WO2020043028A1 - 一种利用历史空气质量数据特征预测空气污染的方法 - Google Patents
一种利用历史空气质量数据特征预测空气污染的方法 Download PDFInfo
- Publication number
- WO2020043028A1 WO2020043028A1 PCT/CN2019/102417 CN2019102417W WO2020043028A1 WO 2020043028 A1 WO2020043028 A1 WO 2020043028A1 CN 2019102417 W CN2019102417 W CN 2019102417W WO 2020043028 A1 WO2020043028 A1 WO 2020043028A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- air quality
- quality data
- historical
- data
- grid
- Prior art date
Links
- 238000000034 method Methods 0.000 title claims abstract description 93
- 238000003915 air pollution Methods 0.000 title claims description 11
- 238000012544 monitoring process Methods 0.000 claims abstract description 10
- 239000011159 matrix material Substances 0.000 claims description 68
- 238000004364 calculation method Methods 0.000 claims description 19
- 238000004458 analytical method Methods 0.000 claims description 16
- 238000001914 filtration Methods 0.000 claims description 4
- 230000003247 decreasing effect Effects 0.000 claims 3
- 239000013598 vector Substances 0.000 description 67
- 238000009792 diffusion process Methods 0.000 description 17
- 239000003344 environmental pollutant Substances 0.000 description 16
- 231100000719 pollutant Toxicity 0.000 description 16
- 238000010586 diagram Methods 0.000 description 8
- 238000012067 mathematical method Methods 0.000 description 5
- 238000012216 screening Methods 0.000 description 4
- 238000003672 processing method Methods 0.000 description 3
- 238000011161 development Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 230000007613 environmental effect Effects 0.000 description 2
- 238000012417 linear regression Methods 0.000 description 2
- 238000012821 model calculation Methods 0.000 description 2
- 239000000126 substance Substances 0.000 description 2
- 239000000809 air pollutant Substances 0.000 description 1
- 231100001243 air pollutant Toxicity 0.000 description 1
- 210000003888 boundary cell Anatomy 0.000 description 1
- 210000004027 cell Anatomy 0.000 description 1
- 238000011109 contamination Methods 0.000 description 1
- 238000012937 correction Methods 0.000 description 1
- 238000007405 data analysis Methods 0.000 description 1
- 230000007423 decrease Effects 0.000 description 1
- 238000012905 input function Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000002277 temperature effect Effects 0.000 description 1
- 238000012876 topography Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
Definitions
- the invention relates to a method for predicting air pollution by using characteristics of historical air quality data, and belongs to the field of environmental monitoring.
- Air pollution directly endangers people's physical health. Therefore, people have a need to know the air quality of a more accurate future date in order to arrange their work and life, and this demand is relatively strong.
- Air quality data generally come from measured values from environmental protection agencies.
- the current technology also has technology to predict air quality, which is mainly divided into two types: one is to perform prediction based on chemical prediction model calculation, that is, to establish a diffusion model, and the method to perform prediction based on chemical prediction model calculation requires extremely high calculations Resources, it is difficult to achieve; the other is based on local historical meteorological data and historical air quality data, combined with meteorological data at the predicted time, to predict the air quality data at the local predicted time, this method
- the selected historical meteorological data and historical air quality data are data at the same time as the predicted time in the historical date. For example, if the predicted time is 15:00, the selected historical meteorological data and historical air quality data are some historical data. Or the data at 15:00 on some dates, the accuracy of the air quality data predicted by this method is relatively low.
- the above two methods of air quality prediction have many disadvantages.
- the first one requires a large amount of parameter data input and has extremely high requirements for computing resources and weather monitoring networks.
- the second one is based on historical meteorological data and historical air quality data. Two types of data are needed, and this simple prediction method based on historical meteorological data and historical air quality data is less accurate. Both methods require meteorological data, and they cannot make accurate and real-time predictions on small and medium scales; they cannot accurately predict the movement direction and trajectory of pollution drift.
- Invention name Method and device for predicting air quality.
- the invention uses historical meteorological data and historical air quality data from all areas within the predicted area and surrounding preset distances, selects certain historical meteorological data and historical air quality data, and combines these data with current meteorological data and air quality data.
- the input function is used to predict the air quality.
- the invention still requires meteorological data and cannot predict the path and trend of pollutant drift.
- invention name Air quality prediction method and device.
- the invention uses historical air quality data and a target date, and simply predicts by analyzing the probability of historical air quality possible air conditions on the target date. Not using existing monitoring data at all, resulting in greatly reduced prediction accuracy and the possibility that predictions that do not conform to the situation at the time may occur.
- Historical air quality data Air pollutant concentration at a certain time in history.
- Historical air quality data characteristics A vector that expresses the contribution of the monitored area from the previous moment to the surrounding area pollutants at a certain time in the past.
- Historical air quality data feature database A database composed of vectors that characterize historical air quality data features.
- Air quality data frame The area to be predicted is divided into a grid with a set density according to the predicted demand, and the historical air quality data is assigned to the grid based on time and geographical location to generate time series based geographic information Air quality data set.
- Air quality data matrix The matrix obtained by expressing the data contained in the data frame in the form of a matrix is the air quality data matrix.
- Recent air quality data The air quality data calculated before the pollution process is predicted to be compared with historical data.
- the vector of the pollution process to be predicted is up to a certain time before comparison with historical data, and a vector expressing the contribution of the monitored area from the previous moment to the surrounding area's pollutants at the next moment.
- Matching degree coefficient A coefficient indicating the degree of matching between the characteristics of recent air quality data of the pollution process to be predicted and the characteristics of historical air quality data to be matched.
- Vector group A vector group calculated based on the air quality data frames at two moments and including the characteristic vectors in each grid to represent the diffusion path of pollutants.
- Air quality data characteristics over time Vectors characterizing the information of pollutant diffusion between two times.
- Ratio comparison matrix After the ratio of the historical air quality data frame and the current air quality data frame to form the ratio comparison data frame, the matrix corresponding to the ratio comparison data frame is the ratio comparison matrix.
- Similar grid interval A piece of historical air quality data similar to the recent development process of air quality data obtained through the feature vector matching method.
- Difference comparison matrix After forming a difference comparison data frame from the difference between the data of the historical air quality data frame and the current air quality data frame, the matrix corresponding to the difference comparison data frame is the difference comparison matrix.
- the invention provides a method for predicting air pollution by using characteristics of historical air quality data, and accurately predicts air pollution and pollution drift path trends by using less computing resources and not using difficult-to-obtain near-surface weather data.
- the present invention provides the following technical solutions:
- the area to be predicted is divided into a grid with a set density according to the predicted demand, and historical air quality data is assigned to the grid according to time and geographical location, and an air quality data frame with two-dimensional geographic location information and time stamp information is generated.
- a three-dimensional spatio-temporal grid is established with time information as the axis, and data frames with two-dimensional geographic information and air quality data are arranged to generate a three-dimensional spatio-temporal network.
- the air quality data of the grid can be obtained by mathematical methods such as interpolation and diffusion model at the same time from the grid data of adjacent space.
- the air quality data of the grid can also be obtained through mathematical methods such as interpolation and diffusion model in the same space, and the grid data at adjacent times.
- each two-dimensional geographic grid can be 10-1000 meters.
- the same time refers to the air quality data of 1 minute to 1 hour before and after the time point.
- the method is to analyze the pollution contribution of each grid in the previous data frame to the adjacent position of the grid in the next data frame. To obtain the historical air quality data characteristics between these two moments.
- the feature is that each grid contributes to the surrounding pollution during these two moments, and represents the direction of pollution drift between these two moments.
- Features can represent the path of pollution drift.
- the historical air quality data at time T 1 is filled into the divided grid according to the geographic location information, and an air quality data frame at time T 1 is obtained.
- a data frame is an air quality data plane with two-dimensional geographic location information at a time in a three-dimensional space-time network.
- the air quality data frame at time T 1 is converted into an A matrix according to the grid.
- the values of the elements in the first row and first column of the A matrix are the air quality of the first row and first column in the air quality data frame at T 1 .
- such extrapolated values of the matrix elements of a m-th row n-th column is the time T 1 of the air quality data frames air mass m th row and n columns, a matrix representative of time T 1 within the area air quality data frame.
- the air quality data matrix B at time T 2 is similarly obtained by the above method, and the B matrix represents the air quality data frame in the area at time T 2 .
- the pollution contribution of each grid to the adjacent grid at the previous moment is calculated separately, and the pollution contribution of each grid to the neighboring grid is obtained by calculating the vector.
- a mn is the air quality data at the previous moment
- B mn is the air quality data at the next moment.
- T 1 is the air quality data of the previous moment
- T 2 is the air quality data of the moment.
- the pollution drift of the grid around the A mn region at time T 1 at time T 2 can be expressed in a vector manner, and the pollution drift direction can be artificially divided into three ways.
- Method 1 At time T 1 A mn has a contribution to the up, down, left, and right directions at time T 2 . That is, A mn contributes to B (m-1) n , B m (n-1) , B m (n + 1), and B (m + 1) n) .
- the pollution drift vector is Then the calculation method of the pollution drift vector of the first method is as follows:
- T 1 A mn time in their four directions upper left, lower left, upper right, lower right at time T 2 has contributed drift. That is, A mn versus B (m-1) (n-1) , B (m-1) (n + 1) , B (m + 1) (n-1) , B (m + 1) (n + 1) Contribute.
- the pollution drift vector is Then the calculation method of the pollution drift vector of the second method is as follows:
- Method 3 At time T 1 A mn has a drift contribution to its eight directions of up, down, left, right, top left, bottom left, top right, and bottom right at time T 2 . That is, A mn versus B (m-1) (n-1) , B (m-1) n , B (m-1) (n + 1) , B m (n-1) , B m (n + 1) , B (m + 1) (n-1) , B (m + 1) n) , B (m + 1) (n + 1) contribute.
- the pollution drift vector is Then the calculation method of the pollution drift vector of the third method is as follows:
- a mn the contribution of A mn to the surroundings is also the direction of pollution drift between these two times, that is, the air quality data characteristics of A mn between time T 1 and time T 2 ; continuous characteristics between multiple times can represent pollutants Drift path.
- the pollution drift vector is also the feature vector.
- the method of analyzing the characteristics of recent air quality data is similar to the way of building a historical air quality data feature database.
- the air quality data with geographic location information of the air quality monitoring station is assigned to the already-divided grid of the region to be predicted, and the current air quality data frame and one or more groups of a certain time forward based on the current time are generated.
- Data frame and then generate the current air quality and the air quality data matrix at a certain time forward.
- a vector set of recent air quality data changes can be obtained, and this vector set is the recent air quality Data characteristics.
- a c represents the current pollution data matrix
- B c represents the current pollution data matrix at a certain time t 1
- the historical air quality data frame and the recent air quality data frame are as follows.
- the two data frames are compared to determine whether the historical pollution process in which the A data frame is located can predict the development trend of the recent pollution process.
- the corresponding A matrix and A c matrix are obtained from the historical air quality data frame and the current air quality data frame.
- ⁇ represents the ratio of the recent data unit to the historical data unit at the same location.
- the ratio comparison matrix can be obtained from the ratio data frame
- ⁇ represents the absolute value of the difference between the recent data unit and the historical data unit at the same location.
- the difference comparison matrix can be obtained from the difference
- the characteristics of the recent air quality data are compared with the characteristics of the historical air quality data to obtain a matching coefficient ⁇ .
- ⁇ Based on mathematical methods such as vector shortest distance, angle cosine, Mahalanobis distance, and vector similarity, you can find the historical air quality data features that are closest to the recent air quality data features, and you can find the current air quality or current pollution status.
- the historical air quality movement process or pollution process that is closest to the movement path.
- f t is related to the temperature parameter (t temp ). When it is higher than a certain temperature or lower than a certain temperature, it is an extreme high or low temperature, which is not suitable for prediction.
- f s is related to the season parameter (s season ). If the historical air quality data is close to or the same as the season, temperature and humidity in the recent air quality data, the degree of matching is higher; if the historical air quality The data and the recent air quality data in the season, temperature and humidity are greatly different, the impact factor value is smaller.
- f w is related to the weather condition parameter (w weather ).
- the historical air quality data and the recent air quality data have extreme weather at the time.
- the value of the extreme weather impact factor is very low. It cannot be used as a reference for prediction, and the current situation is not suitable for prediction.
- f p is related to the pollution path parameter (p path ).
- the distance of the pollution movement path and the source of the pollution should be considered. If the recent air quality data features have a high degree of matching with the pollutant drift path of multiple historical air quality data features, then the matching degree with the historical air quality data feature paths with a shorter path distance in the recent air quality data features has a higher degree of matching. .
- Factor diffusion factor is set: f d is related to the diffusion degree parameter (d diffusion ).
- the diffusion degree of two pollution processes should be used as a parameter. If the degree of diffusion and the area level of the two pollution processes are similar, the matching degree coefficient should be increased accordingly. If the degree of diffusion and the area level of the two pollution processes are significantly different, the coefficient of the matching system degree Reduce accordingly;
- Repeatability set Factor Factor (factor repeatability): f r and the repetition frequency parameter (r repeatability) For (r repeatability), similar to the historical process contamination should occur as a parameter. If similar historical pollution process models matching the pollution process to be predicted have occurred many times, it is considered that the predictions made using the historical air quality data are more reliable and a higher matching degree coefficient should be set;
- f g is related to the terrain parameter (g geo ), and the terrain factor where the history and the location of the pollution process to be predicted are located should also be used as parameters. If the topography of the two pollution processes differs greatly, for example, one time there were no more buildings in the local area a few years ago, and the other time there were no buildings matching the level a few years ago with many buildings The coefficient decreases accordingly.
- Influencing factors Influence parameter condition Impact factor humidity h humid Humidity> 90% f h ⁇ 0 humidity h humid Humidity ⁇ 90% f h ⁇ 1 temperature t temp Temperature > 40 °C or temperature ⁇ -20 °C f t ⁇ 0 temperature t temp -20 °C ⁇ temperature ⁇ 40 °C f t ⁇ 1 extreme weather w weather Sandstorm, heavy rain, snowstorm, etc.
- f w ⁇ 0 path p path The path is repeated more than 5 times f p ⁇ 1 Diffusion, sphere of influence d diffusion Diffusion and influence range are similar f d ⁇ 1 terrain g geo Surrounded by tall buildings f g ⁇ 0
- Cosine similarity uses the cosine of the angle between two vectors in a vector space as a measure of the difference between two individuals. The closer the cosine value is to 1, the closer the angle is to 0 degrees.
- a vector characterizing the recent and historical air quality characteristics of the monitored area at a specific time versus The comparison can also be done using the angle cosine method, and the calculation process is the same as above (Note: for all air quality data obtained during the monitored period)
- the angle cosine matching coefficient ⁇ t is applied over the entire grid range at time t :
- Matching coefficient of angle cosine applied to the m-th row and n-th row of the square at a certain time in a period of time T q represents the number of matches in the T period:
- the matching coefficient ⁇ T of the included angle cosine method is applied over the entire grid range within a period of time T, and q represents the number of matching times in the time period of T:
- ⁇ ′ is the modified matching coefficient
- Selection of historical air quality data characteristics The selection of historical air quality data characteristics can be determined by ranking the matching coefficients directly, or by filtering and ranking.
- the direct ranking method is to arrange the matching coefficients or modified matching coefficients from large to small. The closer the matching coefficient or modified matching coefficient is to 1, the higher the ranking, and finally the top 10%, 20%, or ranking is selected. A certain percentage of the matching coefficient, the historical air quality data characteristics corresponding to this matching coefficient can be used to predict the future air quality.
- Matching coefficient ⁇ 5 ⁇ 3 ⁇ 1 ⁇ 4 ⁇ 2 ⁇ 7 ⁇ 6 ⁇ 9 ⁇ 8 ⁇ 10 Matching coefficient value 0.98 0.97 0.95 0.9 0.89 0.8 0.8 0.75 0.72 0.7 Rank 1 2 3 4 5 6 7 8 9 10
- the top 20% of the matching coefficients are selected, that is, ⁇ 5 and ⁇ 3 are selected.
- the historical air quality data characteristics corresponding to the selected matching coefficients can be used to predict future air quality data.
- the method of ranking after screening is to first conduct a round of screening of historical air quality characteristics through objective factors, and then rank the matching coefficients after screening to determine the required matching coefficients and corresponding historical air quality data characteristics.
- the objective factors used for screening include the historical weather characteristics of the historical air quality data (such as extreme weather conditions at the historical moment: heavy rain, extreme cold, high temperature, etc., the historical moment is excluded), the historical air quality data is a long time from the current moment (such as historical data over 3 years).
- rank and select the matching coefficient After filtering and excluding, rank and select the matching coefficient.
- the ranking method can be arranged from large to small. The closer the matching coefficient is to 1, the higher the ranking. Finally, the top 10%, 20%, or a certain percentage of the top ranking matching coefficient is selected. The historical air quality data corresponding to this matching coefficient. Features can then be used to predict future air quality.
- Matching coefficient ⁇ 5 ⁇ 3 ⁇ 1 ⁇ 4 ⁇ 2 ⁇ 7 ⁇ 6 ⁇ 9 ⁇ 8 ⁇ 10 Matching coefficient value 0.98 0.97 0.95 0.9 0.89 0.8 0.8 0.75 0.72 0.7
- Historical objective factors high temperature - 3 years ago - rainstorm - - - - - Rank - 1 - 2 - 3 4 5 6 7 According to the above table, the top 20% of the matching coefficients are selected, that is, ⁇ 3 and ⁇ 4 are selected.
- the historical air quality data characteristics corresponding to the selected matching coefficients can be used to predict future air quality data.
- the ratio of ⁇ mn elements that meet the set requirements is calculated according to mathematical methods such as average, mode, data distribution, and linear regression.
- the historical air quality data frame A and the recent air quality data frame A c are considered similar.
- ⁇ range that meets the requirements as Within, that is, first calculate the average of the ratio data frame elements If the value of ⁇ mn is in Within the range, ⁇ mn is considered to meet the requirements.
- the ⁇ range that meets the requirements can be set artificially, can be set as a multiple of the average value of the ratio, or a fixed range.
- the ratio of the ⁇ mn elements meeting the set requirements is calculated according to mathematical methods such as average, mode, data distribution, and linear regression.
- the values of the elements in the matrix of the difference ⁇ are statistically calculated and calculated. When the proportion of elements that meet the set standard exceeds the certain element ratio, the historical air quality data frame A and the recent air quality data frame A c are considered similar .
- the average method set the ⁇ range that meets the requirements as Within, that is, first calculate the average of the difference data frame If the value of ⁇ mn is between Within, it is considered that ⁇ mn meets the requirements.
- the ⁇ range that meets the requirements can be set artificially, can be set as a multiple of the difference average, or a fixed range.
- the current air quality data is used as a reference (the current air quality data is known), and the pollution drift vector (the subsequent time) at the time subsequent to the time when the historical matching grid is known.
- the calculation method of the pollution drift vector at the moment is still the pollution drift vector calculation method).
- the pollution drift vector calculation method is used in the backward direction to estimate and predict future air quality data to obtain the required future air quality data, that is, using the current air quality data frame
- a matrix operation is performed on the elements and corresponding feature vectors to obtain air quality data information of the surrounding positions of the current element position in the future.
- the matching coefficient or the percentage of the modified matching coefficient can be given to the corresponding prediction situation as the accuracy of the prediction.
- the outermost circle of cells in the boundary of the statistical area is contracted to form a new statistical area.
- the original grid data of the boundary area is not used as the basis for prediction.
- the original grid data of the boundary area is used for the calculation of vector features.
- all the grid data of the new statistical area can be used as the basic data for prediction and participate in the calculation of vector features.
- the data in the first row, the first column, the last row, and the last column are not used as the basic data for prediction, but all can be used as the basic data for vector calculation.
- the basic data regions that can be used for prediction are A c 22 , A c 2 (n-1) , A c (m-1) 2 , A c (m-1) (n-1) Data (including the four coordinates just now).
- the expansion method can be to directly copy the value of the original boundary as the value of the new boundary, or use the difference to expand outward.
- the raw data area can be used for both the basis of prediction and the calculation of vector features.
- the data of the newly extended outward boundary cannot be used as the basis for prediction, but the data of the newly outward extended boundary can be used for the calculation of vector features.
- the data in the first row, the first column, the last row, and the last column are expanded and are not used as the basic data for prediction, but all can be used as the basic data for vector calculation.
- the original data area is the data in the A c 11 , A c 1n , A c m1 , and A c mn grids (including the four coordinates just now).
- the data in this area can also be used as the data basis for prediction.
- the future air quality prediction method after time t is based on the historical air quality data frame and ratio characteristics at time t after the time point of calculating the characteristic of the ratio, and the ratio characteristic analysis method is used to reversely obtain the future after time t.
- Air quality data That is, the matrix of the historical air quality data frame at time t after the time point of the historical air quality data frame and the ratio characteristic value are used to perform matrix multiplication to obtain future air quality data.
- ⁇ is the percentage of the similarity coefficient, which can be given to the corresponding prediction situation as the accuracy of the prediction.
- a c + t air quality data matrix at time t in the future
- a + t matrix of historical air quality data at time t after the historical air quality data frame time
- the future air quality prediction method after time t is based on the historical air quality data frame and difference characteristics at time t after calculating the difference characteristics of historical air quality data frames, and the difference characteristic analysis method is applied backward to obtain time t Future air quality data. That is, the matrix of the historical air quality data frame at time t after the time point of the historical air quality data frame and the difference eigenvalues are used to perform matrix addition operations to obtain future air quality data.
- ⁇ is the percentage of the similarity coefficient, which can be given to the corresponding prediction situation as the accuracy of the prediction.
- a c + t air quality data matrix at time t in the future
- a + t matrix of historical air quality data at time t after the historical air quality data frame time
- Figure 1 is a schematic diagram of an air quality prediction process
- Figure 2 is a vector schematic diagram
- FIG. 3 is a characteristic vector diagram of historical air quality data from time T 1 to time T 2 ;
- Figure 4 is a vector schematic diagram
- 5 is a schematic diagram of air quality data characteristics of each grid from 10:40 to 10:50;
- FIG. 6 is a schematic diagram of air quality data characteristics in the area to be measured from 10:40 to 10:50;
- Figure 7 is versus Schematic diagram of the angle ⁇ .
- the predicted area in this example is a square area centered on a subway station in Jinan City, with a surrounding area of 150m * 150m and an area of 22500m 2.
- the pollution process continues from 9 am on January 23, 2019 to January 23, 2019 At 5 pm, the air quality data frames are taken for monitoring and analysis of pollutant drift paths at 10: 40-10: 50 am.
- the resolution of the air quality data frames is 50M * 50M.
- This embodiment takes a data unit as an example to introduce a method of vector prediction.
- the area is divided into grids in the unit of 50M * 50M, then this area contains 9 grids, and the bottom left grid is taken as an example to analyze the pollutant drift path.
- the historical air quality data at the time of 10:40 and 10:50 on January 23, 2019 are filled into the divided grid based on the geographic location information, and the air quality data frame at 10:40 and the air quality at 10:50 are obtained. Data Frame.
- the geographic location of air pollution data is often not continuous.
- the air quality data in the grid without data can be obtained using mathematical algorithms such as interpolation based on the existing air quality data.
- January 23, 2019 at 10:40 and 10 am The pollution statistics (PM 2.5 concentration value) at the two moments of 50 are shown in the following table:
- the air quality data frame at time 10:40 is converted into the A matrix according to the grid.
- the values of the elements in the first row and the first column of the A matrix are the air quality of the first row and the first column in the data frame at 10:40.
- the value of the element in the mth row and the nth column in the A matrix is the air quality of the mth row and the nth column in the data frame at time 10:40, and the A matrix represents the data frame of the degree of pollution in the area at 10:40 time.
- the pollution data frame at 10:50 is converted into a B matrix.
- the grid number direction is consistent with the matrix arrangement direction, with the y axis going down).
- the grids are numbered according to the grid's ranking order in the x and y directions).
- the coordinate grid table is as follows:
- Figure 4 is a vector schematic diagram.
- FIG. 5 shows the above vector, which is a feature of air quality data of each grid.
- the air quality data characteristics of the predicted area After determining the recent air quality data characteristics of the predicted area, compare the air quality data characteristics with the historical air quality data characteristics in the database, find the historical air quality data characteristics that are closest to the recent air quality data characteristics, and find a matching history. After the characteristics of the air quality data, the current air quality data is used as the basic data input, and vector calculation and fitting are performed to predict the future air quality.
- the current time is 8:00 on January 28, 2019, and the example of pollutant prediction using the lower left corner of the grid analyzes and predicts the air pollution situation on January 28, 2019 at 8:10.
- the first step in forecasting is to compare and fit the recent vector with the historical vector. Based on the historical air quality data feature database, the characteristics of the air quality data during the historical period are obtained.
- x 1 , y 1 is Vector coordinate values; x 2 , y 2 are The coordinate value of the vector.
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Computing Systems (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Molecular Biology (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
- Air Conditioning Control Device (AREA)
Abstract
现有空气质量预测方式有很多弊端，第一种需要大量的参数数据输入，并且对计算资源以及气象监控网络有极高的要求；第二种由于基于历史气象数据以及历史空气质量数据，同时需要两种数据，而且这种简单的基于历史气象数据和历史空气质量数据的预测方法准确度较低。本方案提供了对中小尺度进行精确和实时的预测的方法，并且可以准确预测污染飘移的移动方向和轨迹。
Description
本发明涉及一种利用历史空气质量数据特征预测空气污染的方法，属于环境监测领域。
在工业化过程中，人类对自然环境的破坏比较严重，产生了各种污染，而大气污染便是其中主要的污染之一。大气污染直接危害着人们的身体健康，因此，人们存在能够获知较为准确的未来日期的空气质量的需求，以便安排自己的工作和生活，并且，这一需求比较强烈。
空气质量数据一般主要来自环保部门的实测值。而目前的技术也有对空气质量进行预测的技术，主要分为两种：一种是基于化学预报模式计算进行预报，即建立扩散模型，基于化学预报模式计算进行预报的方法，需要极高的计算资源，难以实现；另一种是基于当地历史气象数据及历史空气质量数据，结合预测时刻气象数据，对当地预测时刻的空气质量数据进行预测，这种方法，对当地预测时刻的空气质量数据进行预测时，所选取的历史气象数据及历史空气质量数据为历史日期中与预测时刻相同时刻的数据，比如预测时刻为15：00，则所选取的历史气象数据及历史空气质量数据为历史某个或某些日期15：00时的数据，采用这种方法所预测得到的空气质量的数据准确度比较低。
上述两种空气质量预测方式有很多弊端，第一种需要大量的参数数据输入，并且对计算资源以及气象监控网络有极高的要求；第二种由于基于历史气象数据以及历史空气质量数据，同时需要两种数据，而且这种简单的基于历史气象数据和历史空气质量数据的预测方法准确度较低。这两种方法都需要气象数据，而且无法对中小尺度进行精确和实时的预测；无法准确预测污染飘移的移动方向和轨迹。
有相关人员对污染物预测进行了一些研究。
中国专利申请号：201510287229.2，发明名称：空气质量的预测方法和装置。该发明从利用预测地区和周边预设距离范围内所有地区的历史气象数据和历史空气质量数据，从中选取一定的历史气象数据和历史空气质量数据，将这些数据和当前的气象数据与空气质量数据一并作为输入参数，输入函数进行空气质量预测。该发明仍然需要气象数据，而且无法预测污染物飘移的路径和走向趋势。
中国专利申请号：201710818682.0，发明名称：空气质量预测方法及装置。该发明利用空气质量历史数据以及目标日期，简单通过分析历史空气质量在目标日期所可能的空气情况的概率进行预测。完全不利用现有监测数据，导致预测准确性大为降低并且有很可能出现不符合当时情况的预测。
发明内容
在先申请：PCT/CN2019/051245
术语
1.历史空气质量数据：历史某时刻的空气污染物浓度。
2.历史空气质量数据特征：历史某两个时刻间，表达被监测区域前一时刻对后一时刻周围区域污染物的贡献的向量。
3.历史空气质量数据特征库：由表征历史空气质量数据特征的向量组成的数据库。
4.空气质量数据帧：将待预测地区根据预测需求划分为设定密度的网格，将历史空气质量数据根据时间以及地理位置赋予网格内，生成以时间序列为基础的带有地理位置信息的空气质量数据集。
5.空气质量数据矩阵：将数据帧所包含的数据以矩阵的形式表达得到的矩阵即为空气质量数据矩阵。
6.近期空气质量数据：待预测污染过程截止到与历史数据进行比对之前所统计得到的空气质量数据。
7.近期空气质量数据特征：待预测污染过程截止到与历史数据进行比对之前的某两个时刻间，表达被监测区域前一时刻对后一时刻周围区域污染物的贡献的向量。
8.匹配程度系数：表示待预测污染过程的近期空气质量数据特征与待匹配历史空气质量数据特征的匹配程度的系数。
9.向量组：根据两时刻空气质量数据帧计算得到的包含每一个网格内特征向量表征污染物扩散路径的向量组。
10.时刻间空气质量数据特征：表征两时刻间污染物扩散信息的向量。
11.比值对比矩阵：由历史空气质量数据帧和当前空气质量数据帧的数据对比形成比值对比数据帧后，比值对比数据帧所对应的矩阵为比值对比矩阵。
12.相似网格区间：通过特征向量匹配方法，得到的一段与近期空气质量数据发展过程相似的历史空气质量数据。
13.差值对比矩阵：由历史空气质量数据帧和当前空气质量数据帧的数据的差值形成差值对比数据帧后，差值对比数据帧所对应的矩阵为差值对比矩阵。
为了克服现有技术中对空气质量预测需要风速风向等气象参数对计算资源要求很高的弊端，或者现有技术中依靠历史数据进行的简单概率预测的不足。本发明提供了一种利用历史空气质量数据特征预测空气污染的方法，利用较少的计算资源、不使用不易获得的近地面气象数据情况下，准确地预测空气污染以及污染飘移路径趋势。为实现上述目的，本发明提供如下技术方案：
建立历史空气质量数据库
获得历史空气质量数据、地理位置信息和待预测区域等信息，建立历史空气质量数据库，具体做法为：
确定待预测地区的地理区域范围；获取待预测区域范围内的带有地理位置信息的历史空气质量数据，这些带有地理位置信息的历史空气质量数据可以是国控站、超级站、空气质量监测微站和移动站等采集的数据。
建立网格化数据库
将待预测地区根据预测需求划分为设定密度的网格，将历史空气质量数据根据时间以及地理位置赋予网格内，生成带有二维地理位置信息的和时间戳信息的空气质量数据帧。
以时间信息为轴建立三维时空网格，将带有二维地理位置信息和空气质量数据的数据帧排列，生成三维时空网络。
网格内赋予带有地理位置信息和时间戳信息的空气质量方法：
1)在同一时刻内仅有一个空气质量信息落入了一个网格，该网格的空气质量数据是这一个空气质量信息。
2)在同一时刻内有多个空气质量信息落入了一个网格，该网格的空气质量数据为这些空气质量信息的平均值。
3)没有空气质量信息落入的网格，该网格的空气质量数据可以通过同一时刻，相邻空间的网格数据经插值、扩散模型等数学方法得来。
4)没有空气质量信息落入的网格，该网格的空气质量数据还可以通过同一空间，相邻时刻的网格数据经插值、扩散模型等数学方法得来。
5)每个二维地理网格的边长可以是10米-1000米。
6)同一时刻指该时刻点前后1分钟～1小时的空气质量数据。
向量化分析网格化数据库
分析历史空气质量数据库的特征
分析历史空气质量数据库特征，分析带有地理位置信息的空气质量数据帧，做法为对前一时刻数据帧内每个网格对后一时刻数据帧内的该网格相邻位置的污染贡献分析，得到这两个时刻间的历史空气质量数据特征，该特征是这两个时刻间每一个网格对周边的污染贡献，代表了这两个时刻间污染飘移的方向，多个时刻间的连续特征即能表示污染飘移的路径。
空气质量数据矩阵的获得方法：
将在T
1时刻的历史空气质量数据根据地理位置信息填充至划分的网格内，得到T
1时刻空气 质量数据帧。数据帧即是三维时空网络中，一个时刻下的带有二维地理位置信息的空气质量数据平面。
将T
1时刻空气质量数据帧根据网格转化为A矩阵，A矩阵中第一行第一列的元素的值即为T
1时刻空气质量数据帧中第一行第一列的空气质量，以此类推A矩阵中第m行第n列的元素的值即为T
1时刻空气质量数据帧中第m行第n列的空气质量，A矩阵代表T
1时刻区域内空气质量数据帧。
T
2时刻空气质量数据矩阵B同理通过上述方法获得，B矩阵代表T
2时刻区域内空气质量数据帧。
历史空气质量数据特征的获得方法：
分别计算前一时刻每一个网格对下一时刻相邻网格的污染贡献，每一个网格对相邻网格的污染贡献通过向量的计算方式获得。
| B (m-1)(n-1) | B (m-1)n | B (m-1)(n+1) |
| B m(n-1) | A mn | B m(n+1) |
| B (m+1)(n-1) | B (m+1)n | B (m+1)(n+1) |
得到分析矩阵
A
mn为前一时刻的空气质量数据，B
mn为后一时刻空气质量数据。上文中T
1为前一时刻空气质量数据，T
2后一时刻空气质量数据。
历史空气质量数据的向量特征：T
1时刻中A
mn区域对T
2时刻周边网格污染漂移可以用向量的方式表示，根据的污染漂移方向可以人为的分为三种方式。
方式一：T
1时刻中A
mn对其上、下、左、右四个方向在T
2时刻有漂移的贡献。也就是A
mn对B
(m-1)n、B
m(n-1)、B
m(n+1)和B
(m+1)n)有贡献。污染漂移向量为
那么方式一的污染漂移向量计算方法如下：
综合特征向量：
方式二：T
1时刻中A
mn对其左上、左下、右上、右下四个方向在T
2时刻有漂移的贡献。也就是A
mn对B
(m-1)(n-1)、B
(m-1)(n+1)、B
(m+1)(n-1)、B
(m+1)(n+1)有贡献。污染漂移向量为
那么方式二的污染漂移向量计算方法如下：
综合特征向量：
方式三：T
1时刻中A
mn对其上、下、左、右、左上、左下、右上、右下八个方向在T
2时刻有漂移的贡献。也就是A
mn对B
(m-1)(n-1)、B
(m-1)n、B
(m-1)(n+1)、B
m(n-1)、B
m(n+1)、B
(m+1)(n-1)、B
(m+1)n)、B
(m+1)(n+1)有贡献。污染漂移向量为
那么方式三的污染漂移向量计算方法如下：
综合特征向量：
分析近期空气质量数据特征
分析近期空气质量数据特征的方法与建立历史空气质量数据特征库的方式类似。首先将空气质量监测站的带有地理位置信息的空气质量数据赋值于待预测区域已经划分好的网格内，生成当前空气质量数据帧和以当前时刻为基准向前一定时刻的一个或多组数据帧，进而生 成当前空气质量和向前一定时刻的空气质量数据矩阵。
利用上述建立的历史空气质量数据特征库中提到的建立历史空气质量数据特征的获得方法和空气质量数据矩阵的获得方法，可以得到近期空气质量数据变化的向量组，这个向量组就是近期空气质量数据特征。
A
c表示当前污染数据矩阵
B
c表示当前向前一定时刻t
1的污染数据矩阵
数值分析网格化数据库
用比值或差值分析的方法对近期空气质量数据与历史空气质量数据进行分析。
将近期空气质量数据单元与历史空气质量数据单元进行比较，找到和选定近期空气质量数据帧最接近的，或者针对近期空气质量数据帧中每一个数据都成一定比例放大或缩小的历史空气质量数据帧，也就找到了与当前空气质量或者说当前污染状况、运动路径最接近的历史空气质量运动过程或者污染过程。
近期与历史空气质量数据的分析举例：
有如下历史空气质量数据帧与近期空气质量数据帧，现将两数据帧进行对比，以确定A数据帧所在历史污染过程是否可以对近期污染过程的发展趋势进行预测。
历史空气质量数据帧
近期空气质量数据帧
由历史空气质量数据帧和当前空气质量数据帧得到对应A矩阵和A
c矩阵。
比值分析方法
ε代表近期数据单元与同一位置历史数据单元的比值。首先选取近期数据帧中的单个数据A
c
11与历史对应地理位置数据A
11进行运算求得比值
同理，求得该区域内所有网格数据的
分别对比值ε的矩阵中的元素的数值进行统计计算ε的平均值，
可以为比值特征值。
A-A
c比值对比数据帧
由比值数据帧可以得到比值对比矩阵
ε的对比矩阵
差值分析方法
δ代表近期数据单元与同一位置历史数据单元的差值绝对值。首先选取近期数据帧中的单个数据A
c
11与历史对应地理位置数据A
11进行运算求得当前数据与历史数据的差值δ
11＝A
c
11-A
11，同理，求得该区域内所有网格数据的δ
mn＝A
c
mn-A
mn，分别对比值δ的矩阵中的元素的数值进行统计计算δ的平均值，δ即为差值特征值。
A-A
c差值对比数据帧
由差值可以得到差值对比矩阵
差值对比矩阵
匹配近期空气质量数据与历史空气质量数据
特征向量匹配方法
确定了近期空气质量数据特征后，将近期空气质量数据特征与历史空气质量数据特征进行比较，得到匹配系数η。可以根据向量最短距离、夹角余弦、马氏距离、向量相似度等数学方法，找到和近期空气质量数据特征最接近的历史空气质量数据特征，也就找到了与当前空气质量或者说当前污染状况、运动路径最接近的历史空气质量运动过程或者污染过程。
在匹配过程中，应当考虑其他影响因子f如下：
设定温度影响因子(factor temperature)：f
t与温度参数(t
temp)有关，当高于一定温度时候，或者低于一定温度时候，是极端的高温或低温，不适合预测。
设定湿度影响因子(factor humidity)：f
h，与湿度参数(h
humid)有关，当高于一定湿度时候，不适合预测。
季节影响因子(factor season)：f
s与季节参数(s
season)有关，若历史空气质量数据与近 期空气质量数据所处的季节、温湿度接近或相同，则匹配程度较高；若历史空气质量数据与近期空气质量数据所处的季节、温湿度相差较大，则影响因子值较小。
设定极端天气影响因子(factor weather)：f
w与天气情况参数(w
weather)有关，历史空气质量数据与近期空气质量数据所处时间有极端天气，则极端天气影响因子数值很低，当时数据不能作为预测参考，当前情况也不适合进行预测。
设定路径影响因子(factor path)：f
p与污染路径参数(p
path)有关，历史空气质量数据与近期空气质量数据中，污染移动路径的以及污染源头的距离应做考虑。若近期空气质量数据特征与多个历史空气质量数据特征的污染物漂移路径有较高的匹配度，那么与近期空气质量数据特征中路径距离更近的历史空气质量数据特征路径的匹配度更高。
设定扩散程度影响因子(factor diffusion)：f
d与扩散程度参数(d
diffusion)有关，在有相似的污染物扩散路径的情况下，两次污染过程的扩散程度应当作为参数。若两次污染过程的扩散程度、影响范围的区域面级相似，则匹配程度系数应相应增大，若两次污染过程的扩散程度、影响范围的区域面级相差较大，则匹配系程度系数相应减小；
设定重复性影响因子(factor repeatability)：f
r与重复次数参数(r
repeatability)有关(r
repeatability)，历史相似污染过程发生的次数应当作为参数。若待预测污染过程所匹配的相似历史污染过程模型曾多次发生，则认为使用该历史空气质量数据所进行预测的可靠性较强，应设置较高的匹配程度系数；
设定地形影响因子(factor geo)：f
g与地形参数(g
geo)有关，历史与待预测污染过程的产生地点所处的地形因素也应当作为参数。若两次污染过程的产生地地形相差较大，例如其中一次是在数年前当地没有较多建筑，而另一次是在较过去有了很多建筑的情况下，数年前没有建筑的匹配程度系数相应减小。
其他影响因子判定方法表：
| 影响因素 | 影响参数 | 条件 | 影响因子 |
| 湿度 | h humid | 湿度＞90％ | f h→0 |
| 湿度 | h humid | 湿度＜90％ | f h→1 |
| 温度 | t temp | 温度＞40℃或温度＜-20℃ | f t→0 |
| 温度 | t temp | -20℃＜温度＜40℃ | f t→1 |
| 极端天气 | w weather | 沙尘暴、暴雨、暴风雪等 | f w→0 |
| 路径 | p path | 路径重复次数大于5次 | f p→1 |
| 扩散、影响范围 | d diffusion | 扩散、影响范围相似 | f d→1 |
| 地形 | g geo | 周边有高大建筑物遮挡 | f g→0 |
夹角余弦对比匹配方法：
要进行特征对比匹配则需要将近期空气质量数据特征与历史空气质量数据特征进行对比，即将
与
进行对比。
t时刻第m行，第n列方格应用夹角余弦法的匹配系数
余弦相似度用向量空间中两个向量夹角的余弦值作为衡量两个个体间差异的大小。余弦值越接近1，就表明夹角越接近0度。同理，表征被监测区域在特定时刻近期与历史空气质量特征的向量
与
的对比也可以用夹角余弦法完成，计算过程同上(注：针对被监测时段获得的所有空气质量数据。
t时刻的整个网格范围内应用夹角余弦法匹配系数η
t：
一段时间T内某一时刻第m行，第n列方格应用夹角余弦法的匹配系数
q代表T时间段内进行匹配的次数：
一段时间T内整个网格范围内应用夹角余弦法的匹配系数η
T，q代表T时间段内进行匹配的次数：
考虑到其他影响因子，匹配系数的修正公式为：
η′＝f×η
η′为修正后匹配系数
匹配系数越接近于1，则匹配性越好，当前特征与历史空气质量数据特征库的某一段特征越相似，未来发生的可能性越大；匹配系数越接近于0，则匹配性越差，当前特征与历史空气质量数据特征库的某一段特征越不相似，越不可能发生。
| 匹配系数 | 匹配程度 |
| η′→0 | 越低 |
| η′→1 | 越高 |
历史空气质量数据特征的选择：历史空气质量数据特征的选择可以通过对匹配系数直接排名的方式确定，还可以通过筛选后排名的方式确定。
直接排名的方式为将匹配系数或者修正后的匹配系数进行从大到小进行排列，匹配系数或者修正后的匹配系数越接近1，排名越靠前，最终选取排名前10％、20％或者排名前一定比例的匹配系数，该匹配系数所对应的历史空气质量数据特征即可用于对未来空气质量的预测。
例如：总共有10个匹配系数，这些匹配系数及排名情况见下表：
| 匹配系数 | η 5 | η 3 | η 1 | η 4 | η 2 | η 7 | η 6 | η 9 | η 8 | η 10 |
| 匹配系数值 | 0.98 | 0.97 | 0.95 | 0.9 | 0.89 | 0.8 | 0.8 | 0.75 | 0.72 | 0.7 |
| 排名 | 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 | 10 |
根据上表，排名前20％的匹配系数入选，即η
5和η
3入选，入选的匹配系数所对应的历史空气质量数据特征可用于对未来空气质量数据的预测。
筛选后排名的方式为先通过客观因素对历史空气质量特征的情况进行一轮筛选，再对筛选后的匹配系数进行排名的方式，确定所需的匹配系数以及对应的历史空气质量数据特征。用于筛选的客观因素包括历史空气质量数据特征当时的天气情况(如历史时刻出现极端天气情况：暴雨、极寒、高温等，则该历史时刻排除)、历史空气质量数据距离当前时刻时间久远(如3年以上的历史数据)。经过筛选排除后，再进行排名并选择对匹配系数。排名方式可以为从大到小进行排列，匹配系数越接近1，排名越靠前，最终选取排名前10％、20％或者排名前一定比例的匹配系数，该匹配系数所对应的历史空气质量数据特征即可用于对未来空气质量的预测。
例如：总共有10个匹配系数，这些匹配系数及排名情况见下表：
| 匹配系数 | η 5 | η 3 | η 1 | η 4 | η 2 | η 7 | η 6 | η 9 | η 8 | η 10 |
| 匹配系数值 | 0.98 | 0.97 | 0.95 | 0.9 | 0.89 | 0.8 | 0.8 | 0.75 | 0.72 | 0.7 |
| 历史客观因素 | 高温 | - | 3年前数据 | - | 暴雨 | - | - | - | - | - |
| 排名 | - | 1 | - | 2 | - | 3 | 4 | 5 | 6 | 7 |
比值特征值匹配方法
在求得比值对比矩阵之后，根据平均值，众数，数据分布，线性回归等数学方法统计符合设定要求的ε
mn元素的比例。
当达到设定标准的ε的元素比例占ε元素总数量比例超过一定限度，则认为历史空气质量数据帧A与近期空气质量数据帧A
c相似。
此处以平均值法举例：设定符合要求的ε范围的为在
之内，即先计算比值数据帧元素的平均值
如果ε
mn的数值在
范围内，则认为ε
mn符合规定。符合要求的ε范围可以人为设定，可以设定为比值平均值的倍数，或者固定的范围。
差值特征值匹配方法
在求得差值对比矩阵之后，根据平均值，众数，数据分布，线性回归等数学方法统计符合设定要求的δ
mn元素的比例。
分别对差值δ的矩阵中的元素的数值进行统计和计算，当达到设定标准的元素比例占总元素比例超过一定限度，则认为历史空气质量数据帧A与近期空气质量数据帧A
c相似。
此处以平均值法举例：设定符合要求的δ范围为在
之内，即先计算差值数据帧的平均值
如果δ
mn的数值在
之内，则认为δ
mn符合规定。符合要求的δ范围可以人为设定，可以设定为差值平均值的倍数，或者固定的范围。
定义λ为相似系数，此处λ为对比矩阵中符合规定的元素占对比矩阵中元素总数量的比值。κ为相似度参数，代表设定的相似度标准，可以人为设定不同数值，如设定为κ＝0.8，则认为λ超过κ＝0.8两张数据帧相似，即两张数据帧匹配，大于零小于κ则不相似，即不匹配。λ反映了预测的准确率，λ越高代表预测准确率越高。
κ为相似度参数，k＝0.8、0.9等
基于匹配结果和当前空气质量情况对未来空气质量预测
向量特征预测
当得到匹配度最高的一段或几段相似网格区间后，以当前空气质量数据为基准(已知当前空气质量数据)，以及已知历史匹配网格的时刻的后续时刻的污染飘移向量(后续时刻的污染飘移向量的计算方式仍为污染飘移向量计算方法)，逆向应用污染飘移向量计算方法对未来空气质量数据进行推算和预测，得到所需的未来空气质量数据，即使用当前空气质量数据帧的元素与对应的特征向量进行矩阵运算，得到在未来，当前元素位置周边位置的空气质量数据信息，以此类推计算每个数据帧的元素，即可得到未来空气质量的数据。
在使用向量特征进行预测时，如果当有多段相似的区间可以用于未来空气质量的预测，匹配系数或者修正后匹配系数的百分数可以作为预测的准确率赋予对应的预测情况。
在进行预测的时候，需要计算以目标网格为中心的对周围网格对其的贡献值，然而处在统计区域边界的数据(网格数据)则会因为四周数据不全而出现计算障碍的情况，针对这种情况，以下为两种处理方法：
改进边界处理方法1：数据边界内缩法
将统计区域边界最外侧的一圈单元格内缩形成新的统计区域，原始的边界区域网格数据不作为预测的基础，原始的边界区域的网格数据用于向量特征的计算。则新统计区域的所有网格数据可以作为预测的基础数据以及参与向量特征计算。
A
c数据帧内缩后举例
举例说明，这种情况下，第一行、第一列，最后一行和最后一列的数据都不作为预测的基础数据，但是都可以作为向量计算的基础数据。内缩后，可以用于预测的基础数据区域为A
c
22、A
c
2(n-1)、A
c
(m-1)2、A
c
(m-1)(n-1)网格内的数据(包含刚才四个坐标)。
改进边界处理方法2：数据边界扩展法
将统计区域边界单元格外扩一圈形成新的统计区域，扩展的方法可以是原始边界数值直接复制作为新边界的数值，或者使用差值的方式向外扩展一圈。原始数据区域既可以用于预测的基础，还可以用于向量特征的计算。向外新扩展边界的数据不可作为预测的基础，但新向外新扩展边界的数据可以用于向量特征的计算。
举例说明，这种情况下，第一行、第一列，最后一行和最后一列的数据为扩展出来的，都不作为预测的基础数据，但是都可以作为向量计算的基础数据。原始数据区域为A
c
11、A
c
1n、A
c
m1、A
c
mn网格内的数据(包含刚才四个坐标)，这个区域内的数据还可以用于预测的数据基础。
比值特征预测
当根据相似系数，得到与近期空气质量数据匹配的一张或者多张历史空气质量数据帧的时候。t时刻后的未来空气质量预测方法，是以计算比值特征的历史空气质量数据帧时间点之后t时刻的历史空气质量数据帧及比值特征为基础，逆向应用比值特征分析方法得到t时刻后的未来的空气质量数据。即使用历史空气质量数据帧时间点之后t时刻的历史空气质量数据帧的矩阵，以及比值特征值，进行矩阵乘法运算得到未来空气质量数据。
λ为相似系数的百分数可以作为预测的准确率赋予对应的预测情况。
A
c+t：未来t时刻的空气质量数据矩阵
A
+t：历史空气质量数据帧时间点之后t时刻的历史空气质量数据的矩阵
差值特征预测
当根据相似系数，得到与近期空气质量数据匹配的一张或者多张历史空气质量数据帧的时候。t时刻后的未来空气质量预测方法，是以计算差值特征的历史空气质量数据帧时间点之后t时刻的历史空气质量数据帧及差值特征为基础，逆向应用差值特征分析方法得到t时刻后的未来的空气质量数据。即使用历史空气质量数据帧时间点之后t时刻的历史空气质量数据帧的矩阵，以及差值特征值，进行矩阵加法运算得到未来空气质量数据。
λ为相似系数的百分数可以作为预测的准确率赋予对应的预测情况。
A
c+t：未来t时刻的空气质量数据矩阵
A
+t：历史空气质量数据帧时间点之后t时刻的历史空气质量数据的矩阵
附图简要说明
图1为空气质量预测流程示意图；
图2为向量
示意图；
图3为T
1时刻到T
2历史空气质量数据特征向量示意图；
图4为向量
示意图；
图5为10：40时刻至10：50时刻每网格空气质量数据特征示意图；
图6为10：40时刻至10：50时刻待测区域内空气质量数据特征示意图；
图7为
与
的夹角θ示意图。
此实施例中被预测区域为以济南市某地铁站为中心，周围150m＊150m，面积为22500m
2的正方形区域，污染过程自2019年1月23日上午9时持续到2019年1月23日下午5时，取上午10：40-10：50空气质量数据帧进行污染物漂移路径监测分析，空气质量数据帧的分辨率为50M＊50M。本实施方式以一个数据单元为例介绍向量预测的方法。
将区域内以50M＊50M为单位划分成为网格状，则本区域内包含9个网格，取左下角网格为例进行污染物漂移路径分析。将2019年1月23日上午10：40与10：50两时刻的历史空气质量数据根据地理位置信息填充至划分的网格内，得到10：40时刻空气质量数据帧和10：50时刻空气质量数据帧。空气污染数据的地理位置时常不是连续的，没有数据的网格内的空气质量数据可以依据已有空气质量数据利用插值等数学算法得来，最终得到2019年1月23日上午10：40与10：50两时刻污染统计数据(PM
2.5浓度值)如下表所示：
10：40分空气质量数据表格
| 100 | 75 | 50 |
| 135 | 110 | 85 |
| 170 | 145 | 120 |
10：50分空气质量数据表格
| 150 | 135 | 120 |
| 165 | 150 | 135 |
| 180 | 165 | 150 |
将10：40时刻空气质量数据帧根据网格转化为A矩阵，A矩阵中第一行第一列的元素的值即为10：40时刻数据帧中第一行第一列的空气质量，以此类推A矩阵中第m行第n列的元素的值即为10：40时刻数据帧中第m行第n列的空气质量，A矩阵代表10：40时刻区域内污染程度数据帧。同理，将10：50时刻污染数据帧转化为B矩阵。
分别计算10：40时刻每一个网格对10：50时刻相邻网格的污染贡献，每一个网格对相邻网格的污染贡献通过向量的计算方式获得。如下表格：
| B (m-1)(n-1) | B (m-1)n | B (m-1)(n+1) |
| B m(n-1) | A mn | B m(n+1) |
| B (m+1)(n-1) | B (m+1)n) | B (m+1)(n+1) |
根据编号规则对10：40与10：50两个网格区域进行编号(以网格区域左上角网格的左上角顶点为原点，横向为x轴，纵向为y轴建立坐标网格(为保证网格编号方向与矩阵排列方向相符，y轴正向向下)，按照网格在x轴与y轴方向上的排位次序对网格进行编号)，坐标网格表格如下：
10：40时刻空气质量坐标网格
| A 11 100 | A 12 75 | A 13 50 |
| A 21 135 | A 22 110 | A 23 85 |
| A 31 170 | A 32 145 | A 33 120 |
10：50时刻空气质量坐标网格
| B 11 150 | B 12 135 | B 13 120 |
| B 21 165 | B 22 150 | B 23 135 |
| B 31 180 | B 32 165 | B 33 150 |
则10：40时刻左下角网格区域对10：50时刻周边区域的污染贡献为
向量计算方式如下：
同理分别计算向量
分别进行计算，图5表示上述向量，该向量图为每个网格的空气质量数据特征。
在确定了被预测区域近期空气质量数据特征后，将空气质量数据特征与数据库中历史空气质量数据特征进行比较，找到和近期空气质量数据特征最接近的历史空气质量数据特征，在找到匹配的历史空气质量数据特征之后，以当前的空气质量数据为基础数据输入，进行向量的运算和拟合，即可对未来空气质量进行预测。
当前时刻为2019年1月28日8：00，以左下角网格进行污染物预测的实施例分析预测2019年1月28日8：10空气污染情况。进行预测的第一步即将近期向量与历史向量进行对比拟 合，根据历史空气质量数据特征库，得到历史同期时刻间空气质量数据特征
利用同样的计算方式，得到2019年1月28日07：50至2019年1月28日8：00间空气质量数据特征。将
与
进行对比。
根据夹角余弦公式：
x
1，y
1为
向量的坐标值；x
2，y
2为
向量的坐标值。
则
0.9998≈1，所以认为这两个向量的匹配度极高，可以利用此历史数据对污染物进行预测，已知与
相匹配的历史空气质量数据在下一10分钟梯度的时刻间空气质量数据特征
其中向量的分量信息为
通过监测数据已知08：00时刻B
31网格的污染物浓度为180，则根据以上信息预测得到11：00各网格的污染浓度为B
22＝160，B
21＝174，B
32＝174。
8：10左下网格周边的空气污染情况预测结果：
| 174 | 160 | |
| 174 |
Claims (11)
- 一种利用历史空气质量数据特征预测空气污染的方法，包含如下步骤：1)建立历史空气质量数据库：所述历史空气质量数据库中包含各种空气质量监测设备获取的历史空气质量数据；所述历史空气质量数据包含二维地理位置信息和时间戳信息；2)建立三维时空网格：所述三维时空网格以二维地理位置和时间为轴；3)建立网格化数据库：每个网格内产生唯一一个数据单元；所述数据单元由落入该网格的历史空气质量数据计算得出；或者由该网格的邻近网格推算得出；4)对网格化数据库的数据进行比值分析，对近期空气质量数据与历史空气质量数据进行分析得到比值对比矩阵和比值特征值；5)将网格化数据库中当前空气质量的数据帧和历史空气质量数据帧进行匹配，所述匹配方法为对比值对比矩阵的元素进行统计，统计比值对比矩阵中符合设定要求的元素比例，所述比例超过设定值，则得到与当前空气质量的数据帧相似的历史空气质量数据帧；6)以相似的历史空气质量数据帧之后一段时间的数据单元的空气质量数据和比值特征值为基础，预测未来时段的数据单元的数值；所述比值分析包含的步骤如下：1)先建立分析矩阵由历史空气质量数据帧和当前空气质量数据帧的数据得到对应的A矩阵和A c矩阵；
2)计算比值与比值特征值计算矩阵中对应位置的比值并对得到的ε计算平均值
为比值特征值。
- 一种利用历史空气质量数据特征预测空气污染的方法，包含如下步骤：1)建立历史空气质量数据库：所述历史空气质量数据库中包含各种空气质量监测设备获取的历史空气质量数据；所述历史空气质量数据包含二维地理位置信息和时间戳信息；2)建立三维时空网格：所述三维时空网格以二维地理位置和时间为轴；3)建立网格化数据库：每个网格内产生唯一一个数据单元；所述数据单元由落入该网格的历史空气质量数据计算得出；或者由该网格的邻近网格推算得出；4)对网格化数据库的数据进行差值分析，对近期空气质量数据与历史空气质量数据进行分析得到差值对比矩阵和差值特征值；5)将网格化数据库中当前空气质量的数据帧和历史空气质量数据帧进行匹配，所述匹配方法为对差值对比矩阵的元素进行统计，统计差值对比矩阵中符合设定要求的元素比例，所述比例超过设定值，则得到与当前空气质量的数据帧相似的历史空气质量数据帧；6)以相似的历史空气质量数据帧之后一段时间的数据单元的空气质量数据和差值特征值为基础，来\预测未来时段的数据单元的数值；所述差值分析包含的步骤如下：1)先建立分析矩阵由历史空气质量数据帧和当前空气质量数据帧的数据得到对应的A矩阵和Ac矩阵；
2)计算差值与差值特征值计算矩阵中对应位置的差值δ mn＝A c mn-A mn，并对得到的δ计算平均值为差值特征值。
- 如权利要求1或2所述的方法，其特征在于，所述匹配方法为计算符合规定的元素数量与对比矩阵元素总数量的比值(λ)，计算和匹配方法如下：
λ为相似系数，κ为相似度参数，κ的值为0.7、0.8，或0.9。 - 如权利要求1所述的方法，其特征在于，所述设定要求是如下之一：1)ε的平均值加减0.1；2)ε的平均值加减0.3；3)ε的平均值加减0.4；4)0.3～1；5)0.5～1；6)0.7～1。
- 如权利要求2所述的方法，其特征在于，所述设定要求是如下之一：1)δ的平均值加减0.1倍δ；2)δ的平均值加减0.15倍δ；3)δ的平均值加减0.25倍δ；4)δ的平均值加减0.3倍δ；5)δ的平均值加减0.4倍δ。
- 如权利要求1或2所述的方法，其特征在于，所述三维时空网格中每个网格的特征为：二维地理边长：10米～1000米；时长：1分钟～1小时。
- 如权利要求1或2所述的方法，其特征在于，所述部分数据单元不包括位于三维时空网格边界处的网格的数据单元。
- 如权利要求1或2所述的方法，其特征在于，步骤5)中所述相似的历史空气质量数据帧要经过筛选，所述筛选方式为历史空气质量数据帧所在时刻如出现极端天气，则该历史空气质量数据帧不相似，所述极端天气包括如暴雨、极寒或高温；所述筛选方式为历史空气质量时刻距离当前时刻超过一定时间，则该历史空气质量数据帧不相似，所述一定时间为1年、2年或3年。
- 如权利要求1所述的方法，其特征在于，所述预测方法为：
A c+t：未来t时刻的空气质量数据矩阵；A +t：历史空气质量数据帧时间点之后t时刻的历史空气质量数据的矩阵；比值特征值。
- 如权利要求2所述的方法，其特征在于，所述预测方法为：
A c+t：未来t时刻的空气质量数据矩阵；A +t：历史空气质量数据帧时间点之后t时刻的历史空气质量数据的矩阵；差值特征值。
- 如权利要求3所述的方法，其特征在于，相似系数λ的百分数作为预测的准确率赋予对应的预测情况。
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201980089816.7A CN113330456B (zh) | 2018-08-25 | 2019-08-25 | 一种利用历史空气质量数据特征预测空气污染的方法 |
Applications Claiming Priority (4)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810965453 | 2018-08-25 | ||
| CN201810965453.6 | 2018-08-25 | ||
| PCT/IB2019/051245 WO2020044127A1 (zh) | 2018-08-25 | 2019-02-15 | 一种大气污染预测的方法 |
| IBPCT/IB2019/051245 | 2019-02-15 |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| WO2020043028A1 true WO2020043028A1 (zh) | 2020-03-05 |
Family
ID=69643172
Family Applications (3)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| PCT/CN2019/102416 WO2020043027A1 (zh) | 2018-08-25 | 2019-08-25 | 一种大气污染预测中对区域边界优化的方法 |
| PCT/CN2019/102418 WO2020043029A1 (zh) | 2018-08-25 | 2019-08-25 | 一种通过向量化分析预测大气污染的方法 |
| PCT/CN2019/102417 WO2020043028A1 (zh) | 2018-08-25 | 2019-08-25 | 一种利用历史空气质量数据特征预测空气污染的方法 |
Family Applications Before (2)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| PCT/CN2019/102416 WO2020043027A1 (zh) | 2018-08-25 | 2019-08-25 | 一种大气污染预测中对区域边界优化的方法 |
| PCT/CN2019/102418 WO2020043029A1 (zh) | 2018-08-25 | 2019-08-25 | 一种通过向量化分析预测大气污染的方法 |
Country Status (1)
| Country | Link |
|---|---|
| WO (3) | WO2020043027A1 (zh) |
Families Citing this family (15)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113361556B (zh) * | 2020-03-06 | 2023-12-26 | 北京市环境保护监测中心 | 污染相似案例自动识别的方法和装置 |
| CN111428918B (zh) * | 2020-03-12 | 2023-06-30 | 北京农业信息技术研究中心 | 一种重金属衰减污染源的土壤污染范围预测方法及系统 |
| CN111353653B (zh) * | 2020-03-13 | 2020-12-11 | 大连理工大学 | 一种光伏出力短期区间预测方法 |
| CN111523717B (zh) * | 2020-04-15 | 2023-05-16 | 北京工业大学 | 一种大气污染物排放清单的反演估算方法 |
| CN111612245A (zh) * | 2020-05-18 | 2020-09-01 | 北京中科三清环境技术有限公司 | 大气污染状况预测方法、装置、电子设备及存储介质 |
| CN114020804A (zh) * | 2020-08-06 | 2022-02-08 | 安徽摩力孚再生资源有限公司 | 一种利用时序数据、视频数据监控排废的方法 |
| CN111932146B (zh) * | 2020-09-01 | 2024-09-17 | 平安国际智慧城市科技股份有限公司 | 污染成因的分析方法、装置、计算机设备和可读存储介质 |
| CN112634113B (zh) * | 2020-12-22 | 2023-09-26 | 山西大学 | 一种基于动态滑窗的污染废气关联分析方法 |
| CN113536577B (zh) * | 2021-07-21 | 2023-08-22 | 南通大学 | 一种基于空间代表性的pm2.5观测站点优化布局方法 |
| CN114646137A (zh) * | 2022-03-11 | 2022-06-21 | 珠海格力电器股份有限公司 | 净化规划方法、非易失性存储介质以及空气净化器 |
| CN117542434B (zh) * | 2023-11-20 | 2024-04-05 | 复旦大学 | 一种基于数据分析的空气污染物浓度分析方法 |
| CN118425423A (zh) * | 2024-04-28 | 2024-08-02 | 江苏中科三可科技有限公司 | 大气污染物浓度数据预测方法、装置、设备以及存储介质 |
| CN118313518B (zh) * | 2024-06-06 | 2024-08-23 | 中科三清科技有限公司 | 预报网格区域的未来贡献量的方法与装置 |
| CN118314482B (zh) * | 2024-06-07 | 2024-08-27 | 山东益来环保科技有限公司 | 一种无人机智能环保监测系统及方法 |
| CN118428823B (zh) * | 2024-07-02 | 2024-09-27 | 中科三清科技有限公司 | 监测站点位置的确定方法、装置、存储介质与电子设备 |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20130110399A1 (en) * | 2011-10-31 | 2013-05-02 | Insurance Bureau Of Canada | System and method for predicting and preventing flooding |
| CN107292417A (zh) * | 2017-05-09 | 2017-10-24 | 北京市环境保护监测中心 | 基于重污染序列案例库的区域重污染判别预报方法及装置 |
| CN107748932A (zh) * | 2017-10-20 | 2018-03-02 | 杭州尚青科技有限公司 | 一种融合序列模式挖掘和代价敏感学习的空气质量等级预测方法 |
| CN107991722A (zh) * | 2017-12-25 | 2018-05-04 | 北京墨迹风云科技股份有限公司 | 天气预测模型的建立方法、预测方法和预测装置 |
Family Cites Families (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| KR100920529B1 (ko) * | 2007-09-28 | 2009-10-09 | 한국전력공사 | 염해 오손 예측 방법 |
| US20160125307A1 (en) * | 2013-06-05 | 2016-05-05 | Yu Zheng | Air quality inference using multiple data sources |
| CN104850734B (zh) * | 2015-04-21 | 2017-09-15 | 武大吉奥信息技术有限公司 | 一种基于时空分布特征的空气质量指数预测方法 |
| CN104850913B (zh) * | 2015-05-28 | 2019-03-26 | 深圳先进技术研究院 | 一种空气质量pm2.5预测方法及系统 |
| CN106408417A (zh) * | 2016-11-18 | 2017-02-15 | 洪志令 | 一种基于近期预测涨跌幅及其一致性统计的股票推荐方法 |
| CN106780245B (zh) * | 2016-11-28 | 2020-06-02 | 中国科学院深圳先进技术研究院 | 热带气旋引发沿海地区灰霾概率的确定、预报方法及系统 |
| CN106874674B (zh) * | 2017-02-20 | 2019-04-19 | 上海海洋大学 | 一种面向海洋领域的多元时间序列相似性度量方法 |
-
2019
- 2019-08-25 WO PCT/CN2019/102416 patent/WO2020043027A1/zh active Application Filing
- 2019-08-25 WO PCT/CN2019/102418 patent/WO2020043029A1/zh active Application Filing
- 2019-08-25 WO PCT/CN2019/102417 patent/WO2020043028A1/zh active Application Filing
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20130110399A1 (en) * | 2011-10-31 | 2013-05-02 | Insurance Bureau Of Canada | System and method for predicting and preventing flooding |
| CN107292417A (zh) * | 2017-05-09 | 2017-10-24 | 北京市环境保护监测中心 | 基于重污染序列案例库的区域重污染判别预报方法及装置 |
| CN107748932A (zh) * | 2017-10-20 | 2018-03-02 | 杭州尚青科技有限公司 | 一种融合序列模式挖掘和代价敏感学习的空气质量等级预测方法 |
| CN107991722A (zh) * | 2017-12-25 | 2018-05-04 | 北京墨迹风云科技股份有限公司 | 天气预测模型的建立方法、预测方法和预测装置 |
Also Published As
| Publication number | Publication date |
|---|---|
| WO2020043029A1 (zh) | 2020-03-05 |
| WO2020043027A1 (zh) | 2020-03-05 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| WO2020043028A1 (zh) | 一种利用历史空气质量数据特征预测空气污染的方法 | |
| CN113348471B (zh) | 一种大气污染预测中对区域边界优化的方法 | |
| CN109492830B (zh) | 一种基于时空深度学习的移动污染源排放浓度预测方法 | |
| Du et al. | Return period and risk analysis of nonstationary low-flow series under climate change | |
| CN110782093B (zh) | 融合ssae深度特征学习和lstm的pm2.5小时浓度预测方法及系统 | |
| CN111260111B (zh) | 基于气象大数据的径流预报改进方法 | |
| CN106651036A (zh) | 空气质量预报系统 | |
| CN112465243A (zh) | 一种空气质量预报方法及系统 | |
| CN113108918B (zh) | 一种极轨气象卫星热红外遥感数据反演气温方法 | |
| Martínez-Comesaña et al. | Optimisation of thermal comfort and indoor air quality estimations applied to in-use buildings combining NSGA-III and XGBoost | |
| CN101893674A (zh) | 一种区域电网污闪指数预测方法 | |
| Saxena et al. | A review study of weather forecasting using artificial neural network approach | |
| CN114723188A (zh) | 水质预测方法、装置、计算机设备和存储介质 | |
| CN114997534A (zh) | 基于视觉特征的相似降雨预报方法和设备 | |
| CN113836808A (zh) | 一种基于重污染特征约束的pm2.5深度学习预测方法 | |
| CN113591215A (zh) | 基于不确定性的异常卫星组件布局检测方法 | |
| CN110580552B (zh) | 一种普适的区域环境信息移动感知及预测方法 | |
| CN113688506B (zh) | 基于微站等多维数据的潜在大气污染源识别方法 | |
| CN104239446A (zh) | 一种利用辅助变量数据进行采样点布局的方法 | |
| Freire et al. | Predicting building's corners hygrothermal behavior by using a Fuzzy inference system combined with clustering and Kalman filter | |
| CN115510767B (zh) | 基于深度时空网络的区域气温预测方法 | |
| CN116796291A (zh) | 一种基于lstm-mea-svr空气质量预报的系统 | |
| CN116525135A (zh) | 基于气象因素的时空模型对疫情发展态势进行预测的方法 | |
| CN117077353A (zh) | 基于多变量bp神经网络的碳排放测算建模、方法及装置 | |
| Ho et al. | Multivariate convolutional LSTMs for relative humidity forecasting |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| 121 | Ep: the epo has been informed by wipo that ep was designated in this application |
Ref document number: 19853407 Country of ref document: EP Kind code of ref document: A1 |
|
| WPC | Withdrawal of priority claims after completion of the technical preparations for international publication |
Ref document number: 201810965453.6 Country of ref document: CN Date of ref document: 20201110 Free format text: WITHDRAWN AFTER TECHNICAL PREPARATION FINISHED |
|
| NENP | Non-entry into the national phase |
Ref country code: DE |
|
| 122 | Ep: pct application non-entry in european phase |
Ref document number: 19853407 Country of ref document: EP Kind code of ref document: A1 |