CN114360739B - Dengue risk prediction method based on remote sensing cloud computing and deep learning - Google Patents
Dengue risk prediction method based on remote sensing cloud computing and deep learning Download PDFInfo
- Publication number
- CN114360739B CN114360739B CN202210010627.XA CN202210010627A CN114360739B CN 114360739 B CN114360739 B CN 114360739B CN 202210010627 A CN202210010627 A CN 202210010627A CN 114360739 B CN114360739 B CN 114360739B
- Authority
- CN
- China
- Prior art keywords
- dengue
- data
- week
- prediction
- cloud computing
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Classifications
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
- Y02A90/00—Technologies having an indirect contribution to adaptation to climate change
- Y02A90/10—Information and communication technologies [ICT] supporting adaptation to climate change, e.g. for weather forecasting or climate simulation
Landscapes
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a dengue risk prediction method based on remote sensing cloud computing and deep learning, which is characterized in that a buffer zone of a water impermeable layer is set as a main transmission range of dengue, geographic space big data borne by a Google earth engine and a cloud computing method are applied, multiple driving factors are computed in the dengue transmission range to form a time sequence with epidemiological weeks as a unit, and the time difference value of natural logarithm of the number of cases per week is used as a label to directly predict the change of the number of cases per epidemiological week through a model; according to the invention, the calculation range of the climate environment characteristics is precisely an irregular area block of human activities from a wide administrative boundary, the high efficiency caused by the large data cloud calculation of the geographic space is highlighted, and the time difference value of the natural logarithm of the number of week cases is used as a label, so that the evaluation of the model prediction capability is truly and effectively carried out, the negative influence of the autocorrelation on model training is removed to a certain extent, and the timely and accurate prediction of dengue fever propagation risk is realized.
Description
Technical Field
The invention relates to the technical field of infectious disease risk prediction, in particular to a dengue fever risk prediction method based on remote sensing cloud computing and deep learning.
Background
Dengue is an infectious disease caused by the transmission of dengue virus by aedes, which can lead to a recessive infection, dengue fever, dengue hemorrhagic fever after infection. With global urbanization and climate change, the number and geographic extent of dengue affected population is increasing, and at present, nearly one hundred countries worldwide and more than half of the world population are affected by dengue. Cities are a major area where dengue fever occurs, where various dengue agents inhabit and reproduce in cities, where urban population densities are high, and where population flows between cities are frequent, resulting in a great deal of concern for urban dengue fever. To date, controlling the number of agents and reducing human-to-agent contact are the primary means of dengue control, and in this context, accurate and timely dengue risk prediction is an important information base for the implementation of preventive measures.
Existing dengue case prediction techniques are typically based on autoregressive integral moving averages, traditional machine learning, deep learning, etc. methods process time series data to predict the number of dengue week cases or month cases on a specific spatial scale (e.g., district, city, province, or country). The existing method generally takes a target area as a whole (such as administrative units of cities, provinces and the like), combines weather site data or remote sensing satellite data to calculate input features required by a dengue risk prediction model, and does not consider how to reasonably describe the dengue propagation space range; and secondly, the calculation of the climate environment characteristics by the existing method is based on a large amount of weather site data and multi-source remote sensing image acquisition. The number of meteorological sites is limited and the spatial distribution is not balanced, so that the data cannot represent the whole situation of dengue transmission range. The downloading and processing of the multi-source remote sensing data are tedious and time-consuming, so that the prediction efficiency is greatly reduced; finally, existing methods consider training models using historical dengue case data as one of the input features. Since the change in the number of dengue cases is based on a continuous change in the number of cases at the previous time and is affected by various external climatic factors, there is a strong autocorrelation of the time series of cases, which is mostly a non-stationary series and exhibits large fluctuations. However, the existing method has not deeply considered the stability of the driving factor time series and the label time series, and the prediction capability of the case number with large fluctuation needs to be improved. The dengue risk prediction method based on remote sensing cloud computing and deep learning is used for solving the problems in the prior art.
Disclosure of Invention
Aiming at the problems, the invention aims to provide a dengue risk prediction method based on remote sensing cloud computing and deep learning, which is based on public geospatial big data, specifies a reasonable dengue propagation space range according to the flight distance of dengue media and the main range of crowd activities, efficiently forms time series data of each driving factor according to cloud computing capacity, and considers input characteristics and tag time series stability inspection, so that data computation of a model is more reasonable and efficient, and timely and accurate prediction of dengue propagation risk is realized.
In order to achieve the purpose of the invention, the invention is realized by the following technical scheme: dengue risk prediction method based on remote sensing cloud computing and deep learning comprises the following steps:
step one: firstly, acquiring a water-impermeable layer of a city with risk to be predicted, and then defining a buffer zone of one kilometer around the acquired water-impermeable layer as a main transmission area of dengue according to the flight range of dengue media;
step two: obtaining epidemiology Zhou Xulie over a dengue spread period and obtaining a start time and an end time for each week;
step three: the method comprises the steps of calculating geospatial big data of dengue fever driving factors by using a Google earth engine cloud computing platform, firstly obtaining average daytime surface temperature, average nighttime surface temperature, average air temperature, average relative humidity, rainfall total, vegetation normalization index and vegetation enhancement index in each epidemiology week, and forming a week synthetic image set of each driving factor;
step four: spatially aggregating the driving factor images of the weekly composite image set with a pixel average value within a main transmission range of dengue to form a time series of each driving factor;
step five: analyzing the correlation between the weather environment characteristic time series data and removing redundant characteristics;
step six: calculating the number of dengue fever cases in each epidemiology week as tag data, correcting the number of cases per week by adopting a natural logarithmic transformation and time difference method of adding one to the number of cases per week in sequence, and finally selecting a log difference sequence of the number of cases per week as a tag;
step seven: combining the driving factors of dengue environment climate, the log history case number and the label data to design two prediction scenes, wherein the two prediction scenes comprise the prediction scenes which only consider the driving factors of dengue environment climate and the prediction scenes which consider the environment climate and the history case at the same time;
step eight: dividing the time sequence data set into a training set and a verification set, constructing early prediction models under different time nodes for the two prediction scenes designed in the seventh step, and training the early prediction models of the change of the number of the week cases through the training set;
step nine: and calculating the precision indexes RMSE and MAE of the early prediction model by combining the dengue risk prediction value and the true value, comparing and analyzing the prediction curves of twelve early models under two prediction scenes, and selecting an optimal model as a deep learning model for directly predicting the number of dengue week cases.
The further improvement is that: in the first step, the dengue fever medium is aedes aegypti and aedes albopictus, and the flying ranges of the aedes aegypti and the aedes albopictus are 100-200 m and 100-1000 m respectively.
The further improvement is that: in the third step, the daytime and nighttime surface temperature characteristics are calculated by adopting remote sensing data MOD11A1 products, the spatial resolution is 1000 meters, and the time resolution is daily.
The further improvement is that: in the third step, the average air temperature and the average relative humidity are calculated by adopting a global land data assimilation system GLDAS-2.1, the spatial resolution is 0.25 degree, and the temporal resolution is daily.
The further improvement is that: in the third step, the total rainfall amount is calculated by adopting a TRMM3B42 rainfall measuring product, the spatial resolution is 0.25, and the time resolution is every 3 hours.
The further improvement is that: in the third step, the vegetation normalization index and the vegetation enhancement index are calculated by using MOD09GA surface reflectivity products, the spatial resolution is 500 meters, and the time resolution is daily.
The further improvement is that: in the step six, the specific steps of the autocorrelation elimination of the dengue weekly case time series are as follows: the number of cases per week is first added by a natural logarithmic transformation, then the number of the previous week is subtracted from the number of the next week, and the obtained difference is assigned to the previous week.
The further improvement is that: in the eighth step, after data enter a model, firstly, the data dimension is adjusted from three dimensions to two dimensions, the weight matrix of the data and the number of the specified nodes is a matrix product, and a 0.1 bias matrix is added to serve as a preliminary feature extraction, then the data dimension is adjusted to three dimensions again, the data is transmitted into an LSTM layer to operate, the time series feature of the data is extracted, if the model is in a training mode, the data is subjected to dropout processing, namely, each feature extraction point is endowed with a fixed probability to disable the feature extraction point, then the data dimension is adjusted to two dimensions comprising the number of the specified nodes, finally, the weight matrix of the data and the number of the specified nodes is a matrix product, and a bias matrix of 0.1 is added to obtain an output result of the model.
The beneficial effects of the invention are as follows: according to the method, the main human activity area and the dengue medium mosquito flight distance are considered, the impermeable layer kilometer buffer area is set to serve as a main dengue transmission range, the geographic space big data borne by the Google earth engine and the cloud computing method are applied, multiple driving factors are computed in the dengue transmission range, and a time sequence taking the epidemiological week as a unit is formed.
Drawings
In order to more clearly illustrate the embodiments of the invention or the technical solutions of the prior art, the drawings which are used in the description of the embodiments or the prior art will be briefly described, it being obvious that the drawings in the description below are only some embodiments of the invention, and that other drawings can be obtained according to these drawings without inventive faculty for a person skilled in the art.
FIG. 1 is a schematic flow chart of a risk prediction method in an embodiment of the present invention;
FIG. 2 is a schematic diagram of a climate environment variable time series and a historical case pair time series in an embodiment of the invention;
fig. 3 is a schematic diagram of early prediction model results for two prediction scenarios in an embodiment of the present invention.
Detailed Description
The following description of the embodiments of the present invention will be made clearly and completely with reference to the accompanying drawings, in which it is apparent that the embodiments described are only some embodiments of the present invention, but not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
Referring to fig. 1, 2 and 3, the embodiment provides a dengue risk prediction method based on remote sensing cloud computing and deep learning, which includes the following steps:
step one: considering that the transmission of dengue mainly occurs in the areas where the human beings are dominant, and that the flight ranges of dengue medium aedes aegypti and aedes albopictus are 100-200 meters and 100-1000 meters respectively, firstly acquiring a water impermeable layer of a city to be predicted at risk, and defining a kilometer buffer zone around the acquired water impermeable layer as the main transmission area of dengue;
step two: obtaining epidemiology Zhou Xulie over a dengue spread period and obtaining a start time and an end time for each week;
step three: acquiring data (shown in the following table one) such as average daytime surface temperature, average nighttime surface temperature, average air temperature, average relative humidity, rainfall total amount, vegetation normalization index, vegetation enhancement index and the like in each epidemiology week on a Google Earth engine cloud computing platform (GEE) to form a week synthetic image set of each driving factor, wherein the surface temperature characteristic is calculated by using remote sensing data MOD11A1 products, the spatial resolution is 1000 meters, the time resolution is daily, the average air temperature and the average relative humidity are calculated by using a global land data assimilation system GLDAS-2.1, the spatial resolution is 0.25 degree, the time resolution is daily, the rainfall total amount calculation is calculated by using TRMM3B42 rainfall measuring products, the spatial resolution is 0.25, the time resolution is every 3 hours, the vegetation normalization index and the vegetation enhancement index are calculated by using MOD09GA surface reflectivity products, the spatial resolution is 500 meters, and the time resolution is daily;
step four: analyzing the correlation between the weather environment characteristic time series data and removing redundant characteristics;
step five: calculating the number of dengue fever cases in each epidemiology week as tag data, correcting the number of cases per week by adopting a natural logarithmic transformation and time difference method of adding one to the number of cases per week in sequence, and finally selecting a log difference sequence of the number of cases per week as a tag;
step six: combining dengue environmental climate driving factors, the number of logarithmic history cases (the base of the change in predicted cases) and the tag data to design two prediction scenarios, including a prediction scenario taking only dengue environmental climate driving factors into account and a prediction scenario taking both environmental climate and history cases into account;
step seven: dividing a time sequence data set into a training set and a verification set, constructing early prediction models (namely, a prediction label is a week case number logarithmic difference value from one week to twelve weeks after the delay) under different time nodes for the two prediction scenes, and training the early prediction models of the week case number change through the training set;
after data enter the model, three-dimensionally adjust the data dimension from (batch size time step) characteristic number to ((batch size time step) characteristic number) two-dimension at first, make matrix multiplication with the weight matrix of the number of appointed nodes of characteristic number initialized by random normal distribution of data, and add 0.1 bias matrix, as preliminary characteristic extraction, then readjust the data dimension to ((batch size time step) appointed nodes) three-dimension, and transmit the data to LSTM (long short term memory network) layer for operation, extract the time series characteristic of the data, if the model is in training mode, make the data do dropout processing, namely assign a fixed probability to each characteristic extraction point, make it invalid, the method can lighten the overfitting phenomenon of the model, make the result of the model have better generalization, then readjust the data dimension to include the number of appointed nodes, finally, make bias matrix multiplication with the weight matrix of the number of appointed nodes 1 of data and 0.1, and output the result of the model;
step eight: and calculating the precision indexes RMSE and MAE of the early prediction model by combining the dengue risk prediction value and the true value, comparing and analyzing the prediction curves of twelve early models under two prediction scenes, and selecting an optimal model as a deep learning model for directly predicting the number of dengue week cases.
List one
Calculating dengue climate environmental factors according to epidemiological weeks and dengue propagation ranges can also be realized based on other remote sensing or geographic big data cloud computing platforms, such as PIE-Engine remote sensing and geographic information cloud service platforms and geospatial data cloud Geospatial data cloud;
other autocorrelation elimination methods may replace the case number logarithmic difference method proposed in the present embodiment, such as a method of a ratio of two adjacent time case numbers, a second order difference of three adjacent time case numbers, and the like.
The foregoing description of the preferred embodiments of the invention is not intended to be limiting, but rather is intended to cover all modifications, equivalents, alternatives, and improvements that fall within the spirit and scope of the invention.
Claims (7)
1. The dengue risk prediction method based on remote sensing cloud computing and deep learning is characterized by comprising the following steps of:
step one: firstly, acquiring a water-impermeable layer of a city with risk to be predicted, and then defining a buffer zone of one kilometer around the acquired water-impermeable layer as a main transmission area of dengue according to the flight range of dengue media;
step two: obtaining epidemiology Zhou Xulie over a dengue spread period and obtaining a start time and an end time for each week;
step three: the method comprises the steps of calculating geospatial big data of dengue fever driving factors by using a Google earth engine cloud computing platform, firstly obtaining average daytime surface temperature, average nighttime surface temperature, average air temperature, average relative humidity, rainfall total, vegetation normalization index and vegetation enhancement index in each epidemiology week, and forming a week synthetic image set of each driving factor;
step four: spatially aggregating the driving factor images of the weekly composite image set with a pixel average value within a main transmission range of dengue to form a time series of each driving factor;
step five: analyzing the correlation between the weather environment characteristic time series data and removing redundant characteristics;
step six: calculating the number of dengue fever cases in each epidemiology week as tag data, correcting the number of cases per week by adopting a natural logarithmic transformation and time difference method of adding one to the number of cases per week in sequence, and finally selecting a log difference sequence of the number of cases per week as a tag;
the specific steps of the autocorrelation elimination of the dengue week case time series are as follows: firstly, adding one to the number of cases per week to perform natural logarithmic transformation, then subtracting the value of the previous week from the value of the next week, and then assigning the obtained difference value to the previous week;
step seven: combining the driving factors of dengue environment climate, the log history case number and the label data to design two prediction scenes, wherein the two prediction scenes comprise the prediction scenes which only consider the driving factors of dengue environment climate and the prediction scenes which consider the environment climate and the history case at the same time;
step eight: dividing the time sequence data set into a training set and a verification set, constructing early prediction models under different time nodes for the two prediction scenes designed in the seventh step, and training the early prediction models of the change of the number of the week cases through the training set;
step nine: and calculating the precision indexes RMSE and MAE of the early prediction model by combining the dengue risk prediction value and the true value, comparing and analyzing the prediction curves of twelve early models under two prediction scenes, and selecting an optimal model as a deep learning model for directly predicting the number of dengue week cases.
2. The dengue risk prediction method based on remote sensing cloud computing and deep learning as claimed in claim 1, wherein: in the first step, the dengue fever medium is aedes aegypti and aedes albopictus, and the flying ranges of the aedes aegypti and the aedes albopictus are 100-200 m and 100-1000 m respectively.
3. The dengue risk prediction method based on remote sensing cloud computing and deep learning as claimed in claim 1, wherein: in the third step, the daytime and nighttime surface temperature characteristics are calculated by adopting remote sensing data MOD11A1 products, the spatial resolution is 1000 meters, and the time resolution is daily.
4. The dengue risk prediction method based on remote sensing cloud computing and deep learning as claimed in claim 1, wherein: in the third step, the average air temperature and the average relative humidity are calculated by adopting a global land data assimilation system GLDAS-2.1, the spatial resolution is 0.25 degree, and the temporal resolution is daily.
5. The dengue risk prediction method based on remote sensing cloud computing and deep learning as claimed in claim 1, wherein: in the third step, the total rainfall amount is calculated by adopting a TRMM3B42 rainfall measuring product, the spatial resolution is 0.25, and the time resolution is every 3 hours.
6. The dengue risk prediction method based on remote sensing cloud computing and deep learning as claimed in claim 1, wherein: in the third step, the vegetation normalization index and the vegetation enhancement index are calculated by using MOD09GA surface reflectivity products, the spatial resolution is 500 meters, and the time resolution is daily.
7. The dengue risk prediction method based on remote sensing cloud computing and deep learning as claimed in claim 1, wherein: in the eighth step, after data enter a model, firstly, the data dimension is adjusted from three dimensions to two dimensions, the weight matrix of the data and the number of the specified nodes is a matrix product, and a 0.1 bias matrix is added to serve as a preliminary feature extraction, then the data dimension is adjusted to three dimensions again, the data is transmitted into an LSTM layer to operate, the time series feature of the data is extracted, if the model is in a training mode, the data is subjected to dropout processing, namely, each feature extraction point is endowed with a fixed probability to disable the feature extraction point, then the data dimension is adjusted to two dimensions comprising the number of the specified nodes, finally, the weight matrix of the data and the number of the specified nodes is a matrix product, and a bias matrix of 0.1 is added to obtain an output result of the model.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202210010627.XA CN114360739B (en) | 2022-01-05 | 2022-01-05 | Dengue risk prediction method based on remote sensing cloud computing and deep learning |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202210010627.XA CN114360739B (en) | 2022-01-05 | 2022-01-05 | Dengue risk prediction method based on remote sensing cloud computing and deep learning |
Publications (2)
Publication Number | Publication Date |
---|---|
CN114360739A CN114360739A (en) | 2022-04-15 |
CN114360739B true CN114360739B (en) | 2023-07-21 |
Family
ID=81106524
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202210010627.XA Active CN114360739B (en) | 2022-01-05 | 2022-01-05 | Dengue risk prediction method based on remote sensing cloud computing and deep learning |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN114360739B (en) |
Family Cites Families (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
GB2515215A (en) * | 2012-02-13 | 2014-12-17 | Kamran Khan | Warning system for infectious diseases and method therefor |
CN108766585A (en) * | 2018-05-31 | 2018-11-06 | 平安科技(深圳)有限公司 | Generation method, device and the computer readable storage medium of influenza prediction model |
CN110459329B (en) * | 2019-07-11 | 2022-11-18 | 广东省公共卫生研究院 | Dengue fever risk comprehensive assessment method |
CN111554408B (en) * | 2020-04-27 | 2024-04-19 | 中国科学院深圳先进技术研究院 | City internal dengue space-time prediction method, system and electronic equipment |
-
2022
- 2022-01-05 CN CN202210010627.XA patent/CN114360739B/en active Active
Non-Patent Citations (1)
Title |
---|
基于GA-BP神经网络模型的登革热时空扩散模拟;李卫红;陈业滨;闻磊;;中国图象图形学报(第07期) * |
Also Published As
Publication number | Publication date |
---|---|
CN114360739A (en) | 2022-04-15 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Han et al. | Simulating the spatial dynamics of urban growth with an integrated modeling approach: A case study of Foshan, China | |
Wang et al. | Spatio-temporal urban growth dynamics of Lagos Metropolitan Region of Nigeria based on Hybrid methods for LULC modeling and prediction | |
Bagan et al. | Analysis of urban growth and estimating population density using satellite images of nighttime lights and land-use and population data | |
Liao et al. | Urban sprawl scenario simulations based on cellular automata and ordered weighted averaging ecological constraints | |
Luo et al. | Understanding ecological groups under landscape fragmentation based on network theory | |
Dupuy et al. | ARPEGE cloud cover forecast postprocessing with convolutional neural network | |
Yadav et al. | Assessment and prediction of urban growth for a mega-city using CA-Markov model | |
CN116343103A (en) | Natural resource supervision method based on three-dimensional GIS scene and video fusion | |
CN114120145A (en) | Monitoring method, monitoring device, electronic equipment and computer readable storage medium | |
JP7023434B1 (en) | Eddy forecasting system and eddy forecasting method | |
CN114360739B (en) | Dengue risk prediction method based on remote sensing cloud computing and deep learning | |
CN116863097A (en) | Space environment data model construction method and device | |
Rocha et al. | Modelling coastal and land use evolution patterns through neural network and cellular automata integration | |
Xi et al. | A Bayesian model for estimating the effects of human disturbance on wildlife habitats based on nighttime light data and INLA-SPDE | |
CN115983522B (en) | Rural habitat quality assessment and prediction method | |
CN114842163A (en) | Multi-scale dynamic gridding model construction method for disaster assessment | |
Aqbelaghi et al. | Environmental approach in modelling of urban growth: Tehran City, Iran | |
Xu et al. | Monthly NDVI prediction using spatial autocorrelation and nonlocal attention networks | |
CN114139984A (en) | Urban traffic accident risk prediction method based on flow and accident collaborative perception | |
Suo et al. | Coastline carrying capacity monitoring and assessment based on GF-1 satellite remote sensing images | |
Benchelha et al. | Modeling Dynamic Urban Growth Using Cellular Automata and Geospatial Technique: Case of Casablanca in Morocco | |
Sangpradid | APPLICATION OF A MULTI-LAYER PERCEPTRON NEURAL NETWORK TO SIMULATE SPATIAL-TEMPORAL LAND USE AND LAND COVER CHANGE ANALYSIS BASED ON CELLULAR AUTOMATA IN BURIRAM PROVINCE, THAILAND. | |
Handayani et al. | Modelling of land use change in indramayu district, west Java Province | |
Getachew et al. | Modeling Land-Use/Land-Cover Change, Using Multi-Layer Perceptron and Markov Chain Analysis: A Study on Bahir Dar City, Ethiopia | |
Zhao et al. | A modified cyanobacteria prediction model based on cellular automata model using N and P concentration reverse data: a case study in Taihu Lake |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |