Air pollution prediction method and system based on deep learning

Abstract

The invention discloses an air pollution prediction method and system based on deep learning, and relates to the technical field of air pollution prediction. Comprising the steps that building information and construction information stored in a terminal storage device are obtained and read, the building information comprises construction material information and building height information marked in a construction blueprint, and the construction information comprises construction progress information marked in a construction schedule; at least one monitoring point wind power and wind direction sensor is arranged at a monitoring point, and at least eight prediction points are arranged on the periphery of the monitoring point with the monitoring point as the center. According to the method, the material dust discharge amount M and the current building height P are introduced into the input sample feature set of the deep learning model, and the introduction purpose is to consider the enhancement influence of the construction material on the dust concentration of the monitoring point in the interval time period and the weakening influence of the building height on the wind power in the interval time period, so that the prediction accuracy is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of air pollution prediction, and specifically to an air pollution prediction method and system based on deep learning. Background Technique

[0002] In air quality prediction, deep learning, as a powerful machine learning technology, is gradually emerging. By simulating the structure and function of the human brain neural network, deep learning can automatically learn complex patterns from a large amount of data, which gives it significant advantages in dealing with complex problems such as air quality prediction.

[0003] Chinese invention patent, application publication number CN114626580A discloses a method for predicting the dust concentration at a construction site. By inputting the real-time environmental information of the detection point to be predicted into the dust concentration prediction model, the real-time dust concentration correction data of the detection point can be obtained, and then by processing the dust concentration correction data, the dust concentration prediction data of the detection point to be predicted can be obtained. This invention can predict the dust concentration at a construction site according to the real-time environmental information of the construction site.

[0004] Another example is that Chinese invention patent, application publication number CN116756549A discloses an air pollution concentration prediction method and system based on discrete wavelet and deep learning, including the following steps: S1. By performing missing value processing on the original input data and performing discrete wavelet transform on the air pollution concentration sequence to be predicted, extracting the periodic information of air pollution, and performing tensor stacking and normalization processing on the obtained periodic information; S2. By using a convolutional layer for feature extraction to obtain a feature matrix; S3. Inputting the feature matrix into an encoder, the encoder encodes the feature matrix into the final cell state Ce, and transmits the information in the feature matrix to the decoder. After receiving it, the decoder reconstructs its internal features as the output of the decoder. After the linear transformation of the full connection layer, hd is the final predicted value of the model, which is applied to the technical field of air pollution prediction. It can provide support for the prediction and prevention and control of overall air pollution.

[0005] The above solutions solve the technical problems proposed in their background technology, but in actual applications, the above solutions still have certain problems, such as:

[0006] The existing air pollution prediction at construction sites only establishes a prediction model through some historical data to predict the pollution in the detection area, without considering the impact of construction progress and construction materials on the pollution in the monitoring area, resulting in a decrease in prediction accuracy;

[0007] In an actual environment, the strength and direction changes of wind can significantly alter the dust diffusion pattern. Strong winds can quickly blow up the dust on the ground, while weak winds may cause dust to accumulate in local areas. The change in wind direction also leads to uneven distribution of dust concentration among different regions. Ignoring these factors will result in a large deviation between the prediction results and the actual situation, thus reducing the reliability of the overall prediction. Summary of the Invention

[0008] The purpose of the present invention is to provide an air pollution prediction method and system based on deep learning to solve the problems raised in the above background technology.

[0009] To achieve the above purpose, the present invention provides the following technical solutions: An air pollution prediction method and system based on deep learning, including:

[0010] Obtain the building information and construction information stored in the terminal storage device, and read the building information and construction information. Among them, the building information includes the construction material information and building height information marked in the construction blueprint, and the construction information includes the construction progress information marked in the construction schedule;

[0011] Set at least one monitoring point wind force and wind direction sensor at the monitoring point. Centered on the monitoring point, at least eight prediction points are set outside the monitoring point, and at least one prediction point wind force and wind direction sensor is set at each prediction point;

[0012] Monitor the wind force and wind direction at the prediction points through the prediction point wind force and wind direction sensors set at the prediction points to obtain the wind force and wind direction data at the prediction points, and monitor the wind force and wind direction at the monitoring point through the monitoring point wind force and wind direction sensors set at the monitoring point to obtain the wind force and wind direction data at the monitoring point;

[0013] Calculate the time when the wind reaches the monitoring point through the time model to adjust the time Tn when the wind reaches the monitoring point in the input sample feature set;

[0014] Calculate the predicted wind force when the wind reaches the monitoring point through the predicted wind force model to adjust the predicted wind force in the input sample feature set;

[0015] Calculate the material dust emission amount and the current building height to adjust the material dust emission amount and the current building height in the input sample feature set;

[0016] Obtain the current dust concentration data at the monitoring point and input the current dust concentration data into the input sample feature set to predict the dust concentration at the monitoring point position after time Tn;

[0017] Establish a deep learning model, and the deep learning model is one of a feedforward neural network and a recurrent neural network;

[0018] Input the data obtained at several intervals into the sample feature set for data input of the deep learning model. After obtaining the Tn time corresponding to the interval time through the deep learning model, predict the dust concentration. Compare the predicted dust concentration with the actual dust concentration data at the time monitoring point location, and optimize the deep learning model through the optimization function to improve the accuracy of the prediction result.

[0019] Furthermore, the method for establishing the time model is as follows:

[0020] The first step: Determine the target prediction point according to the wind direction, the positions of the prediction point and the monitoring point.

[0021] The second step: Calculate the straight-line distance between the wind direction and wind speed sensors at the prediction point within the target prediction point and the wind direction and wind speed sensors at the monitoring point through the satellite map system.

[0022] The third step: Detect the wind force and wind direction at the target prediction point through the wind direction and wind speed sensors at the prediction point within the target prediction point every t1 time interval.

[0023] The fourth step: Input the wind force and wind direction data obtained by the wind direction and wind speed sensors at the prediction point within the target prediction point into the numerical weather prediction model every t1 time point to obtain the time Tn when the wind reaches the monitoring point.

[0024] Furthermore, in the first step, the method for determining the target prediction point is as follows:

[0025] Determine the wind direction through the wind direction and wind speed sensors set at the monitoring point or the prediction point, and determine the positions of several wind direction and wind speed sensors at the prediction points that are downwind according to the wind direction.

[0026] In the two-dimensional coordinate system, connect all the determined wind direction and wind speed sensors at the prediction points that are downwind and the wind direction and wind speed sensors at the monitoring point.

[0027] Calculate the angles between the wind direction and the connections of all the wind direction and wind speed sensors at the prediction points that are downwind and the wind direction and wind speed sensors at the monitoring point, and determine the position of the wind direction and wind speed sensor at the prediction point with the smallest angle as the target prediction point.

[0028] Furthermore, in the fourth step, if the sum of the time Tn2 when the wind reaches the monitoring point at the later time point and the interval time t1 is less than or equal to the time Tn1 when the wind reaches the monitoring point at the previous time point for adjacent interval time points, then the time when the wind reaches the monitoring point for the adjacent interval time points is all Tn2.

[0029] Furthermore, the method for establishing the predicted wind force model is as follows:

[0030] Determine the wind distance attenuation coefficient and the wind direction attenuation coefficient , and then establish a predicted wind power model through the formula Build a predicted wind power model.

[0031] Furthermore, the determination formula of the wind distance attenuation coefficient is: ;

[0032] Wherein, is the straight-line distance between the wind power and wind direction sensors at the prediction points within the target prediction point and the wind power and wind direction sensors at the monitoring point, is the normalization parameter, The value of is greater than the value of

[0033] Furthermore, the determination method of the wind direction attenuation coefficient is:

[0034] The first step: Taking the monitoring point as the origin, the east-west direction as the X-axis, and the north-south direction as the Y-axis, establish a two-dimensional geographical direction coordinate system;

[0035] The second step: Obtain the angle between the wind direction at the target prediction point and the X-axis, and determine the wind vector;

[0036] The third step: Obtain the angle between the connection line between the wind power and wind direction sensors at the prediction points within the target prediction point and the wind power and wind direction sensors at the monitoring point and the X-axis, and determine the side vector;

[0037] The fourth step: Establish a determination formula for the wind direction attenuation coefficient, .

[0038] Furthermore, the method for calculating the material dust emission amount and the current building height is:

[0039] The first step: Obtain the construction time D of the day and the total building height Hn, obtain the construction material consumption An of the day, the previous building height Hp, the current construction building height Hq according to the construction progress, and obtain the emission coefficient qn of the current construction material;

[0040] The second step: Calculate the material dust emission amount M and the current building height P within the interval time t1, wherein, , ;

[0041] An air pollution prediction system based on deep learning, applied to the air pollution prediction method based on deep learning, includes:

[0042] Storage module: Used to store building information and construction information, predicted point wind power and wind direction data, monitoring point wind power and wind direction data, input sample feature set data, and predicted dust concentration after Tn time;

[0043] Calculation module: Calculate the time when the wind reaches the monitoring point, which is used to adjust the time Tn when the wind reaches the monitoring point in the input sample feature set; calculate the predicted wind force when the wind reaches the monitoring point, which is used to adjust the predicted wind force in the input sample feature set; calculate the material dust emission and the current building height, which are used to adjust the material dust emission and the current building height in the input sample feature set.

[0044] Deep learning model: Input the data obtained at several intervals into the input sample feature set for data input of the deep learning model. After obtaining the Tn time matching the interval time through the deep learning model, predict the dust concentration, compare the predicted dust concentration with the actual dust concentration data at the time monitoring point location, and optimize the deep learning model through the optimization function to improve the accuracy of the prediction result.

[0045] Compared with the prior art, the beneficial effects of the present invention are:

[0046] In the air pollution prediction method and system based on deep learning, the material dust emission M and the current building height P are introduced into the input sample feature set of the deep learning model. The purpose of the introduction is to consider the enhanced influence of construction materials on the dust concentration at the monitoring point during the interval period and the weakened influence of the building height on the wind force during the interval period, thereby increasing the prediction accuracy.

[0047] At the same time, determine the wind distance attenuation coefficient and the wind direction attenuation coefficient, and calculate the predicted wind force when the wind reaches the monitoring point through the predicted wind force model, which is used to adjust the predicted wind force in the input sample feature set, thereby fully considering the influence of wind force and wind direction on the dust concentration and further improving the prediction accuracy.

[0048] In addition, in this solution, a calculation module is set, and the collected relevant data is pre-calculated through the calculation module to obtain the time when the wind reaches the monitoring point, the predicted wind force, the material dust emission, and the current building height, reducing the number of input sample features of the deep learning, and being able to speed up the prediction efficiency of the deep learning model while improving the prediction accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 It is a schematic diagram for establishing the deep learning model of the present invention;

[0050] Figure 2 It is the first distribution schematic diagram of the predicted wind force and wind direction sensor and the monitored wind force and wind direction sensor of the present invention;

[0051] Figure 3 It is the second distribution schematic diagram of the predicted wind force and wind direction sensor and the monitored wind force and wind direction sensor of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0052] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without making creative efforts shall fall within the protection scope of the present invention.

[0053] In air quality prediction, deep learning, as a powerful machine learning technology, is gradually emerging. By simulating the structure and function of the human brain neural network, deep learning can automatically learn complex patterns from a large amount of data, which gives it significant advantages in dealing with complex problems such as air quality prediction.

[0054] Specifically in this solution, it should be noted that in construction sites, dust control is mainly achieved by watering when the dust is relatively large. Currently, there are already dust concentration detection devices used in construction sites on the market, but there is no prediction for the dust concentration in construction sites, and it is impossible to make preparations in advance when the environment changes.

[0055] Embodiment 1: As Figures 1-3 shown, the present invention provides a technical solution: a method and system for air pollution prediction based on deep learning, including: obtaining the construction information and construction information stored in the terminal storage device, and reading the construction information and construction information. Among them, the construction information includes the construction material information and building height information marked in the construction blueprint, and the construction information includes the construction progress information marked in the construction schedule. Through the construction progress information, the construction material information entering the site at this construction progress can be confirmed. Among them, the construction material information includes, but is not limited to, the amount of concrete required for construction, the amount of bricks required, the amount of steel bars required, and the amount of soil during earth excavation, and the height of the current building can be determined by combining the building height information and the construction progress information.

[0056] Set at least one monitoring point wind direction sensor at the monitoring point. Centered on the monitoring point, at least eight prediction points are set outside the monitoring point, and at least one prediction point wind direction sensor is set at each prediction point. As Figure 2 shown, specifically in this solution, the number of prediction points is set to eight, which are respectively arranged in the east, west, south, north, southeast, southwest, northwest, and northeast directions. It can be understood that the wind direction sensor is a device used to measure wind speed and wind direction, and is widely used in meteorological monitoring, environmental monitoring, wind power generation, agriculture, marine research, aviation and navigation and other fields. It can provide wind speed and wind direction data in real time and accurately. The specific working principle thereof will not be elaborated in this solution.

[0057] In this solution, the wind speed and direction sensors at the prediction points and the wind speed and direction sensors at the monitoring points are respectively used to monitor the wind speed and direction at the prediction points and the wind speed and direction at the monitoring points, so as to obtain the wind speed and direction data at the prediction points and the wind speed and direction data at the monitoring points.

[0058] Determine the interval time, set the interval time as t1, and then calculate the time when the wind reaches the monitoring point through the time model, which is used to adjust the time Tn when the wind reaches the monitoring point in the input sample feature set.

[0059] Specifically, the method for establishing the time model is as follows:

[0060] The first step: According to the wind direction and the positions of the prediction points and the monitoring points, determine the target prediction points. In this solution, the distances between each prediction point and the monitoring point can be different. However, considering that the wind speed and direction are affected by obstacles, resulting in a decrease in wind speed and a change in wind direction, the wind speed and direction sensors at the prediction points need to be 2 - 10 meters higher than the obstacles.

[0061] Among them, the method for determining the target prediction points is as follows:

[0062] Determine the wind direction through the wind speed and direction sensors set at the monitoring points or the prediction points, and determine the positions of several wind speed and direction sensors at the prediction points that are downwind according to the wind direction. It should be noted that in this solution, the definition of downwind is the direction in which the wind blows from the position of the prediction point to the monitoring point.

[0063] In the established two-dimensional coordinate system, connect all the wind speed and direction sensors at the prediction points that are downwind and the wind speed and direction sensors at the monitoring points. The position of the wind speed and direction sensor at the monitoring point is the origin of the two-dimensional coordinate system. In this way, the wind speed and direction sensors at the prediction points set in the east-west direction are respectively located on the positive and negative axes of the X-axis of the coordinate system, and the wind speed sensors at the prediction points set in the north-south direction are respectively located on the negative and positive axes of the Y-axis of the coordinate system;

[0064] Calculate the angles between the wind direction and the connections of all the wind speed and direction sensors at the prediction points that are downwind and the wind speed and direction sensors at the monitoring points, and determine the position of the wind speed and direction sensor at the prediction point with the smallest angle as the target prediction point. Specifically, as Figure 2 shown, if the wind is northwest, the three wind speed and direction sensors at the prediction points above the X-axis are the several wind speed and direction sensors at the prediction points that are downwind. The angles between the wind and the connections of the three wind speed and direction sensors at the prediction points above the X-axis and the wind speed and direction sensor at the monitoring point are a1, a2, and a3 respectively. Since a1 is less than a2 and a2 is less than a3, the position of the wind speed and direction sensor at the prediction point at the a1 angle is the target prediction point.

[0065] Step 2: Calculate the straight-line distance between the wind speed and direction sensor at the predicted point and the wind speed and direction sensor at the monitoring point within the target predicted point through the satellite map system. Of course, other methods can also be used to calculate the straight-line distance. For example, after determining the coordinates of the wind speed and direction sensor at the predicted point and the wind speed and direction sensor at the monitoring point, calculate through the vector norm formula.

[0066] Step 3: At every interval of t1 time, detect the wind speed and direction of the target predicted point through the wind speed and direction sensor at the predicted point within the target predicted point. In this solution, the value range of t1 is 3 - 10 minutes, and the preferred interval time is 6 minutes.

[0067] Step 4: At every interval time point of t1, input the wind speed and direction data obtained by the wind speed and direction sensor at the predicted point within the target predicted point into the numerical weather prediction model to obtain the time Tn when the wind reaches the monitoring point. It should be noted that the numerical weather prediction model (NWP) is an existing meteorological model, which can simulate the spatio-temporal changes of the wind by combining data such as air pressure, humidity, and temperature, so as to accurately calculate the time Tn when the wind reaches the monitoring point under a certain wind speed. In this solution, there may be a sudden change in wind speed at adjacent interval time points. When the sum of the time Tn2 when the wind reaches the monitoring point at the later time point and the interval time t1 is less than or equal to the time Tn1 when the wind reaches the monitoring point at the previous time point, then the time when the wind reaches the monitoring point at adjacent interval time points is all Tn2, that is, the interval time input into the sample feature set is all Tn2.

[0068] Embodiment 2: On the basis of Embodiment 1, calculate the predicted wind speed when the wind reaches the monitoring point through the predicted wind speed model, and use it to adjust the predicted wind speed in the input sample feature set. By calculating the predicted wind speed when the wind reaches the monitoring point through the predicted wind speed model and using it to adjust the predicted wind speed in the input sample feature set, the influence of wind speed and direction on the dust concentration is fully considered, and the prediction accuracy is further improved.

[0069] Specifically, the method for establishing the predicted wind speed model is as follows:

[0070] Determine the wind distance attenuation coefficient and the wind direction attenuation coefficient , and then establish the predicted wind speed model through the formula .

[0071] Among them, the formula for the wind distance attenuation coefficient is: , is the straight-line distance between the wind speed and direction sensor at the predicted point and the wind speed and direction sensor at the monitoring point within the target predicted point, is the normalization parameter, The value of is greater than The value of, and The value of is The value is multiplied by an increment factor, which is manually assigned and depends on the numerical range, that is, the increment factor is different for each numerical range.

[0072] For example, when the straight-line distance between the wind direction and speed sensor at the prediction point and the wind direction and speed sensor at the monitoring point within the target prediction point is 2.5 - 3 km, the increment factor is 1.66. If the straight-line distance between the wind direction and speed sensor at the prediction point and the wind direction and speed sensor at the monitoring point within the target prediction point is 3 km, then the normalized parameter is 5 km, then , and the calculation result is 0.55, which means the influence weight of the distance from the prediction point to the monitoring point is 55%. The larger the value, the greater the obtained distance influence weight. By quantifying the spatial attenuation law, it helps the model to more realistically reflect the influence of the prediction position on the interaction intensity.

[0073] Furthermore, as Figure 2 shown, the method for determining the wind direction attenuation coefficient is as follows:

[0074] Step 1: Taking the monitoring point as the origin, with the east-west direction as the X-axis and the north-south direction as the Y-axis, establish a two-dimensional geographical direction coordinate system;

[0075] Step 2: Obtain the angle between the wind direction at the target prediction point and the X-axis to determine the wind vector;

[0076] Step 3: Obtain the angle between the line connecting the wind direction and speed sensor at the prediction point and the wind direction and speed sensor at the monitoring point within the target prediction point and the X-axis to determine the side vector;

[0077] Step 4: Establish a wind direction attenuation coefficient formula, .

[0078] Combined with Figure 2 and Figure 3 for illustration, through the angle between the wind direction at the target prediction point and the X-axis, the determined wind vector is (sin a1, cos a1), and the determined side vector is (sin b1, cos b1). Assuming a1 = 30° and b1 = 45°, then, , that is, the wind direction attenuation is (sin30°×sin45° + cos30°×cos45° + 1) / 2, and the calculated result is 0.982. Assuming the detected wind force is 6 m / s, then the finally obtained predicted wind force is 0.55×0.982×6 m / s, and the result is 3.24 m / s.

[0079] Embodiment 3: Based on Embodiment 1 or both Embodiment 1 and Embodiment 2, calculate the material dust emission and the current building height to adjust the material dust emission and the current building height in the input sample feature set;

[0080] Specifically, the method for calculating the material dust emission and the current building height is as follows:

[0081] First step: Obtain the construction time D of the day and the total building height Hn, and obtain the material consumption An for the day's construction, the previous building height Hp, the current building height Hq for the day's construction, and the emission coefficient qn of the current construction material according to the construction progress;

[0082] Second step: Calculate the material dust emission M and the current building height P within the interval time t1. Among them, , , it can be known that An can be the concrete pouring volume required for construction, or the brick requirement or the soil volume during earth excavation. qn is the emission coefficient of An. For example, if An is the concrete pouring volume for the day's construction, according to the material-dust emission coefficient (EPA construction dust guide), qn can be obtained as 0.5 g / m³. Through the formula, it can be known that the material dust emission M is the sum of the dust generated by several construction material consumptions. The purpose of introducing the material dust emission M and the current building height P into the input sample feature set is to consider the enhanced impact of construction materials on the dust concentration at the monitoring point during the interval time period and the weakened impact of the building height on the wind force during the interval time period, thereby increasing the accuracy of the prediction.

[0083] Build a deep learning model. The deep learning model is one of the feedforward neural network and the recurrent neural network. Specifically, in the feedforward neural network, after normalizing the data such as the current dust concentration data at the monitoring point, the time when the wind reaches the monitoring point, the predicted wind force when the wind reaches the monitoring point, the material dust emission within the interval time, and the current building height data, input them into the input sample feature set. Flatten all the features into a single vector, then perform data loading and training, and perform backpropagation on the output result to optimize the model parameters. Finally, obtain the prediction result. Different from the feedforward neural network model, in the deep learning model of the recurrent neural network, retain the time series dimension for the above data and process the features step by step for each time step.

[0084] Finally, input the data obtained at several intervals into the sample feature set for data input of the deep learning model. After obtaining the Tn time corresponding to the interval time through the deep learning model, predict the dust concentration. Compare the predicted dust concentration with the actual dust concentration data at the time monitoring point location, and optimize the deep learning model through the optimization function to improve the accuracy of the prediction result. For example, the interval time can be the time data of one construction day or multiple construction days. To improve the prediction accuracy of the model, the number of interval times is not less than 100. For example, on the fifth, tenth, thirteenth, and twentieth days of construction, select continuous or discontinuous interval times every day, and the number of interval times is not less than 20. For example, on the fifth day, select 10:00 - 10:06 as the first time interval, 10:00 as the first monitoring time. Corresponding to the first monitoring time, the time when the wind reaches the monitoring point at this monitoring time is T01, and the monitoring interval time each time is 6 minutes, that is, 10:06 is the second monitoring time. In this way, input the time when the wind reaches the monitoring point obtained at 10:00, the predicted wind force when the wind reaches the monitoring point, the calculated material dust emission amount, and the current building height data into the deep learning model to obtain the predicted dust concentration. If the difference between the predicted dust concentration and the current dust concentration data when the wind reaches the monitoring point at time T01 is within the error threshold, optimize the deep learning model through the optimization function, so that the difference between the predicted dust concentration and the current dust concentration data when the wind reaches the monitoring point at time T01 is within the error threshold.

[0085] This solution also discloses an air pollution prediction system based on deep learning, which is applied to the air pollution prediction method based on deep learning, and includes:

[0086] Storage module: used to store building information and construction information, predicted point wind force and wind direction data, monitoring point wind force and wind direction data, input sample feature set data, and predicted dust concentration after Tn time;

[0087] Calculation module: calculates the time when the wind reaches the monitoring point, which is used to adjust the time Tn when the wind reaches the monitoring point in the input sample feature set, calculates the predicted wind force when the wind reaches the monitoring point, which is used to adjust the predicted wind force in the input sample feature set, calculates the material dust emission amount and the current building height, which are used to adjust the material dust emission amount and the current building height in the input sample feature set. It can be known that the setting of the calculation module is to pre-calculate the collected relevant data through the calculation module, so as to obtain the time when the wind reaches the monitoring point, the predicted wind force, the material dust emission amount, and the current building height, reducing the number of input sample features of the deep learning, and being able to speed up the prediction efficiency of the deep learning model while improving the prediction accuracy.

[0088] Deep learning model: Use the input sample feature set for data input of the deep learning model, compare the predicted dust concentration after a certain number of Tn time periods with the actual dust concentration data at the time monitoring point locations, and optimize the deep learning model to improve the prediction results.

[0089] Although the embodiments of the present invention have been shown and described, those of ordinary skill in the art can understand that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and spirit of the present invention. The scope of the present invention is defined by the appended embodiments and their equivalents.

Claims

1. A deep learning-based air pollution prediction method, characterized in that, Including: Obtain the building information and construction information stored in the terminal storage device, and read the building information and construction information. Among them, the building information includes the construction material information and building height information marked in the construction blueprint, and the construction information includes the construction progress information marked in the construction schedule; Set at least one monitoring point wind direction and speed sensor at the monitoring point. Centered on the monitoring point, at least eight prediction points are set around the monitoring point, and at least one prediction point wind direction and speed sensor is set at each prediction point; Monitor the wind force and wind direction at the prediction points through the prediction point wind direction and speed sensors set at the prediction points to obtain the wind force and wind direction data at the prediction points. Monitor the wind force and wind direction at the monitoring point through the monitoring point wind direction and speed sensors set at the monitoring point to obtain the wind force and wind direction data at the monitoring point; Determine the interval time, and calculate the time when the wind reaches the monitoring point through the time model, which is used to adjust the time Tn when the wind reaches the monitoring point in the input sample feature set; Calculate the predicted wind force when the wind reaches the monitoring point through the predicted wind force model, which is used to adjust the predicted wind force in the input sample feature set; Calculate the material dust emission amount and the current building height, which are used to adjust the material dust emission amount and the current building height in the input sample feature set; Obtain the current dust concentration data at the monitoring point, and input the current dust concentration data into the input sample feature set, which is used to predict the dust concentration at the monitoring point position after time Tn; Establish a deep learning model, and the deep learning model is one of the feedforward neural network and the recurrent neural network; Input the time when the wind reaches the monitoring point, the predicted wind force when the wind reaches the monitoring point, the calculated material dust emission amount and the current building height data obtained at several interval times into the input sample feature set for data input of the deep learning model. Obtain the predicted dust concentration after time Tn matching the interval time through the deep learning model. Compare the predicted dust concentration with the actual dust concentration data at the monitoring point position at the time, and optimize the deep learning model through the optimization function to improve the accuracy of the prediction result.

2. The air pollution prediction method based on deep learning according to claim 1, wherein, The method for establishing the time model is as follows: The first step: Determine the target prediction point according to the wind direction and the positions of the prediction points and the monitoring point; The second step: Calculate the straight-line distance between the prediction point wind direction and speed sensors and the monitoring point wind direction and speed sensors within the target prediction point through the satellite map system; The third step: Detect the wind force and wind direction at the target prediction point through the prediction point wind direction and speed sensors within the target prediction point every t1 time interval; The fourth step: At every t1 time point, input the wind force and wind direction data obtained by the prediction point wind direction and speed sensors within the target prediction point into the numerical weather prediction model to obtain the time Tn when the wind reaches the monitoring point.

3. The air pollution prediction method based on deep learning according to claim 2, characterized in that, In the first step, the method for determining the target prediction point is as follows: Determine the wind direction through the wind direction and speed sensors set at the monitoring point or the prediction points, and determine the positions of several prediction point wind direction and speed sensors that are downwind according to the wind direction; In the two-dimensional coordinate system, connect all the determined prediction point wind direction and speed sensors that are downwind and the monitoring point wind direction and speed sensors; Calculate the angle between the wind direction and the line connecting the wind sensor at the predicted point and the wind sensor at the monitoring point for all downwind directions, and determine the position of the wind sensor at the predicted point with the minimum angle as the target predicted point.

4. A method for predicting air pollution based on deep learning according to claim 2, characterized in that, In the fourth step, if the sum of the time Tn2 when the wind at the later time point arrives at the monitoring point and the interval time t1 is less than or equal to the time Tn1 when the wind at the previous time point arrives at the monitoring point for adjacent time intervals, then the time when the wind at the adjacent time intervals arrives at the monitoring point is Tn2.

5. The air pollution prediction method based on deep learning according to claim 2, wherein, The method for establishing the predicted wind force model is as follows: Determine the wind distance attenuation coefficient and the wind direction attenuation coefficient , and then establish a predicted wind force model through the formula .

6. The air pollution prediction method based on deep learning according to claim 5, wherein The determination formula for the wind distance attenuation coefficient is as follows: ; Among them, is the straight-line distance between the wind speed and direction sensor at the prediction point and the wind speed and direction sensor at the monitoring point within the target prediction point, is the normalization parameter, The value is greater than value.

7. The air pollution prediction method based on deep learning according to claim 5, wherein The method for determining the wind direction attenuation coefficient is as follows: Step 1: Establish a two-dimensional geographical direction coordinate system with the monitoring point as the origin, the east-west direction as the X-axis, and the north-south direction as the Y-axis. Step 2: Obtain the angle between the wind direction at the target predicted point and the X-axis, and determine the wind vector. Step 3: Obtain the angle between the line connecting the wind sensor at the predicted point within the target predicted point and the wind sensor at the monitoring point and the X-axis, and determine the side vector. Step 4: Establish the determination formula for the wind direction attenuation coefficient, .

8. The air pollution prediction method based on deep learning according to claim 2, characterized in that, The method for calculating the material dust emission and the current building height is as follows: Step 1: Obtain the construction time D of the day and the total building height Hn, obtain the material consumption An of the day according to the construction progress, the previous building height Hp, the building height Hq under construction on the day, and obtain the emission coefficient qn of the current construction material. Step 2: Calculate the material dust emission M and the current building height P within the interval time t1, where, , .

9. A deep learning-based air pollution prediction system, applied to the deep learning-based air pollution prediction method described in any one of claims 1-8, characterized in that, It includes: Storage module: used to store building information and construction information, predicted wind force and wind direction data, monitored wind force and wind direction data, input sample feature set data, and predicted dust concentration after time Tn. Calculation module: calculates the time when the wind arrives at the monitoring point, used to adjust the time Tn when the wind arrives at the monitoring point in the input sample feature set, calculates the predicted wind force when the wind arrives at the monitoring point, used to adjust the predicted wind force in the input sample feature set, calculates the material dust emission and the current building height, used to adjust the material dust emission and the current building height in the input sample feature set. Deep learning model: inputs the data obtained at several interval times into the input sample feature set for data input of the deep learning model, obtains the predicted dust concentration after time Tn matching the interval time through the deep learning model, compares the predicted dust concentration with the actual dust concentration data at the time monitoring point position, and optimizes the deep learning model through the optimization function to improve the accuracy of the prediction result.

Images