neural network traffic controller, and train the micro-meteorological traffic controller of the expressway. Reduce the training error to a predetermined threshold or reach the number of training times to obtain an optimal controller;

Step 4: Input the real-time collected traffic flow information and micro-meteorological information into the optimal fuzzy neural network traffic controller to generate a highway traffic control plan for the traffic flow conditions and meteorological information at that moment, including early warning measures, speed limit values and Safety distance value;

Step 5: Arrange a variable information board upstream of each traffic and micro-meteorological monitoring point to release the safe driving warning information output by the traffic controller.

In the said step 2, the fuzzy neural network constructs the highway micro-meteorological traffic controller and is divided into five layers:

The first layer is the input layer, and the input value is the parameter data collected by highway traffic flow and micro-meteorological detection points, expressed as x={x ₁ ,x ₂ ,x ₃ ,...,x _n }, where n represents The number of input parameters, n is an integer greater than or equal to 1, and x ₁ , x ₂ ,...x _n represent traffic flow parameters, namely speed, flow, occupancy rate and micro-meteorological parameters, namely visibility, temperature, humidity, air pressure , wind direction, wind speed, ground temperature;

Divide the input parameters into 5 grades according to the order from small to large, which are {NB negative large, NS negative small, Z zero, PS positive small, PB positive large}, which means that the corresponding parameter index value is small and small , medium, large, large; the system output of the controller is the control mode, the speed limit value and the safety distance; among them, the control mode is divided into three types, the whole line closed, the area closed and the ramp control are determined according to the traffic and weather conditions. Control the flow of entering the expressway; the division method of speed limit and safety distance is the same as that of the input parameters;

式中,j为1至m_i的整数，m_i是x_i的模糊分割数，此处m_i＝5；c_ij和σ_ij分别表示隶属度函数的中心和宽度；The second layer is the fuzzy layer, which is used to indicate that the input quantities belong to {NB, NS, Z, PS, PB} and the degree of membership is

In the formula, j is an integer from 1 to m _i , and m _i is the fuzzy division number of x _i , where m _i =5; c _ij and σ _ij represent the center and width of the membership function respectively;

The third layer is the rule antecedent layer. Each node represents a fuzzy rule. Its function is to match the antecedent of the fuzzy rule and calculate the applicability a _j of each rule. The formula is:

or

${\mathrm{<span\; class="notranslate">\; a</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; \&mu;</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}^{\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; )</span>{\mathrm{<span\; class="notranslate">\; \&mu;</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}^{\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; .</span>{\mathrm{<span\; class="notranslate">\; \&mu;</span>}}_{\mathrm{<span\; class="notranslate">\; no</span>}}^{\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; no</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ;</span>$

In the formula: a _j —applicability of fuzzy rule j;

—Input the degree of membership that x _i belongs to the jth level, j is an integer between 1 and m, and m is an integer greater than or equal to 1;

The fourth layer implements normalized calculation, the formula is:

其中，j为1至m之间的整数，m为大于等于1的整数；

Wherein, j is an integer between 1 and m, and m is an integer greater than or equal to 1;

The fifth layer output layer, the formula is

$\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; r</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; m</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{\mathrm{<span\; class="notranslate">\; w</span>}}_{\mathrm{<span\; class="notranslate">\; r</span>}}\stackrel{<span\; class="notranslate">\; \&OverBar;</span>}{{\mathrm{<span\; class="notranslate">\; a</span>}}_{\mathrm{<span\; class="notranslate">\; r</span>}}}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; r</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1,2</span>}<span\; class="notranslate">\; ,</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; m</span>}$

Wherein, w _r is the connection weight, r is an integer between 1 and m, and m is an integer greater than or equal to 1.

The training of the micro-meteorological traffic controller based on fuzzy neural network in described step 3, specific content is as follows:

(1) Selection of training samples

The training samples are considered to be composed of three parts: the first part, the meteorological part, the historical data of the highway traffic management department, including historical meteorological parameter information, traffic flow parameter information, early warning information taken at that time and the effect after its release; the second part, Questionnaire for experts in the field of traffic engineering, by setting different weather and traffic flow scenarios, and asking experts for possible early warning measures; in the third part, after the system is completed, the implementation effect of traffic early warning schemes under different weather and traffic flow conditions will be added to the historical database ;

(2) Training algorithm

The parameters that the network needs to learn are the central value c _ij and width σ _ij of the membership function of the second layer in step 2 and the connection weight w _r of the fifth layer network. The learning algorithm of this network is the backpropagation BP algorithm; It consists of two processes: forward propagation and error backpropagation; in the process of forward propagation, the input information is processed layer by layer from the input layer through the hidden layer unit, and after passing through all hidden layers, it is transmitted to the output layer; in the hidden layer layer by layer During the processing, the state of neurons in each layer only affects the state of neurons in the next layer; compare the actual output with the expected output in the output layer, if the difference between the actual output and the expected output is no longer acceptable In the process of reverse propagation, the error between the actual value and the expected output is returned along the original connection path, and the error is reduced by modifying the connection weights of neurons in each layer, and then transferred to the forward propagation process. This calculation is repeated until the error is less than the set value;

The learning algorithm of the network selects the backpropagation BP algorithm, and the steps are as follows:

① Let x _k be the input vector, x _k = (x ₁ , x ₂ ,..., x _n ), k is an integer between 1 and K, where: K is the number of samples, n represents the number of characteristic parameters number; the output vector corresponding to the traffic mode is y _k , and the network weights and thresholds are initialized;

② The input of each unit in the second layer is

${\mathrm{<span\; class="notranslate">\; S</span>}}_{\mathrm{<span\; class="notranslate">\; ij</span>}}^{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 2</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; ,</span>$

In the formula, x _i represents the system input, that is, traffic flow parameters and meteorological parameter values, i is an integer between 1 and n, and n represents the number of characteristic parameters; j is an integer between 1 and m _i , and m _i represents The number of fuzzy divisions of the i-th feature parameter;

The transfer function between the first layer and the second layer is the membership function:

${\mathrm{<span\; class="notranslate">\; \&mu;</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; exp</span>}<span\; class="notranslate">\; [</span>{<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; c</span>}}_{\mathrm{<span\; class="notranslate">\; ij</span>}}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; /</span>{\mathrm{<span\; class="notranslate">\; \&sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; ij</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; ]</span>$

Then the output of the second layer unit is:

${\mathrm{<span\; class="notranslate">\; the\; y</span>}}^{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 2</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; =</span><span\; class="notranslate">\; \{</span>{\mathrm{<span\; class="notranslate">\; \&mu;</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \}</span>$

③ The input of each unit in the third layer is the output of the corresponding units in the second layer, which is

${\mathrm{<span\; class="notranslate">\; S</span>}}_{\mathrm{<span\; class="notranslate">\; r</span>}}^{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 3</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; =</span><span\; class="notranslate">\; \{</span>{\mathrm{<span\; class="notranslate">\; \&mu;</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{\mathrm{<span\; class="notranslate">\; r</span>}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \}</span>$

The output of the output layer is

${\mathrm{<span\; class="notranslate">\; the\; y</span>}}_{\mathrm{<span\; class="notranslate">\; r</span>}}^{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 3</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; min</span>}<span\; class="notranslate">\; \{</span>{\mathrm{<span\; class="notranslate">\; \&mu;</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{\mathrm{<span\; class="notranslate">\; l</span>}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \}</span>$

In the formula, l is an integer between 1 and m, and m represents the number of fuzzy rules;

④ The input of each unit in the fourth layer is the output of the corresponding unit in the third layer, which is

${\mathrm{<span\; class="notranslate">\; S</span>}}_{\mathrm{<span\; class="notranslate">\; r</span>}}^{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 4</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; =</span><span\; class="notranslate">\; \{</span>{\mathrm{<span\; class="notranslate">\; the\; y</span>}}_{\mathrm{<span\; class="notranslate">\; r</span>}}^{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 3</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; \}</span>$

The output of the output layer is

${\mathrm{<span\; class="notranslate">\; the\; y</span>}}_{\mathrm{<span\; class="notranslate">\; r</span>}}^{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 4</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; =</span>\frac{{\mathrm{<span\; class="notranslate">\; the\; y</span>}}_{\mathrm{<span\; class="notranslate">\; r</span>}}^{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 3</span>}<span\; class="notranslate">\; )</span>}}{\underset{\mathrm{<span\; class="notranslate">\; r</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; m</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{\mathrm{<span\; class="notranslate">\; the\; y</span>}}_{\mathrm{<span\; class="notranslate">\; r</span>}}^{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 3</span>}<span\; class="notranslate">\; )</span>}}$

⑤ The input of each unit in the fifth layer is the output of the corresponding unit in the fourth layer, which is

${\mathrm{<span\; class="notranslate">\; S</span>}}_{\mathrm{<span\; class="notranslate">\; r</span>}}^{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 5</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; =</span><span\; class="notranslate">\; \{</span>{\mathrm{<span\; class="notranslate">\; the\; y</span>}}_{\mathrm{<span\; class="notranslate">\; r</span>}}^{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 4</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; \}</span>$

The output layer output is

$\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; w</span>}}_{\mathrm{<span\; class="notranslate">\; r</span>}}{\mathrm{<span\; class="notranslate">\; the\; y</span>}}_{\mathrm{<span\; class="notranslate">\; r</span>}}^{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 4</span>}<span\; class="notranslate">\; )</span>}$

At this point, a forward pass process is completed;

⑥In the process of error backpropagation, the error calculation must be performed first,

For the fuzzy neural network, it is assumed that the error function E _t of the tth sample pair is defined as:

${\mathrm{<span\; class="notranslate">\; E.</span>}}_{\mathrm{<span\; class="notranslate">\; t</span>}}<span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; 2</span>}}{<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; the\; y</span>}}_{\mathrm{<span\; class="notranslate">\; 0</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}$

In the formula: y ₀ (t) is the expected output value of the system, y(t) is the actual output value of the system, t is an integer greater than or equal to 1, and represents the label of the sample.

The idea of backpropagation BP is used to supervise learning. By adjusting the weight values of the network, the value of the error function is minimized, so as to achieve the purpose of modifying the parameters of the membership function and the weight of the network connection;

⑦ Randomly select the next sample pair to provide to the network, and repeat the calculation process until the global error function of the network Less than a preset minimum value, that is, the network converges; or the number of learning times is less than a preset value, that is, the network cannot converge; where K is the number of learning samples;

⑧End of study.

The fuzzy neural network of Takagi-Sugeno (Takagi-Sugeno) reasoning in the present invention is a known technology, and will not be repeated here.

The beneficial effect of the present invention is that the micro-meteorological traffic control method of the expressway is based on the real-time weather and traffic flow conditions of the expressway, and adaptively adjusts the traffic control scheme, thereby improving the driving safety and traffic efficiency of the expressway under bad weather. Compared with other traffic warning methods in severe weather, it has the following differences:

1. The control rules do not need to be given in advance, and historical data and expert experience are used to train the model, which improves the accuracy of the control scheme and abandons the subjectivity of traffic managers in traffic management;

2. Generate a traffic control plan based on the information collected in real time, and use the variable information board in its upstream to implement information release, which improves the efficiency of severe weather traffic warning.

Description of drawings

Fig. 1 schematic diagram of micro-meteorological detection traffic early warning system;

Figure 2 is a schematic diagram of the layout scheme of traffic flow and micro-meteorological detection points;

Detailed ways

The present invention will be further described below in conjunction with accompanying drawings and examples.

The structure diagram of micro-meteorological traffic early warning system based on fuzzy neural network is shown in Figure 1. The system consists of three subsystems: Subsystem 1, traffic and micro-meteorological data acquisition system; Subsystem 2, fuzzy neural network controller; Subsystem 3, traffic early warning information release system. The relationship between these three subsystems is as follows: The traffic and micro-meteorological information collected by subsystem 1 is used as input value, which is input to the fuzzy neural network controller of subsystem 2, and the traffic warning scheme is output through the calculation of the controller, and the traffic warning scheme is input into the subsystem 3. Release it to the variable information board as early warning information.

Step 1: Layout of traffic flow and micro-meteorological detection points

The layout scheme of traffic flow and micro-meteorological detection points is shown in Figure 2. Module 1 in Fig. 1 represents the traffic flow and micro-weather monitoring module, which is arranged on the roadside at certain intervals. Geomagnetic induction coils are laid at the traffic flow monitoring points to detect the traffic speed, flow rate and occupancy rate of the section; the micro weather detection points are equipped with visibility detectors, temperature detectors, ground temperature detectors, humidity detectors, rainfall detectors, Freeze detectors, air pressure detectors, sand and dust detectors, hail detectors, snow detectors and sand and dust detectors and other sensors to detect the weather information in the area. Traffic flow and micro-meteorological data collection will prepare for the next step of fuzzy neural network control.

Step 2: Design of Fuzzy Neural Network Traffic Controller

The fuzzy neural network traffic controller realizes the release of traffic early warning information under severe weather, which includes some historical data, prior knowledge or expert experience in the field of traffic engineering in the fuzzy rules, so as to obtain reasonable information related to weather and traffic flow information. Adaptive traffic warning information. The fuzzy neural network traffic controller is divided into five layers:

The first layer is the input layer, and the input value is the parameter data collected by highway traffic flow and micro-meteorological detection points, expressed as x={x ₁ ,x ₂ ,x ₃ ,...,x _n }, where n represents The number of input parameters, and x ₁ , x ₂ ,...x _n represent traffic flow parameters (speed, flow rate, occupancy rate) and micro-meteorological parameters (visibility, temperature, humidity, air pressure, wind direction, wind speed, ground temperature wait).

Divide the input parameters into 5 levels according to the order from small to large, which are {NB negative large, NS negative small, Z zero, PS positive small, PB positive large}, which means that the corresponding parameter index value is small and small , Medium, Large, Larger. The output of the system is control mode, speed limit value and safety distance. Among them, there are three control modes, full line closure, area closure and ramp control (to control the flow of entering the expressway according to traffic and weather conditions); the speed limit and safety distance classification methods are the same as the input parameters.

式中j＝1,2,...,m_i,m_i是x_i的模糊分割数，此处m_i＝5。c_ij和σ_ij分别表示隶属度函数的中心和宽度；The second layer of the network is the fuzzy layer, which is used to indicate that the input quantities belong to {NB, NS, Z, PS, PB} and the degree of membership is

In the formula, j=1,2,...,m _i , m _i is the fuzzy division number of x _i , where m _i =5. c _ij and σ _ij represent the center and width of the membership function, respectively;

The third layer is the rule antecedent layer. Each node represents a fuzzy rule. Its function is to match the antecedent of the fuzzy rule and calculate the applicability of each rule. The formula is:

or

${\mathrm{<span\; class="notranslate">\; a</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; \&mu;</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}^{\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; )</span>{\mathrm{<span\; class="notranslate">\; \&mu;</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}^{\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; .</span>{\mathrm{<span\; class="notranslate">\; \&mu;</span>}}_{\mathrm{<span\; class="notranslate">\; no</span>}}^{\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; no</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ;</span>$

In the formula: a _j —applicability of fuzzy rule j;

—输入x_i隶属于第j个等级的隶属度，j为1至m之间的整数，m为大于等于1的整数；

—Input the degree of membership that x _i belongs to the jth level, j is an integer between 1 and m, and m is an integer greater than or equal to 1;

The fourth layer implements normalized calculation, the formula is:

$\stackrel{<span\; class="notranslate">\; \&OverBar;</span>}{{\mathrm{<span\; class="notranslate">\; \&alpha;</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}}<span\; class="notranslate">\; =</span>\frac{{\mathrm{<span\; class="notranslate">\; \&alpha;</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}}{\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; m</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{\mathrm{<span\; class="notranslate">\; \&alpha;</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1,2</span>}<span\; class="notranslate">\; ,</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; m</span>}$

The fifth layer output layer, the formula is

$\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; r</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; m</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{\mathrm{<span\; class="notranslate">\; w</span>}}_{\mathrm{<span\; class="notranslate">\; r</span>}}\stackrel{<span\; class="notranslate">\; \&OverBar;</span>}{{\mathrm{<span\; class="notranslate">\; a</span>}}_{\mathrm{<span\; class="notranslate">\; r</span>}}}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; r</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1,2</span>}<span\; class="notranslate">\; ,</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; m</span>}$

Among them, w _r is the connection weight,

Step 3: Training of micrometeorological traffic controller based on fuzzy neural network

(3) Selection of training samples

The training samples are considered to be composed of three parts: the first part, the meteorological part, the historical data of the expressway traffic management department, including historical meteorological parameter information, traffic flow parameter information, early warning information taken at that time and the effect after its release; the second part, Questionnaire for experts in the field of traffic engineering, by setting different weather and traffic flow scenarios, asking experts for possible early warning measures; the third part, after the system is completed, the implementation effect of traffic early warning schemes under different weather and traffic flow conditions will be added to the historical database .

(4) Training algorithm

The parameters that the network needs to learn are mainly the center value c _ij and width σ _ij of the second layer membership function and the fifth layer network connection weight w _r . The learning algorithm of the network chooses the Back Propagation BP (Back Propagation) algorithm. The BP algorithm consists of two processes: forward propagation and error back propagation. In the process of forward propagation, the input information is processed layer by layer from the input layer through the hidden layer unit, and after passing through all the hidden layers, it is transmitted to the output layer. In the process of hidden layer processing layer by layer, the state of neurons in each layer only affects the state of neurons in the next layer. In the output layer, the actual output is compared with the expected output. If the difference between the actual output and the expected output is no longer within the acceptable range, it will turn to the back propagation process, and the error between the actual value and the expected output will be along the original The connection path is returned, and the error is reduced by modifying the connection weight of neurons in each layer, and then transferred to the forward propagation process, and the calculation is repeated until the error is less than the set value.

The specific algorithm steps of BP neural network training are as follows:

① Let x _k be the input vector, x _k =(x ₁ ,x ₂ ,...,x _n ), k=1,2,...K, where: K is the number of samples, n represents the characteristic parameters The number of ; the output vector corresponding to the traffic mode is y _k . Initialize network weights, thresholds and related parameters.

② The input of each unit in the second layer is

${\mathrm{<span\; class="notranslate">\; S</span>}}_{\mathrm{<span\; class="notranslate">\; ij</span>}}^{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 2</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1,2</span>}<span\; class="notranslate">\; ,</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; ;</span>\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1,2</span>}<span\; class="notranslate">\; ,</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; .</span>{\mathrm{<span\; class="notranslate">\; m</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}$

In the formula, m _i represents the fuzzy segmentation number of the i-th feature parameter.

The transfer function between the first layer and the second layer is the membership function:

${\mathrm{<span\; class="notranslate">\; \&mu;</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; exp</span>}<span\; class="notranslate">\; [</span>{<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; c</span>}}_{\mathrm{<span\; class="notranslate">\; ij</span>}}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; /</span>{\mathrm{<span\; class="notranslate">\; \&sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; ij</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; ]</span>$

Then the output of the second layer unit is:

${\mathrm{<span\; class="notranslate">\; the\; y</span>}}^{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 2</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; =</span><span\; class="notranslate">\; \{</span>{\mathrm{<span\; class="notranslate">\; \&mu;</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \}</span>$

③ The input of each unit in the third layer is the output of the corresponding units in the second layer, which is

${\mathrm{<span\; class="notranslate">\; S</span>}}_{\mathrm{<span\; class="notranslate">\; r</span>}}^{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 3</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; =</span><span\; class="notranslate">\; \{</span>{\mathrm{<span\; class="notranslate">\; \&mu;</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{\mathrm{<span\; class="notranslate">\; r</span>}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \}</span>$

The output of the output layer is

${\mathrm{<span\; class="notranslate">\; the\; y</span>}}_{\mathrm{<span\; class="notranslate">\; r</span>}}^{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 3</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; min</span>}<span\; class="notranslate">\; \{</span>{\mathrm{<span\; class="notranslate">\; \&mu;</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{\mathrm{<span\; class="notranslate">\; l</span>}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \}</span>$

In the formula, l=1, 2, ..., m. m represents the number of fuzzy rules.

④ The input of each unit in the fourth layer is the output of the corresponding unit in the third layer, which is

${\mathrm{<span\; class="notranslate">\; S</span>}}_{\mathrm{<span\; class="notranslate">\; r</span>}}^{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 4</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; =</span><span\; class="notranslate">\; \{</span>{\mathrm{<span\; class="notranslate">\; the\; y</span>}}_{\mathrm{<span\; class="notranslate">\; r</span>}}^{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 3</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; \}</span>$

The output of the output layer is

${\mathrm{<span\; class="notranslate">\; the\; y</span>}}_{\mathrm{<span\; class="notranslate">\; r</span>}}^{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 4</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; =</span>\frac{{\mathrm{<span\; class="notranslate">\; the\; y</span>}}_{\mathrm{<span\; class="notranslate">\; r</span>}}^{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 3</span>}<span\; class="notranslate">\; )</span>}}{\underset{\mathrm{<span\; class="notranslate">\; r</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; m</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{\mathrm{<span\; class="notranslate">\; the\; y</span>}}_{\mathrm{<span\; class="notranslate">\; r</span>}}^{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 3</span>}<span\; class="notranslate">\; )</span>}}$

⑤ The input of each unit in the fifth layer is the output of the corresponding unit in the fourth layer, which is

${\mathrm{<span\; class="notranslate">\; S</span>}}_{\mathrm{<span\; class="notranslate">\; r</span>}}^{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 5</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; =</span><span\; class="notranslate">\; \{</span>{\mathrm{<span\; class="notranslate">\; the\; y</span>}}_{\mathrm{<span\; class="notranslate">\; r</span>}}^{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 4</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; \}</span>$

The output layer output is

$\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; w</span>}}_{\mathrm{<span\; class="notranslate">\; r</span>}}{\mathrm{<span\; class="notranslate">\; the\; y</span>}}_{\mathrm{<span\; class="notranslate">\; r</span>}}^{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 4</span>}<span\; class="notranslate">\; )</span>}$

At this point, a forward pass process is completed.

⑥ In the process of error backpropagation, the error calculation must be performed first.

For the fuzzy neural network, it is assumed that the error function of the tth sample pair is defined as:

${\mathrm{<span\; class="notranslate">\; E.</span>}}_{\mathrm{<span\; class="notranslate">\; t</span>}}<span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; 2</span>}}{<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; the\; y</span>}}_{\mathrm{<span\; class="notranslate">\; 0</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}$

In the formula: y ₀ (t) is the expected output value of the system, y(t) is the actual output value of the system, and the idea of backpropagation BP is used to supervise the learning. By adjusting the weight values of the network, the value of the error function is minimized, so that To achieve the purpose of modifying the parameters of the membership function and the network connection weight.

（其中，K为学习样本数）小于预先设定的一个极小值，即网络收敛；或学习次数小于预先设定的值，即网络无法收敛。⑦ Randomly select the next sample pair to provide to the network, and repeat the calculation process until the global error function of the network

(where K is the number of learning samples) is less than a preset minimum value, that is, the network converges; or the number of learning times is less than the preset value, that is, the network cannot converge.

⑧End of study.

Step 4: Using the optimal fuzzy neural network traffic controller to generate traffic warning information

Input the real-time collected meteorological parameters and traffic flow parameters into the trained fuzzy neural network controller to generate real-time traffic warning information, including expressway control methods (full line closure, area closure, ramp control, etc.), speed limit and safe driving spacing.

Step 5: Publication of early warning information

Using wireless communication technology, the traffic early warning information is sent to the roadside variable information board (module 2 shown in Figure 1), and the information released by the variable information board should be the traffic early warning information obtained by its downstream controller.

Although the specific implementation of the present invention has been described above in conjunction with the accompanying drawings, it does not limit the protection scope of the present invention. Those skilled in the art should understand that on the basis of the technical solution of the present invention, those skilled in the art do not need to pay creative work Various modifications or variations that can be made are still within the protection scope of the present invention.

Claims (4)

1. the highway microclimate traffic prewarning method based on fuzzy neural network is characterized in that, may further comprise the steps: the laying of traffic flow and microclimate check point; Ambiguity in definition neural network traffic controller; The training of fuzzy neural network traffic controller; Utilize optimum fuzzy neural network traffic controller to generate traffic safety driving parameter; The traffic control information issue, the concrete operations step is following:

Step 1: some traffic flows and microclimate monitoring point are set at the every spacing distance of highway trackside; Detect this road traffic delay situation and microclimate supplemental characteristic; Pass through sensor Data Fusion; Obtain road traffic delay and weather information, traffic flow and microclimate data acquisition are prepared for next step Fuzzy Neural-network Control;

Step 2: adopt fuzzy neural network to make up highway microclimate traffic controller based on the Takagi-Sugeno reasoning; Telecommunication flow information and weather information that definition step 1 collects are state variable; Input value as said controller; Definition Highway Control mode, speed limit and safe spacing value are control variable, as the output valve of said controller;

Step 3: the meteorology, traffic flow, control measure and the implementation result historical data base thereof that adopt meteorological department and vehicle supervision department; Make up the training sample of fuzzy neural network traffic controller in conjunction with expertise; Highway microclimate traffic controller is trained; Training error is reduced to predetermined threshold or reaches frequency of training, to obtain optimal controller;

Step 4: the telecommunication flow information that will gather in real time and the optimum fuzzy neural network traffic controller of microclimate information input; Generate Freeway Traffic Control scheme, comprise early warning measure, speed limit and safe spacing value to this moment traffic flow situation and weather information;

Step 5: lay the variable information plate in each traffic and the upper reaches, microclimate monitoring point, the safe driving early warning information of traffic controller output is issued.

2. the highway microclimate traffic prewarning method based on fuzzy neural network as claimed in claim 1 is characterized in that, fuzzy neural network structure highway microclimate traffic controller is divided into five-layer structure in the said step 2:

Ground floor is an input layer, and input value is the supplemental characteristic that freeway traffic flow and microclimate check point are gathered, and is expressed as x={x ₁, x ₂, x ₃..., x _n, wherein, n representes the number of input parameter, n is the integer more than or equal to 1, and x ₁, x ₂... x _nRepresent respectively traffic flow parameter be speed, flow, occupation rate and microclimate parameter can degree of opinion, temperature, humidity, air pressure, wind direction, wind speed, ST;

Input parameter is divided into 5 grades according to from small to large order, is respectively { NB is negative big, and NS is negative little, and Z zero, and PS is just little, and PB is honest }, its meaning for the relevant parameters desired value be little, less, medium, greatly, bigger; The system of the controller of telling is output as control model, speed limit and safe distance; Wherein, control model is divided into three kinds, and sealing completely, zone sealing and ring road control are promptly controlled according to traffic and meteorological condition and sailed the highway flow into; Speed limit and safe distance grade classification mode are identical with input parameter;

The second layer is the obfuscation layer, be used for representing input quantity belong to respectively NB, NS, Z, PS, the degree of membership of PB} does

In the formula, j is 1 to m _iInteger, m _iBe x _iThe fuzzy number of cutting apart, m here _i=5; c _IjAnd σ _IjCenter and the width of representing membership function respectively;

The 3rd layer is regular former piece layer, and each node is represented a fuzzy rule, and its effect is the former piece that is used for mating fuzzy rule, calculates the relevance grade a of every rule _j, formula is:

Or

$a_j={\mathrm{\&mu;}}_1^j\left(x_1\right){\mathrm{\&mu;}}_2^j\left(x_2\right)...{\mathrm{\&mu;}}_n^j\left(x_n\right);$

In the formula: a _jThe relevance grade of-fuzzy rule j;

-input x _iThe degree of membership that is under the jurisdiction of j grade; J is the integer between 1 to m, and m is the integer more than or equal to 1;

Realize normalization calculating for the 4th layer, formula is:

wherein; J is the integer between 1 to m, and m is the integer more than or equal to 1;

The layer 5 output layer, formula does

$y=\underset{r=1}{\overset{m}{\mathrm{\&Sigma;}}}{w}_r\stackrel{\&OverBar;}{a_r},r=\mathrm{1,2},...,m$

Wherein, w _rBe connection weight, r is the integer between 1 to m, and m is the integer more than or equal to 1.

3. the highway microclimate traffic prewarning method based on fuzzy neural network as claimed in claim 2 is characterized in that, based on the training of the microclimate traffic controller of fuzzy neural network, particular content is following in the said step 3:

Selection of training

Training sample is considered to be made up of three parts: first, the historical data of meteorological part, freeway traffic regulation department, comprise historical meteorologic parameter information, traffic flow parameter information, the early warning information of taking at that time and issue after effect; Second portion, traffic engineering domain expert's questionnaire is through being provided with different meteorologies and traffic flow sight, the early warning measure that the inquiry expert possibly adopt; Third part after this system builds up, is added into historical data base with the traffic prewarning scheme implementation effect under different meteorologies, the traffic flow situation;

Training algorithm

The parameter that network need be learnt is the central value c of second layer membership function in the step 2 _IjAnd width cs _IjAnd layer 5 network connection weight w _r, the learning algorithm of this network is selected backpropagation BP algorithm; The BP algorithm is made up of forward-propagating and two processes of error back propagation; In the forward-propagating process, input information is successively handled through latent layer unit from input layer, after all latent layers, passes to output layer; In the process that hidden layer is successively handled, the neuronic state of each layer only exerts an influence to the following neuronic state of one deck; At output layer reality output and desired output are compared; If the difference of actual output and desired output no longer within the tolerance interval, then changes back-propagation process over to, the error between actual value and the desired output is returned along original connecting path; Through revising the neuronic connection weight of each layer error is reduced; And then changing the forward-propagating process over to, repeated calculation like this is till error is less than setting value.

4. the highway microclimate traffic prewarning method based on fuzzy neural network as claimed in claim 3 is characterized in that, the learning algorithm of said network is selected backpropagation BP algorithm, and step is following:

1. establish x _kBe input vector, x _k=(x ₁, x ₂..., x _n), k is the integer between 1 to K, in the formula: K is a number of samples, and n represents the number of characteristic parameter; The output vector of corresponding travel pattern is y _k, initialization network weight, threshold value;

2. each unit of the second layer is input as

$S_{\mathrm{ij}}^{\left(2\right)}=x_i,$

In the formula, x _iThe input of expression system, i.e. traffic flow parameter and meteorologic parameter value; I is the integer between 1 to n, and n represents the number of characteristic parameter; J is 1 to m _iBetween integer, m _iRepresent the fuzzy number of cutting apart of i characteristic parameter;

Transport function between the ground floor and the second layer is membership function:

${\mathrm{\&mu;}}_i^j\left(x_i\right)=\exp [{(x_i-c_{\mathrm{ij}})}^2/{\mathrm{\&sigma;}}_{\mathrm{ij}}^2]$

Then second layer unit is output as:

$y^{\left(2\right)}=\left\{{\mathrm{\&mu;}}_i^j\left(x_i\right)\right\}$

3. the input of the 3rd layer of each unit is the output of corresponding each unit of the second layer, for

$S_r^{\left(3\right)}=\left\{{\mathrm{\&mu;}}_i^r\left(x_i\right)\right\}$

Output layer is output as

$y_r^{\left(3\right)}=\min \left\{{\mathrm{\&mu;}}_i^l\left(x_i\right)\right\}$

In the formula, l is the integer between 1 to m, and m representes number of fuzzy rules;

4. the input of the 4th layer of each unit is the output of the 3rd layer of corresponding each unit, for

$S_r^{\left(4\right)}=\left\{y_r^{\left(3\right)}\right\}$

Output layer is output as

$y_r^{\left(4\right)}=\frac{y_r^{\left(3\right)}}{\underset{r=1}{\overset{m}{\mathrm{\&Sigma;}}}{y}_r^{\left(3\right)}}$

5. the input of each unit of layer 5 is the output of the 4th layer of corresponding each unit, for

$S_r^{\left(5\right)}=\left\{y_r^{\left(4\right)}\right\}$

Output layer is output as

$y=w_r{y}_r^{\left(4\right)}$

So far accomplish a forward pass process;

6. in the error back propagation process, at first to carry out Error Calculation,

For fuzzy neural network, suppose t the error function E that sample is right _tBe defined as:

$E_t=\frac{1}{2}{(y_0\left(t\right)-y\left(t\right))}^2$

In the formula: y ₀(t) be system's desired output, y (t) is system's real output value, and t is the integer more than or equal to 1, the label of expression sample;

Backpropagation BP thought is used to supervised learning, through each weighted value of adjustment network, makes the error function value minimum, thereby reaches the purpose of revising membership function parameter and network connection power;

7. the next sample of picked at random is to offering network; The double counting process; Until network global error function less than predefined minimal value, i.e. a network convergence; Or learn number of times less than predefined value, promptly network can't be restrained; Wherein, K is the learning sample number;

8. finish study.

Images