Photochemical oxidant prediction system
The photochemical oxidant prediction system uses neural networks to analyze atmospheric data for precise, early warnings, addressing the limitations of human-dependent systems and improving warning accuracy and timeliness.
Patent Information
- Application Number
- JP2024047390
- Authority / Receiving Office
- JP · JP
- Patent Type
- Applications
- Current Assignee / Owner
- Filing Date
- 2024-03-24
- Publication Date
- 2025-10-06
AI Technical Summary
Existing photochemical oxidant warning systems issue late warnings and rely heavily on human judgment, leading to psychological burden and potential inaccuracies.
A photochemical oxidant prediction system utilizing observation devices to collect atmospheric data, employing a neural network for predictive analysis of photochemical oxidants based on multiple atmospheric parameters, including NOX, VOCs, hydrocarbons, wind direction, and meteorological factors, to provide advanced warnings.
Enables accurate, timely prediction of photochemical oxidant levels, reducing unnecessary outings and alleviating the workload on warning issuers by leveraging machine learning and multiple data inputs.
Smart Images
Figure 2025147165000001_ABST
Abstract
Description
[Technical Field]
[0001] The present invention relates to a photochemical oxidant prediction system for predicting photochemical oxidants. [Background technology]
[0002] Photochemical oxidants are produced when nitrogen oxides and hydrocarbons contained in factory exhaust smoke and automobile exhaust gases react chemically with ultraviolet rays, and are produced when factors such as temperatures above 25°C, the presence or absence of sunlight, stagnant air, and atmospheric stability occur.
[0003] Conventionally, when issuing such photochemical oxidant warnings or alerts, consideration is given to man-made factors such as factory smoke and automobile exhaust fumes, as well as meteorological observation results in each city, town, or village. When the observed value of photochemical oxidants reaches 0.12 ppm or higher and it is expected that this state will continue for a specified period of time or longer, a warning is issued; when it reaches 0.24 ppm or higher, a warning is issued; and when it reaches 0.40 ppm or higher, a severe warning is issued (see the non-patent literature below). [Prior art documents] [Non-patent literature]
[0004] [Non-Patent Document 1] https: / / www.env.go.jp / air / post_99_00003.html (Ministry of the Environment, Air Quality and Automobile Measures website: confirmed on February 15, 2024) Summary of the Invention [Problem to be solved by the invention]
[0005] However, since such photochemical oxidant warnings are issued when it is predicted that the level will remain above 0.12 ppm for a certain period of time, the warnings may be issued late for citizens who are already out and about. Furthermore, because the judgment of whether the observed level will continue is based on the experience of the person in charge, the predictions may be wrong, which places a psychological burden on the person in charge.
[0006] Therefore, in order to solve the above problems, an object of the present invention is to provide a photochemical oxidant prediction system that can automatically predict photochemical oxidants. [Means for solving the problem]
[0007] In other words, in order to solve the above-mentioned problems, the photochemical oxidant prediction system of the present invention comprises an observation device that observes the values of at least three items of NOX, VOCs, hydrocarbon-based substances, photochemical oxidants, wind direction, wind speed, humidity, solar radiation, and temperature in the atmosphere within a specific area over a predetermined period of time; a prediction means that uses the values of at least three items of NOX, VOCs, hydrocarbon-based substances, photochemical oxidants, wind direction, wind speed, humidity, solar radiation, and temperature in the atmosphere observed by the observation device as input values and predicts the state of photochemical oxidants after a predetermined period of time using a neural network; and an output means that outputs the results predicted by the prediction means.
[0008] This configuration makes it possible to predict the photochemical oxidant situation in advance based on past data, which will encourage people to refrain from unnecessary outings and reduce the human burden on those in charge of issuing warnings, etc.
[0009] Furthermore, in such an invention, if there are multiple observation points within the specific region, all values of at least three of the atmospheric NOX, VOCs, hydrocarbon substances, photochemical oxidants, wind direction, wind speed, humidity, solar radiation, and temperature at all observation points are used as input values for the neural network.
[0010] By configuring it in this way, when there are multiple observation points for meteorological information in an area where a photochemical oxidant warning is issued, the accuracy of the prediction can be improved compared to when a single value is entered for each item.
[0011] Furthermore, when making predictions using the neural network, the input items NOX, VOC, hydrocarbon substances, and photochemical oxidants are weighted more heavily than the other input items.
[0012] With this configuration, substances that cause photochemical oxidants are weighted more heavily, making it possible to predict a situation that is closer to the actual situation of photochemical oxidants.
[0013] Furthermore, when inputting values for input items into the neural network, observed values within the specific area and observed values in the surrounding areas of the specific area are input.
[0014] This configuration makes it possible to predict photochemical oxidants over a wide area, including the conditions in the surrounding areas. [Effects of the Invention]
[0015] According to the present invention, an observation device that observes at least three atmospheric values within a specific region for a predetermined period of time (NOX, VOCs, hydrocarbon-based substances, photochemical oxidants, wind direction, wind speed, humidity, solar radiation, and temperature) is included; a prediction means that uses a neural network to predict the photochemical oxidant status after a predetermined period of time using the at least three atmospheric values (NOX, VOCs, hydrocarbon-based substances, photochemical oxidants, wind direction, wind speed, humidity, solar radiation, and temperature) observed by the observation device as input values; and an output means that outputs the results predicted by the prediction means. This allows the photochemical oxidant status to be predicted in advance based on past data, encouraging people to refrain from unnecessary outings and reducing the burden on personnel responsible for issuing warnings, etc. [Brief explanation of the drawings]
[0016] [Figure 1] Schematic diagram of a photochemical oxidant prediction system according to one embodiment of the present invention. [Figure 2] Functional block diagram of the same configuration [Figure 3] A diagram showing the configuration of a neural network in the same form. [Figure 4] FIG. 10 shows examples of input items in the same form. [Figure 5] Flowchart in the same format DETAILED DESCRIPTION OF THE INVENTION
[0017] Hereinafter, a photochemical oxidant prediction system 1 according to an embodiment of the present invention will be described with reference to the drawings.
[0018] As shown in Figures 1 and 2, the photochemical oxidant prediction system 1 in this embodiment is configured to include observation devices 2 arranged within a specific area such as a prefecture, city, town, or village, transmission means 21 that transmits observation values observed by these observation devices 2, prediction means 3 that collects the observation values transmitted from this transmission means 21 and predicts the photochemical oxidant situation using a neural network, and output means 4 (see Figure 2) that outputs the predicted values of photochemical oxidants predicted by this prediction means 3. In this way, the neural network is used to perform machine learning on past actual observation values and photochemical oxidant values, making it possible to predict the photochemical oxidant situation at a predetermined time, such as in the afternoon of that day, based on the current observation values. This embodiment will be described in detail below.
[0019] First, the observation device 2 observes atmospheric NOX (nitrogen oxides) such as nitric oxide and nitrogen dioxide, VOCs (volatile organic compounds), hydrocarbon substances such as non-methane hydrocarbons and total hydrocarbons, photochemical oxidants, suspended particulate matter, fine particulate matter, wind direction, wind speed, humidity, solar radiation, temperature, etc. Note that while these values are observed here, other values such as carbon monoxide, methane, suspended particulate matter β, carbon dioxide, rainfall, ultraviolet radiation, and radiation balance may also be observed.
[0020] These observation devices 2 are observed by a plurality of measuring instruments installed in the station 20, and observe each value by taking in air from outdoors, or by using a rain gauge, pyranometer, or the like installed outside the station 20. The values observed by the observation devices 2 in this way are transmitted to a remote base unit 30 using a transmission means 21, as shown in Fig. 2. The observation devices 2 are installed at multiple locations within an area such as each prefecture or city, town, or village, and at each location, NOX, VOCs, hydrocarbon-based substances, photochemical oxidants, wind direction, wind speed, humidity, solar radiation, temperature, and the like are observed.
[0021] When multiple observation devices 2 transmit observation values for the same multiple items (e.g., atmospheric NOX, VOCs, hydrocarbon substances, photochemical oxidants, wind direction, wind speed, humidity, solar radiation, temperature, etc.), the master unit 30 uses all of these values as input item values and predicts photochemical oxidants using a neural network, as shown in FIG. 3 . While it is possible to predict each item by using a single value, such as the maximum value, as the input item value, in this example, all observation values are input so that photochemical oxidants can be predicted using as much information as possible. Note that the values observed by these observation devices 2 are those observed a predetermined time ago (e.g., within three hours), and observation values from observation devices 2 within a predetermined distance, including neighboring prefectures (see FIG. 1 ), are also used to predict the occurrence of photochemical oxidants over a wide area.
[0022] This prediction means 3 takes as input three or more items, namely, at least two of NOX, VOC, hydrocarbon substances, photochemical oxidants, wind direction, wind speed, humidity, solar radiation, and temperature in the atmosphere, which are anthropogenic factors, and at least one of wind direction, wind speed, humidity, solar radiation, and temperature, which are natural factors, and predicts the value of photochemical oxidants using a neural network such as that shown in Figure 3. In this case, since photochemical oxidants are generated based on NOX, VOC, and hydrocarbon substances in the atmosphere, the values of these sources of photochemical oxidants are weighted heavily when predicting the value of photochemical oxidants.
[0023] When making predictions using this neural network, multiple past input items and actual photochemical oxidant values at a predetermined time on that day are machine-learned, and weights for the intermediate layer are determined by repeating this process.The observed values on that day are then used as input items, and the photochemical oxidant values at a predetermined time on that day are predicted using the machine-learned weights.
[0024] Furthermore, when the final prediction results are output, the photochemical oxidant status after a specified time is predicted using multiple different prediction methods, and these prediction results are integrated to predict the final photochemical oxidant status. Here, the prediction is made using an ensemble model, and in the first stage, the same training data is fed to models of multiple prediction methods to make predictions, and in the second stage, these values are aggregated to output the final results.
[0025] The output means 4 outputs the photochemical oxidant situation predicted by the prediction means 3, and outputs the situation as to whether or not a photochemical oxidant warning, alert, or serious alert should be issued after a predetermined time, or the photochemical oxidant value after a predetermined time.
[0026] The information output by the output means 4 can then be used to issue photochemical oxidant warnings, alerts, serious alerts, etc. for that day.
[0027] Next, the prediction method in the photochemical oxidant prediction system 1 configured as above will be described with reference to the flowchart of FIG.
[0028] First, when predicting photochemical oxidants, NOX, VOCs, hydrocarbon-based substances, photochemical oxidants, wind direction, wind speed, humidity, solar radiation, and temperature in the atmosphere at a predetermined time are observed (step S1), and transmitted to the parent unit 30 via the transmitting means 21 (step S2). Note that although it is possible to observe all of these items, it is preferable to be able to observe values for a total of three items: at least two items from among NOX, VOCs, hydrocarbon-based substances, and photochemical oxidants, and at least one item from among wind direction, wind speed, humidity, solar radiation, and temperature.
[0029] The master unit 30 then uses these transmitted observation values as input items and predicts the photochemical oxidant situation after a predetermined time using a neural network (step S3). Note that these input items may be observed multiple times within the predetermined time, but all of these observation values are input as input items to the neural network. Furthermore, when these observation values are input, not only are observation values within the prefecture or municipality in question input, but also observation values from neighboring prefectures within a predetermined distance are input.
[0030] Then, using the observed values as input items, a weighted neural network is used to predict the state of photochemical oxidants after a predetermined time, and the predicted state is output by the output means 4 (step S5).
[0031] Based on the output values, officials at prefectures, cities, towns, and villages will be able to issue warnings about photochemical oxidants.
[0032] As described above, the embodiment includes an observation device 2 that observes at least three of atmospheric NOX, VOCs, hydrocarbon-based substances, photochemical oxidants, wind direction, wind speed, humidity, solar radiation, and temperature within a specific region over a predetermined period of time; a prediction means 3 that uses a neural network to predict the photochemical oxidant status after a predetermined period of time using the at least three of atmospheric NOX, VOCs, hydrocarbon-based substances, photochemical oxidants, wind direction, wind speed, humidity, solar radiation, and temperature observed by the observation device 2 as input items; and an output means 4 that outputs the results predicted by the prediction means 3. This allows the photochemical oxidant status to be predicted in advance based on past data, thereby encouraging people to refrain from unnecessary outings. It also reduces the workload of personnel responsible for issuing warnings and other alerts.
[0033] The present invention is not limited to the above-described embodiment, but can be implemented in various forms.
[0034] For example, in the above embodiment, the photochemical oxidant values for the afternoon of that day are predicted based on the observed values at a predetermined time in the morning of that day. In this case, the photochemical oxidant values for the predicted afternoon time and the photochemical oxidant values predicted in the morning may be fed back to train the weighting in the neural network.
[0035] In addition, in the above embodiment, the input items are at least two items of man-made factors such as NOX, VOCs, hydrocarbon substances, and photochemical oxidants, and at least one item of natural factors such as wind direction, wind speed, humidity, solar radiation, and temperature in the area, but predictions may also be made using three items of man-made factors and natural factors such as solar radiation and temperature. [Explanation of symbols]
[0036] 1. Photochemical oxidant prediction system 2. Observation equipment 20, 20a... Station building 21. Transmitting device 3. Prediction methods 30... Base unit 4. Output means
Claims
1. an observation device for observing values of at least three items of NOx, VOC, hydrocarbon-based substances, photochemical oxidants, wind direction, wind speed, humidity, solar radiation, and temperature in the atmosphere within a specific area within a predetermined period of time; a prediction means for predicting the state of photochemical oxidants after a predetermined time using a neural network, using as input values at least three of the atmospheric NOX, VOC, hydrocarbon substances, photochemical oxidants, wind direction, wind speed, humidity, solar radiation, and temperature observed by the observation device; and an output means for outputting the results predicted by the prediction means. A photochemical oxidant prediction system comprising:
2. The photochemical oxidant prediction system of claim 1, wherein when there are multiple observation points within the specific region, all observed values of at least three of atmospheric NOx, VOCs, hydrocarbon substances, photochemical oxidants, wind direction, wind speed, humidity, solar radiation, and temperature at all observation points are used as input values for the neural network.
3. The photochemical oxidant prediction system of claim 1, wherein when making predictions using the neural network, the input items NOx, VOCs, hydrocarbon substances, and photochemical oxidants are weighted more heavily than other input items.
4. 2. The photochemical oxidant prediction system according to claim 1, wherein the input items are observation values within the specific area and observation values in the surrounding areas of the specific area.