JPH1073670A - Weather discrimination method and device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a weather discrimination method and device capable of discriminating local weather state with high accuracy by utilizing a hierarchy type neural network capable of nonlinear processing in discriminating weather state. SOLUTION: Constituted are a weather sensor part 1A detecting and outputting the weather state such as temperature, pressure, road temperature, etc., a trend output part 1B outputting the trend of past weather state, a weather judgment part 2 for discriminating the weather state using a hierarchy type neural network based on the signal input from the weather sensor part 1A and the trend output part 1B and a display 3 indicating the weather state according to the discriminated result of the weather judgment part 2.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a meteorological discrimination method capable of accurately discriminating a peculiar weather condition for each location from factors closely related to changes in weather conditions such as air temperature, air pressure, and road surface temperature. And its equipment.

[0002]

2. Description of the Related Art In order to ensure safe and smooth road traffic, factors related to the road driving environment such as road surface conditions, visibility, traffic flow, and weather conditions are accurately detected, and the information is used for road use. It is desired to provide the information to people and road managers in a form suitable for the purpose. There is a correlation between a change in the state of the road surface, visibility, traffic flow, and the like, and a change in the weather state, and the detection of the weather state plays an important role in a road driving environment and the like.

[0003] The weather information that the user particularly wants to know in the road driving environment and the like is the weather condition of the road driving environment where the road user is currently located or will be located in the very near future. Determination of weather conditions is extremely important.

Conventionally, as a method for determining such a weather condition, linearity is assumed between a weather condition and factors closely related to changes in the weather condition such as air temperature, barometric pressure, road surface temperature, etc. It is common to make a determination using a statistical analysis method represented by analysis.

[0005]

However, the relationship between the change in weather conditions and the temperature, pressure, road surface temperature, etc., is not necessarily linear, but includes non-linear elements. In particular, inland areas such as mountainous areas and foothills, there are many phenomena in which the weather conditions are completely different even though the distance is not so large. Conventional statistical analysis methods cannot sufficiently cope with such local changes in weather conditions.

SUMMARY OF THE INVENTION The present invention has been made to solve the above-described problems. A hierarchical neural network capable of performing non-linear processing is used for determining weather conditions, and local weather conditions can be determined. It is another object of the present invention to provide a weather discriminating method and a device capable of performing discrimination with high accuracy.

[0007]

In order to solve the above-mentioned problems, a weather discriminating method according to the present invention comprises a hierarchical structure of current weather conditions such as air temperature, air pressure, road surface temperature and the like, and past transition trends of the weather conditions. It is characterized in that it is inputted to an input layer of a neural network of the type to determine weather conditions. It is desirable to use temperature and pressure as the past transition tendency.

The weather discriminating apparatus according to the present invention includes a weather sensor unit for detecting and outputting weather conditions such as air temperature, air pressure, and road surface temperature, and a transition trend output unit for outputting past transition trends of the weather conditions. A weather discriminating unit that discriminates a weather condition using a hierarchical neural network based on signals input from the weather sensor unit and the transition tendency output unit, and a weather condition based on the discrimination result of the weather discriminating unit. A display unit for performing display. It is desirable to use temperature and pressure as the past transition tendency.

[0009] In the case of the method and the apparatus having the above-described configuration, the neural network is trained in advance using teacher data, and the weather condition is discriminated by the neural network after the learning, so that nonlinear elements are included. The weather condition can be accurately determined.

[0010]

Embodiments of the present invention will be described below with reference to the drawings. FIG. 1 is a block diagram of an embodiment of a weather discriminating apparatus according to the present invention configured by applying the method of the present invention. This weather discriminating apparatus includes a weather sensor section 1A, a past transition tendency output section 1B, and a weather discriminating section 2.
And a display unit 3. The weather sensor unit 1A includes at least a plurality of sensors for detecting factors closely related to a change in weather conditions, such as a temperature sensor 11, an air pressure sensor 12, and a road surface temperature sensor 13. In addition, since not only the current state, but also the past transition tendency is considered to have a deep relationship with the change in the weather state including the above-described nonlinear element, in the present invention, the temperature, pressure, A physical quantity such as a road surface temperature is processed by averaging or the like, and the obtained data is used as a past transition tendency input 1B which is output to the weather discriminating unit 2 as a signal indicating the past transition tendency.

As shown in detail in FIG. 2, the weather discriminator 2 is constituted by a hierarchical neural network including an input layer 21, an intermediate layer 22, and an output layer 23.
Each signal from the weather sensor unit 1 is input to the input layer 21, the weather state is determined based on the input signal, and the result is output from the output layer 23 to the display unit 3.

In order to perform the discrimination, it is necessary to provide an input teacher signal and an output teacher signal to a neural network for learning, and set a coupling coefficient of a synapse connecting neurons of each hierarchy to an optimum value. . With this learning,
The relationship between input and output is established in the network, and the weather can be determined. At this time, if learning is performed using the teacher data for the place where the weather condition is to be determined, a network incorporating different weather condition habits for each location is constructed, so that local weather condition can be determined. Becomes

As the learning method, for example, a learning algorithm such as a back propagation method is used, and an input teacher signal 24 composed of actual temperature, atmospheric pressure, road surface temperature, past transition tendency, and the like, A data set of the output teacher signal 25 that gives the actual weather condition for the signal is prepared as a teacher signal, and the coupling coefficient of the synapse connecting the neurons of each layer may be determined so that the discrimination error 26 is minimized.

As described above, the current temperature, atmospheric pressure, road surface temperature, and signals from the past transition tendency output unit 1B actually detected by the weather sensor unit 1A are input to the input layer 21 of the neural network learned by the teacher data. By doing so, the weather condition determined from the output layer 23 is output. The weather conditions to be discriminated include “fine weather”, “cloudy weather”, “rainfall”, “snowfall”, “mist”, and the like. The display unit 3 outputs the determination result sent from the weather determination unit 2 in a desired display format to a monitor screen, a printer, or the like for display.

[0015]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS A concrete embodiment in which the weather condition is determined by using the weather determination device having the structure shown in FIGS. 1 and 2 and actual measurement data collected in an actual field will be described below. FIG. 3 shows actual measurement data serving as learning data, and includes actual measurement data 31 consisting of air pressure, air temperature, and road surface temperature.
And a corresponding change state of the actual weather condition 32. The actual measurement data 31 is obtained by collecting the air temperature, the atmospheric pressure, and the road surface temperature every 15 minutes, and the total number of the samples is 163 samples from December 19 to December 21, 1995. For weather condition 32,
We focused on two events, "no snowfall" and "with snowfall".

First, from among the 163 samples shown in FIG.
Twenty samples in the case of snowfall and 20 samples in the case of no snowfall are taken out, for a total of 40 samples, and learning of the neural network of FIG. 2 is performed using the sample data as teacher data. The area indicated by reference numeral 33 in FIG.
Each sample position is indicated by a black vertical bar line.

As described above, factors related to a change in weather conditions input to the neural network include:
In addition to the air temperature, the atmospheric pressure, and the road surface temperature, these past trends are also considered important information. Regarding this transition tendency, if the actual measurement data 31 in FIG. 3 shows that the temperature and the road surface temperature show almost the same transition tendency, the past transition tendency of the road surface temperature is unnecessary, and the temperature and the atmospheric pressure are plotted. By using the sampling method of 5, the average value every 30 minutes is retroactive to the past 6 hours, and 12 pieces of input information are simultaneously provided. That is, in the case of the present embodiment, "with snowfall" and "without snowfall" are determined based on the actual measurement data of the temperature, the atmospheric pressure, and the road surface temperature, and the past transition trends of the temperature and the atmospheric pressure.

Next, the results of the actual weather condition discrimination by the weather discriminator learned as described above will be described. By inputting the temperature, pressure, road surface temperature, and past transition trends of temperature and pressure in FIG. 3 into the input layer 21 of the learned neural network (weather discriminating unit 2), the weather condition can be discriminated. went. FIG. 4 shows the result of determining the weather condition of all the 163 samples shown in FIG. In FIG. 4, reference numeral 41 denotes an actual weather condition, and reference numeral 42 denotes a determination result obtained by the weather determination device of the present invention. Further, Table 1 below shows a discrimination correct answer rate in which the degree of coincidence between the actual weather condition and the discrimination result of the weather condition according to the present invention is represented by a ratio to the total number of samples.

[0019]

[Table 1]

As shown in Table 1, a high discrimination accuracy rate of 95.1% is obtained with the weather discriminator according to the present invention, compared with 82.2% when no past transition tendency information is input. Was. From this, it can be seen that extremely accurate weather discrimination has been performed. In addition, most of the erroneously determined locations in FIG. 4 correspond to portions that are hard to say in either the “with snowfall” state or the “without snowfall” state.

Furthermore, focusing on the discrimination result at the boundary portion where the weather condition changes in FIG.
There is a tendency to make a distinction slightly earlier in time, indicating that a distinction having a predictive effect has been made. Therefore,
If this prediction result is used, it is possible to take measures as soon as possible prior to a change in the weather condition, which is extremely useful in actual road traffic and the like.

In addition to the above-described embodiment, a change in data increase / decrease may be obtained by differentiating data before and after, and a change in a predetermined time in the past may be input to the weather discriminating unit 2 as a change tendency. .

[0023]

As described above, claims 1 and 2
When according to the described weather determination method, the current weather condition such as temperature, pressure, road surface temperature, and the past trend of the weather condition are input to the input layer, and the weather condition is determined using a hierarchical neural network. Since the determination is performed, it is possible to accurately determine the weather condition including a non-linear element based on the temperature, the atmospheric pressure, the road surface temperature, the past transition tendency, and the like. Further, by performing learning using the learning data of the area, local weather discrimination can be performed, and a weather discrimination method that is rich in usefulness and adaptability can be provided.

According to the weather discriminating apparatus according to the third and fourth aspects, a weather sensor section for detecting and outputting weather conditions such as air temperature, air pressure, road surface temperature and the like, and a transition for outputting past trends of the weather conditions. A trend output unit, a weather discriminating unit that discriminates a weather condition using a hierarchical neural network based on a signal sent from the weather sensor unit, and a weather condition based on a discrimination result of the weather discriminating unit. Since the display unit includes a display unit for performing display, it is possible to accurately determine a weather condition including a non-linear element based on the temperature, the atmospheric pressure, the road surface temperature, the past transition tendency, and the like. Further, by performing learning using the learning data of the area, local weather discrimination can be performed, and a weather discriminating apparatus that is highly useful and adaptable can be provided.

[Brief description of the drawings]

FIG. 1 is a block diagram of an embodiment of a weather discriminator according to the present invention configured by applying the method of the present invention.

FIG. 2 is a configuration diagram of a neural network constituting a weather discrimination unit.

FIG. 3 is a diagram showing actually measured data serving as learning data.

FIG. 4 is a diagram illustrating a weather determination result by the weather determination device of FIG. 1;

FIG. 5 is an explanatory diagram showing past transition trends.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1A Meteorological sensor part 1B Past transition tendency output part 2 Weather discrimination part (neural network) 3 Display part 11 Temperature sensor 12 Barometric pressure sensor 13 Road surface temperature sensor 14 Past transition tendency 21 Input layer 22 Middle layer 23 Output layer 24 Input learning signal 25 output learning signal 26 discrimination error

 ──────────────────────────────────────────────────続 き Continued on the front page (72) Inventor Noboru Sugie 35-3, 1-203, Yagotohonmachi, Showa-ku, Nagoya-shi, Aichi Prefecture Inside the Miwa factory of Denki Kogyo Co., Ltd.

Claims (4)

[Claims] 1. A weather condition is determined by inputting a current weather condition such as an air temperature, a barometric pressure, a road surface temperature and the like and a past transition tendency of the weather condition to an input layer of a hierarchical neural network. Weather determination method. 2. The method according to claim 1, wherein
A weather discriminating method using temperature and pressure as the past transition tendency. 3. A weather sensor unit for detecting and outputting weather conditions such as air temperature, barometric pressure, road surface temperature, and the like; a transition trend output unit for outputting a past transition trend of the weather condition; A weather discriminator for discriminating a weather condition using a hierarchical neural network based on a signal output from the output unit; and a display unit for displaying a weather condition in accordance with the discrimination result of the weather discriminator. A weather discriminator characterized by the following. 4. The weather discriminating apparatus according to claim 3,
A weather discriminator characterized by using temperature and pressure as the past transition tendency.