JP2020527788A - Disease prediction methods and devices, computer devices and readable storage media - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a method for predicting a disease. The disease prediction method of the present invention includes a step of acquiring disease monitoring data, weather data, and public opinion data, a step of preprocessing the disease monitoring data, weather data, and public opinion data. A step of constructing a multi-layer LSTM model, a step of training and performing performance verification on the multi-layer LSTM model to obtain an optimized multi-layer LSTM model, and a step of predicting at the time of prediction by the pre-processed multi-layer LSTM model. The step of obtaining the disease prediction result at the time of the prediction is included. The present invention also provides a disease prediction device, a computer device, and a readable storage medium. According to the present invention, highly accurate disease prediction can be realized. [Selection diagram] Fig. 1

Description

The present invention relates to the field of prediction technology, and specifically to disease prediction methods and devices, computer devices and non-volatile readable storage media. This application is filed with the Chinese Patent Office on April 11, 2018, with Chinese application number 201810321868. X, Title of the Invention: Claims priority based on "disease prediction methods and devices, computer devices and readable storage media", the entire contents of which application is incorporated herein by reference.

As the global economic integration process accelerates, economic and communication activities increase and the flow of people becomes more frequent, providing a favorable environment for disease epidemics and outbreaks, and public health problems. Is getting more and more serious. At the same time, changes will occur in society and the natural environment, and an increase in situations that affect people's health such as environmental pollution and natural disasters will increase the possibility of public health emergencies.

How to detect a public health emergency of a disease early, issue an immediate warning, and take appropriate control measures as soon as possible to minimize the loss of a public health emergency has long been a focus of attention. It is also an important content of measures against emergencies. Early warning of a public health emergency is the operation of modern advanced technologies such as computers, networks, and communications by collecting, organizing, analyzing, and integrating relevant data to monitor, identify, and identify signs of the situation. Diagnose and evaluate, and issue immediate warnings, notify relevant departments and the public to take relevant response and preparatory work, and have timely and effective preventive and administrative measures taken to ensure as much emergency as possible. It stops or mitigates the outbreak, or mitigates the harm caused by the situation.

An important task in early warning of public health emergencies is disease prediction, i.e., predicting future disease monitoring data based on past disease monitoring data (ie, patient data). With the development of machine learning technology, more and more machine learning methods have been applied to disease prediction. However, conventional machine learning applied to disease prediction often requires manually defining a feature set and searching for the optimal feature combination in the defined feature set, which is less effective and therefore disease. The accuracy of the prediction is affected.

In view of the above problems, it is necessary to propose a disease prediction method and device, a computer device, and a non-volatile readable storage medium capable of realizing highly accurate disease prediction.

In order to solve the above problems, the first aspect of the present invention provides a disease prediction method, wherein the disease prediction method is described.
The disease monitoring data is acquired, the disease monitoring data is a step that is time-series data, and the weather data associated with the disease monitoring data is acquired, and the weather data is time-series data corresponding to the disease monitoring data. A step and public opinion data related to the disease monitoring data are acquired, and the public opinion data is a step which is time-series data corresponding to the disease monitoring data, the disease monitoring data, the weather data, and the public opinion data. A step of preprocessing the data, a step of constructing a multi-layer long short-term memory recurrent neural network model, that is, a multi-layer RSTM model, and training data and verification data from the preprocessed disease monitoring data, weather data, and public opinion data. The step of obtaining an optimized multi-layer LSTM model by training and performing performance verification on the multi-layer LSTM model using the training data and the verification data, and the preprocessed disease monitoring data. Pre-predicted disease monitoring data, weather data and public opinion data are acquired from the weather data and the public opinion data, and the pre-predicted disease monitoring data, weather data and public opinion data are input to the optimized multi-layer LSTM model. However, the step of obtaining the disease monitoring result at the time of the prediction is included.

A second aspect of the present invention provides a disease prediction device, wherein the disease prediction device
Disease monitoring data is acquired, and the disease monitoring data is the first acquisition unit, which is time-series data, and
The weather data related to the disease monitoring data is acquired, and the weather data includes a second acquisition unit which is time-series data corresponding to the disease monitoring data.
The public opinion data related to the disease monitoring data is acquired, and the public opinion data includes a third acquisition unit which is time-series data corresponding to the disease monitoring data.
A preprocessing unit for preprocessing the disease monitoring data, the weather data, and the public opinion data,
Multi-layer long short-term memory recurrent neural network model, that is, a construction unit for constructing a multi-layer RSTM model,
Training data and verification data are acquired from the preprocessed disease monitoring data, weather data, and public opinion data, and training and performance verification are performed on the multi-layer LSTM model using the training data and the verification data. , An optimization unit that provides an optimized multi-layer RSTM model,
The preprocessed disease monitoring data, the weather data, and the public opinion data are used to acquire the disease monitoring data, the weather data, and the public opinion data before the prediction time point, and the disease monitoring data, the weather data, and the public opinion data before the prediction time point are obtained. Includes a prediction unit that inputs into an optimized multi-layer RSTM model to obtain disease monitoring results at the time of prediction.

A third aspect of the present invention provides a computer device, wherein the computer device comprises a memory and a processor, the memory storing at least one computer-readable command, and the processor is at least one computer. Execute a readable command,
The disease monitoring data is acquired, the disease monitoring data is a step that is time-series data, and the weather data associated with the disease monitoring data is acquired, and the weather data is time-series data corresponding to the disease monitoring data. A step and public opinion data related to the disease monitoring data are acquired, and the public opinion data is a step which is time-series data corresponding to the disease monitoring data, the disease monitoring data, the weather data, and the public opinion data. A step of preprocessing the data, a step of constructing a multi-layer long short-term memory recurrent neural network model, that is, a multi-layer RSTM model, and training data and verification data from the preprocessed disease monitoring data, weather data, and public opinion data. The step of obtaining an optimized multi-layer LSTM model by training and performing performance verification on the multi-layer LSTM model using the training data and the verification data, and the preprocessed disease monitoring data. , The disease monitoring data before the prediction time point, the weather data and the public opinion data are acquired from the weather data and the public opinion data, and the disease monitoring data, the weather data and the public opinion data before the prediction time point are converted into the optimized multi-layer LSTM model. The step of inputting and obtaining the disease monitoring result at the time of the prediction is realized.

A fourth aspect of the present invention provides a non-volatile readable storage medium, wherein the non-volatile readable storage medium stores at least one computer-readable command, and the at least one computer-readable command is by a processor.
The disease monitoring data is acquired, the disease monitoring data is a step that is time-series data, and the weather data associated with the disease monitoring data is acquired, and the weather data is time-series data corresponding to the disease monitoring data. A step and public opinion data related to the disease monitoring data are acquired, and the public opinion data is a step which is time-series data corresponding to the disease monitoring data, the disease monitoring data, the weather data, and the public opinion data. A step of preprocessing the data, a step of constructing a multi-layer long short-term memory recurrent neural network model, that is, a multi-layer RSTM model, and training data and verification data from the preprocessed disease monitoring data, weather data, and public opinion data. The step of obtaining an optimized multi-layer LSTM model by training and performing performance verification on the multi-layer LSTM model using the training data and the verification data, and the preprocessed disease monitoring data. , The disease monitoring data before the prediction time point, the weather data and the public opinion data are acquired from the weather data and the public opinion data, and the disease monitoring data, the weather data and the public opinion data before the prediction time point are converted into the optimized multi-layer LSTM model. The step of inputting and obtaining the disease monitoring result at the time of the prediction is realized.

The present invention acquires disease monitoring data, the disease monitoring data acquires steps that are time-series data, and weather data related to the disease monitoring data, and the weather data corresponds to the disease monitoring data. A step that is time-series data and public opinion data related to the disease monitoring data are acquired, and the public opinion data is a step that is time-series data corresponding to the disease monitoring data, the disease monitoring data, and the weather data. And the step of preprocessing the public opinion data, the step of constructing a multi-layer long short-term memory recurrent neural network model, that is, a multi-layer RSTM model, and training from the preprocessed disease monitoring data, weather data, and public opinion data. The steps of acquiring data and verification data, training and performing performance verification on the multi-layer LSTM model using the training data and the verification data, and obtaining an optimized multi-layer LSTM model, and preprocessing were performed. The disease monitoring data, weather data and public opinion data before the prediction time point were acquired from the disease monitoring data, the weather data and the public opinion data, and the disease monitoring data, the weather data and the public opinion data before the prediction time point were optimized. It includes a step of inputting into a multi-layer LSTM model and obtaining a disease monitoring result at the time of the prediction.

The present invention predicts morbidity based on a multi-layer RSTM model. The LSTM model extracts knowledge directly from the data, builds feature vectors that are useful for prediction, and improves prediction accuracy. Furthermore, compared to the conventional RNN (Recurrent Neural Networks) model, the LSTM model solves the problem that the gradient is eliminated during the long-term dependence caused when the amount of time series data is too large. The present invention uses weather data and public opinion data for disease prediction as influential factors to improve the accuracy of disease prediction. Therefore, the present invention realizes highly accurate disease prediction.

FIG. 1 is a flowchart of a disease prediction method according to Example 1 of the present application.

FIG. 2 is a detailed flowchart for acquiring weather data related to disease monitoring data from the disease prediction method according to the second embodiment of the present application.

FIG. 3 is a block diagram of the disease prediction device according to the third embodiment of the present application.

FIG. 4 is a detailed configuration diagram of the second acquisition unit of the disease prediction device according to the fourth embodiment of the present application.

FIG. 5 is a schematic view of the computer device according to the fifth embodiment of the present application.

Preferably, the disease prediction method of the present invention applies to one or more computer devices. The computer device is a device capable of automatically performing numerical calculation and / or information processing according to preset or stored commands, and its hardware is a microprocessor and an application specific integrated circuit (Application Special Integrated Circuit). , ASIC), Programmable Gate Array (Field-Programmable Gate Array, FPGA), Digital Processor (DSP), Embedded Devices, and the like, but not limited to these.

Example 1
FIG. 1 is a flowchart of a disease prediction method according to Example 1 of the present invention. The disease prediction method is applied to a computer device. The disease prediction method uses a multi-layer long short-term memory recurrent neural network model to predict disease monitoring data and obtain highly accurate disease prediction results.

As shown in FIG. 1, the disease prediction method specifically includes the following steps.

In step 101, the disease monitoring data, which is time series data, is acquired.

The disease monitoring data may include disease data such as influenza, hand-foot-and-mouth disease, measles, and mumps.

In a preset area (for example, province, city, region), a disease monitoring network consisting of a plurality of monitoring bases is established, disease monitoring data is acquired from the monitoring bases, and the disease monitoring data is used for disease monitoring. Time series data can be constructed. Medical institutions, schools, childcare facilities, pharmacies, etc. can be selected as monitoring bases for disease monitoring and data collection for the corresponding targets. You can select a location that meets the preset conditions as a monitoring base. The preset conditions may include the number of people, scale, and the like. For example, select schools and childcare facilities where the number of students reaches a preset number as monitoring bases. Also, select a pharmacy whose scale (for example, statistics based on daily sales) reaches a preset scale as a monitoring base. In addition, a hospital whose scale (for example, statistics based on the number of examinees per day) reaches a preset scale is selected as a monitoring base.

Disease monitoring data at different time points constitutes time series data for disease monitoring. For example, disease monitoring data collected on a daily basis is used as time-series data for disease monitoring. Alternatively, the disease monitoring data collected on a weekly basis is used as time-series data for disease monitoring.

Medical institutions (mainly hospitals) are the best place to detect early signs of disease and are a top priority for disease monitoring. Acquire disease monitoring data based on the patient's consultation status.

Since some patients with some diseases go to pharmacies to buy medicines to relieve early symptoms, disease monitoring data is acquired based on the sales status of medicines at pharmacies.

Since children and adolescents are subjects at high risk of developing the disease and are an important part of disease transmission, surveillance of this target should also be strengthened. Schools and childcare facilities are good places to monitor disease outbreaks in children and adolescents. Obtain disease monitoring data based on the vacation status of children and adolescents in schools and childcare facilities.

Therefore, in the present invention, medical institutions, schools, childcare facilities and pharmacies are mainly selected for collecting disease monitoring data. Of course, for the above data source selection, other notable targets or locations in other embodiments can be added or replaced as data sources for monitoring. For example, the disease monitoring data of the resident of the hotel can be acquired by including the hotel in the disease monitoring area.

If necessary, only the disease monitoring data collected by any type of monitoring base (for example, a medical institution) can be acquired to configure the disease monitoring time series data. For example, disease monitoring data collected in a hospital can be acquired to construct time-series data for disease monitoring. Alternatively, disease monitoring data collected from a plurality of types of monitoring bases can be combined to form time-series data for disease monitoring. For example, time-series data for disease monitoring can be constructed mainly with disease monitoring data collected at a hospital and supplemented with disease monitoring data collected from a pharmacy.

Disease monitoring data may include disease data such as the number of disease visits, consultation rates, morbidity numbers, and morbidity rates. For example, the number of daily consultations for a disease (for example, influenza) may be obtained from a medical institution (for example, a hospital), and the number of daily consultations for a disease (for example, influenza) may be used as disease monitoring data. Alternatively, the daily prevalence of a student's disease (eg influenza) is obtained from the school and the daily prevalence of the disease (eg influenza) is used as disease monitoring data.

In step 102, the weather data associated with the disease monitoring data is acquired, and the weather data is time series data corresponding to the disease monitoring data.

The weather data associated with the disease monitoring data is the weather data that affects the disease monitoring data (that is, the disease morbidity data). The effect of different weather data on the disease monitoring data can be analyzed in advance, and the weather that affects the disease monitoring data or has a more serious effect can be determined based on the analysis result.

The weather data may include humidity, temperature, atmospheric pressure, precipitation, vapor pressure, wind speed, wind direction, and sunshine duration. In a specific embodiment, the weather data may include daily mean temperature, mean barometric pressure, maximum temperature, minimum temperature, average relative humidity, minimum relative humidity, precipitation, average wind speed, sunshine duration, average vapor pressure.

The weather data is the same as the time zone corresponding to the disease monitoring data, and the weather data is the same as the statistical cycle (for example, daily, weekly) of the disease monitoring data. For example, when the disease monitoring data is the number of daily consultations in January-February 2018, the weather data is the daily weather data in January-February 2018. When the disease monitoring data is the number of weekly consultations from January to December 2017, the weather data is the weekly weather data from January to December 2017 (for example, weekly average temperature).

By capturing the weather from weather information websites (eg, China Weather Network, Sina Weather, Sohu Weather, etc.), the reliability of the weather data can be improved. The weather data can be captured from any web page.

It is possible to capture the weather data of a predetermined area. The predetermined area may include provinces, cities, regions and the like. For example, capture weather data for Shenzhen city.

It is possible to capture weather data for a predetermined time. The predetermined time may include a year, a month, a day, and the like. For example, capture daily weather data for January-February 2018.

The weather data can be captured by an internet worm. Internet worms are applications that can automatically extract the content of data information on web pages. Internet worms typically obtain the URL of an initial web page from the URL of one or more initial web pages (also called the seed URL) and follow a particular algorithm or strategy (eg, a depth-priority search strategy). In the process of capturing a web page, new URLs are constantly extracted from the current web page and placed in the corresponding queue until the stop condition is met. URL is an abbreviation for Uniform Resource Locator, that is, a uniform resource locator.

The weather data can be captured by an API interface opened on a weather information website (for example, an API interface opened on the China Weather Network). API is an abbreviation for application interface, and mutual communication between computer software can be realized via the API interface. Data in JSON format or XML format can be returned via the API interface opened on the weather information website.

In a specific embodiment, the weather data can be captured by an internet worm via an API interface opened on a weather information website. See Figure 2 for the specific process of capturing the weather data by an internet worm via the API interface opened on the weather information website.

In step 103, public opinion data that is related to the disease monitoring data and is time-series data corresponding to the disease monitoring data is acquired.

The public opinion data associated with the disease monitoring data is public opinion data indicating the disease monitoring data. As an example, during an epidemic of a disease (eg, influenza), as the number of people affected increases, people searching the Internet for disease-related words (eg, certain words such as influenza, Tamiflu, and high fever). As the number increases, the amount of searches for these words increases significantly. In addition, during the epidemic of a disease (eg, influenza), disease-related content (eg, disease information, treatment information) published on public opinion websites such as news, forums, blogs, and bulletin boards as the number of affected people increases. Etc.) are increasing. Therefore, disease can be predicted based on disease monitoring data and associated public opinion data.

The opinion data may include the number of searches for a particular word. For example, the number of searches for a specific word by a predetermined search engine (for example, the number of searches per day for a specific word by a predetermined search engine in a specific area) can be statistic.

The opinion data may also include the number of opinion information about a particular word contained on a particular opinion website (eg, news, forums, blogs, bulletin boards, etc.).

The specific word is a word related to a predicted disease, for example, if the specific word is a word related to a symptom of a disease, and the predicted disease is influenza, the specific word. May include sudden onset, high fever, chills, headache, weakness, throat irritation, myalgia, dry cough, etc. Also, if the predicted disease is hand-foot-and-mouth disease, the particular word may include mouth pain, anorexia nervosa, low-grade fever, small cold sores on the hands, small ulcers on the mouth, and the like.

The opinion data is the same as the time zone corresponding to the disease monitoring data, and the opinion data is the same as the statistical cycle of the disease monitoring (for example, daily, weekly). For example, if the disease monitoring data is the number of daily consultations from January to February 2018, the opinion data is the daily opinion data from January to February 2018 (for example, a specific word is searched per day). Number of times). Further, if the disease monitoring data is the number of weekly consultations from January to December 2017, the opinion data is the weekly opinion data from January to December 2017 (for example, the number of times a specific word is searched per week). ).

Steps 101-103 can be executed in any order and can be executed in parallel.

In step 104, preprocessing is performed on the disease monitoring data, weather data, and public opinion data.

Preprocessing for disease monitoring data, weather data and public opinion data may include anomalous data processing. The reason why the abnormality data processing is performed on the disease monitoring data, the weather data, and the public opinion data is to correct the abnormality data in the disease monitoring data, the weather data, and the public opinion data, and to improve the reliability and accuracy of the disease prediction. ..

The anomaly data processing may include complementing missing values in the disease monitoring data, weather data and public opinion data. The missing value can be complemented by the average or median of the data before and after the missing value, or the missing value can be complemented by a method called regression fitting.

Furthermore, the anomaly data processing can also include correcting outliers in the disease monitoring data, weather data and public opinion data. The outlier is a numerical value that greatly deviates from other data. The outliers can be corrected by an interpolation method.

Preprocessing for disease monitoring data, weather data and public opinion data may also include conversion of the data format to said disease monitoring data, weather data and public opinion data. For example, by performing standardization processing on disease monitoring data, weather data, and public opinion data, the disease monitoring data, weather data, and public opinion data are provided with a standard format of consistency and used as input data of an LSTM model. Suitable.

In step 105, a multi-layer long short-term memory recurrent neural network (Long Short-term Memory Recurrent Neural Network) model, that is, a multi-layer LSTM model is constructed. The multi-layer LSTM model includes two LSTM cell layers and one fully connected layer, and the first LSTM cell layer is an input composed of input data (for example, the disease monitoring data, weather data and public opinion data). (Data) used to build features and acquire first hidden layer units, the second layer LSTM cell layer combines the first hidden layer units and the second hidden layer unit. The fully connected layer is used to obtain a prediction result (eg, disease prediction result) based on the second hidden layer unit, and the LSTM cell layer is used to obtain an oblivion gate and an input, respectively. The storage state of the LSTM cell layer is controlled by the oblivion gate, the input gate, and the output gate, including the gate and the output gate.

The LSTM model is a time recurrent neural network model. In contrast to traditional Recurrent Neural Network (RNN) models, LSTM models store information by building multiple gates in the LSTM cell layer, which means that the gradient is immediate during the model training process. Does not resolve.

The multi-layer LSTM model used in the method includes two LSTM cell layers and one fully connected layer, the first LSTM cell layer being input data (eg, said disease monitoring data, weather data and public opinion data). The LSTM cell layer of the second layer is a combination of the first hidden layer units, and is used to construct a feature for the input data composed of the first hidden layer unit. (Ii) Used to obtain a hidden layer unit. The fully connected layer obtains a predicted value based on the second hidden layer unit. The first non-display layer unit is a local feature, while the second non-display layer unit is an overall feature. That is, the LSTM cell layer of the first layer is used to extract local data, while the LSTM cell layer of the second layer is used to obtain the overall characteristics according to the local data. The fully connected layer is used to obtain a prediction result (for example, a disease prediction result) based on the overall characteristics.

The LSTM cell layer includes a forgetting gate, an input gate, and an output gate, and the storage state of the LSTM cell layer is controlled by the forgetting gate, the input gate, and the output gate. The input gate determines whether to receive the input at the current time, and the output gate determines whether to output the storage state.

In one embodiment, it is possible to calculate the forgetting gate _{f t} of LSTM cell layer, the input gate _{i t,} the output gate _{o t,} the memory state _{c t} and hidden layer units _{h t} as follows.

Among _{them, W f, U f, b} f is a parameter of the forgetting _{gate, W i, U i, b} i is the parameter of the input _{gate, W o, U o, b} o is an parameter of an output gate , W _c , U _c , _bc are parameters of the storage unit.

In another embodiment, the forgetting gate _{f t} of LSTM cell layer, the input gate _{i t,} the output gate _{o t,} the storage form

In step 106, training data and verification data are acquired from the preprocessed disease monitoring data, weather data, and public opinion data, and the training data and verification data are used to perform training and performance verification on the multi-layer LSTM model, which is optimal. Obtain a multi-layered RSTM model.

Time-series data may be cut out from the preprocessed disease monitoring data, weather data, and public opinion data to form the training data and the verification data.

The input data of the multi-layer RSTM model is a vector having a predetermined dimension (for example, 1000 dimensions). In the cut time series data, a vector of a predetermined dimension is constructed from the preprocessed disease monitoring data, weather data, and public opinion data corresponding to each time point, and the vector corresponding to each time point is sequentially put into the multilayer RSTM model in chronological order. It can be input and used for training and verification of the multi-layer RSTM model.

For example, from the preprocessed disease monitoring data, weather data, and public opinion data, first temporary series data for training the multi-layer RSTM model is cut out. In the cut out first temporary series data, the first vector of a predetermined dimension is constructed from the preprocessed disease monitoring data, weather data, and public opinion data corresponding to each time point, and the first vector corresponding to each time point is arranged in chronological order. It is used to sequentially input data into the multi-layer LSTM model and train the multi-layer LSTM model. From the preprocessed disease monitoring data, weather data and public opinion data, second time series data for verification against the multi-layer LSTM model is cut out. In the cut second time series data, a second vector of a predetermined dimension is constructed from the preprocessed disease monitoring data, weather data, and public opinion data corresponding to each time point, and the second vector corresponding to each time point is arranged in chronological order. It is used to sequentially input data to the multi-layer LSTM model and perform verification on the multi-layer LSTM model.

When training the multi-layer LSTM model, the loss function of the multi-layer LSTM model may be defined as a mean square error, and by adjusting the parameters of the multi-layer LSTM model, the mean square error can be minimized. The value is obtained. The RMSprop algorithm can be used during the training process. RMSprop is an improved stochastic gradient descent algorithm. Since the mean square error and the RMSprop algorithm are conventional techniques, description thereof will be omitted here.

In step 107, the disease monitoring data, weather data, and public opinion data before the prediction time point are acquired from the preprocessed disease monitoring data, weather data, and public opinion data, and the disease monitoring data, weather data, and public opinion data before the prediction time point are acquired. Is input to the optimized multi-layer RSTM model, and the disease monitoring result at the time of the prediction is obtained.

The acquired disease monitoring data, weather data, and public opinion data before the prediction time point are time series data. In the acquired disease monitoring data, weather data, and public opinion data before the predicted time point, a third vector of a predetermined dimension is constructed from the preprocessed disease monitoring data, weather data, and public opinion data corresponding to each time point, and at each time point. The corresponding third vector can be sequentially input into the multi-layer RSTM model in chronological order to predict the disease at the time of prediction.

At the time of disease prediction, from the initial time point to obtaining the predicted value at the predetermined time point, the optimized multi-layer RSTM model combines the input data at the current time point and the hidden layer unit at the previous time point layer by layer. By obtaining each hidden layer unit at the current time point, obtaining the predicted value at the current time point based on the hidden layer unit at the current time point, and constantly performing recursive processing in chronological order, the following Get the hidden layer unit and predicted value at the time of.

Example 1 predicts morbidity data by a multi-layer RSTM model. The LSTM model can directly extract knowledge from the data, construct a feature vector useful for prediction, and improve the prediction accuracy. Further, compared with the conventional RNN model, the LSTM model solves the problem that the gradient disappears at the time of long-term dependence caused when the amount of time series data is too large. Then, in Example 1, weather data and public opinion data are used for disease prediction as influential factors to improve the accuracy of disease prediction.

Example 2
FIG. 2 is a detailed flowchart for acquiring weather data (that is, step 102 in FIG. 1) associated with disease monitoring data from the disease prediction method according to the second embodiment of the present application.

The weather data can be captured by an internet worm via an API interface opened on a weather information website. Specifically, as shown in FIG. 2, the following steps may be included.

In step 201, a seed URL and a subsequent URL for the API interface of the weather information website are generated.

The seed URL is the basis and premise for the Internet worm to do all the work. The seed URL may be one or a plurality. The structural features of the URL of the weather information website can be analyzed and subsequent URLs can be obtained based on the structural features of the URL.

In step 202, an HTTP request is sent to the API interface of the weather information website to request access to the API interface.

The GET method can send an HTTP request to the API interface of the weather information website. When the weather information website agrees to acquire the weather data provided by the website, it notifies that the operation for acquiring the weather data is to be performed by returning an HTTP response.

In step 203, the data content is examined by analyzing and identifying the data content provided by the weather information website.

Since the weather information website provides the data content in a specific format, it is necessary to analyze and identify the data content in the specific format provided by the weather information website in order to examine the data content. For example, the data format provided by the API interface of the weather information website is JSON. JSON is a data exchange format that uses grammatical customs similar to C language. The data content is examined by analyzing and identifying the data content in the JSON format.

In step 204, it is determined whether or not the data content is a predetermined information content.

In order to acquire specific weather data, it must be determined whether or not the data content is a predetermined information content. If the data content is not the predetermined information content, the data content is discarded, and if not, the process proceeds to the next step.

In step 205, when the data content is a predetermined information content, the data content is captured.

The ultimate goal of data capture is to capture and localize the data content in the network. When capturing the data content in the JSON format, the state space search can be performed by the depth-first search strategy.

In step 206, the captured data content is locally stored as the weather data.

A database can be created in a computer device, and the weather data can be stored in the database.

All conventional Internet worms first set one or more entry URLs, and in the process of capturing a web page, follow the capture strategy and extract a new URL from the current web page and queue it. , Acquires the web page content corresponding to the URL, stores the web page content locally, extracts a valid address as the next entry URL, and stops when the crawl is completed. As the number of web pages increases dramatically, traditional Internet worms download a large number of irrelevant web pages. By capturing the weather data by internet worms through the API interface opened on the weather information website, you can avoid downloading irrelevant web pages, get the weather data efficiently, and get the disease. Improve prediction efficiency.

Example 3
FIG. 3 is a block diagram of the disease prediction device according to the third embodiment of the present application. As shown in FIG. 3, the disease prediction device 10 includes a first acquisition unit 301, a second acquisition unit 302, a third acquisition unit 303, a pretreatment unit 304, a construction unit 305, an optimization unit 306, and a prediction unit 307. It may be included.

The first acquisition unit 301 is used to acquire disease monitoring data which is time series data.

The disease monitoring data may include disease data such as influenza, hand-foot-and-mouth disease, measles, and mumps.

Establish a disease monitoring network consisting of multiple monitoring bases in a preset area (for example, province, city, region), acquire disease monitoring data from the monitoring bases, and use the disease monitoring data for disease monitoring. Let it be series data. Medical institutions, schools, childcare facilities, pharmacies, etc. can be selected as monitoring bases for disease monitoring and data collection for the corresponding targets. You can select a location that meets the preset conditions as a monitoring base. The preset conditions may include the number of people, scale, and the like. For example, select a school or childcare facility where the number of students reaches a preset number as a monitoring base. In addition, a pharmacy whose scale (for example, statistics based on daily sales) reaches a preset scale is selected as a monitoring base, and a hospital whose scale (for example, statistics based on the number of examinees per day) reaches a preset scale is selected as a monitoring base.

Disease monitoring data at different time points constitutes time series data for disease monitoring. For example, disease monitoring data collected on a daily basis is used as time-series data for disease monitoring. Alternatively, the disease monitoring data collected on a weekly basis is used as time-series data for disease monitoring.

Medical institutions (mainly hospitals) are the best place to detect early signs of disease and are a top priority for disease monitoring. Acquire disease monitoring data based on the patient's consultation status.

Since some patients with some diseases go to pharmacies to buy medicines to relieve early symptoms, disease monitoring data is acquired based on the sales status of medicines at pharmacies.

Since children and adolescents are at high risk of developing the disease and are an important part of disease transmission, surveillance of this target should also be strengthened. Schools and childcare facilities are good places to monitor disease outbreaks in children and adolescents. Acquire disease monitoring data based on the vacation status of children and adolescents in schools and childcare facilities.

Therefore, in the present invention, medical institutions, schools, childcare facilities and pharmacies are mainly selected for collecting disease monitoring data. Of course, the above data source selection does not add or replace other notable targets or locations as monitoring data sources only in other embodiments. For example, the hotel is included in the disease monitoring area, and the disease monitoring data of the resident of the hotel is acquired.

If necessary, only the disease monitoring data collected by any type of monitoring base (for example, a medical institution) can be acquired to configure the disease monitoring time series data. For example, disease monitoring data collected in a hospital can be acquired to construct time-series data for disease monitoring. Alternatively, disease monitoring data collected from a plurality of types of monitoring bases can be combined to form time-series data for disease monitoring. For example, time-series data for disease monitoring can be constructed mainly with disease monitoring data collected at a hospital and supplemented with disease monitoring data collected from a pharmacy.

Disease monitoring data may include disease data such as the number of disease visits, consultation rates, morbidity numbers, and morbidity rates. For example, the number of daily consultations for a disease (for example, influenza) may be obtained from a medical institution (for example, a hospital), and the number of daily consultations for a disease (for example, influenza) may be used as disease monitoring data. Alternatively, the daily prevalence of a student's disease (eg influenza) is obtained from the school and the daily prevalence of the disease (eg influenza) is used as disease monitoring data.

The second acquisition unit 302 is used to acquire weather data that is related to the disease monitoring data and is time-series data corresponding to the disease monitoring data.

The weather data associated with the disease monitoring data is the weather data that affects the disease monitoring data (that is, the disease morbidity data). The effect of different weather data on the disease monitoring data can be analyzed in advance, and the weather that affects the disease monitoring data or has a more serious effect can be determined based on the analysis result.

The weather data may include humidity, temperature, atmospheric pressure, precipitation, vapor pressure, wind speed, wind direction, and sunshine duration. In a specific embodiment, the weather data may include daily mean temperature, mean barometric pressure, maximum temperature, minimum temperature, average relative humidity, minimum relative humidity, precipitation, average wind speed, sunshine duration, average vapor pressure.

The weather data is the same as the time zone corresponding to the disease monitoring data, and the weather data is the same as the statistical cycle of the disease monitoring data (eg, daily, weekly). For example, when the disease monitoring data is the number of daily consultations in January-February 2018, the weather data is the daily weather data in January-February 2018. When the disease monitoring data is the number of weekly consultations from January to December 2017, the weather data is the weekly weather data from January to December 2017 (for example, weekly average temperature).

By capturing the weather from weather information websites (eg, China Weather Network, Sina Weather, Sohu Weather, etc.), the reliability of the weather data can be improved. As will be appreciated, the weather data can be captured from any web page.

It is possible to capture the weather data of a predetermined area. The predetermined area may include provinces, cities, regions and the like. For example, capture weather data for Shenzhen city.

It is possible to capture weather data for a predetermined time. The predetermined time may include a year, a month, a day, and the like. For example, capture daily weather data for January-February 2018.

The weather data can be captured by an internet worm.

The weather data can be captured by an API interface opened on a weather information website (for example, an API interface opened on the China Weather Network).

In a specific embodiment, the weather data can be captured by an internet worm via an API interface opened on a weather information website. See Figure 2 for the specific process of capturing the weather data by an internet worm via the API interface opened on the weather information website.

The third acquisition unit 303 is used to acquire public opinion data which is time-series data related to the disease monitoring data and corresponding to the disease monitoring data.

The public opinion data associated with the disease monitoring data is public opinion data indicating the disease monitoring data. As an example, during an epidemic of a disease (eg, influenza), as the number of people affected increases, people searching the Internet for disease-related words (eg, certain words such as influenza, Tamiflu, and high fever). As the number increases, the amount of searches for these words increases significantly. In addition, during the epidemic of a disease (eg, influenza), disease-related content (eg, disease information, treatment information) published on public opinion websites such as news, forums, blogs, and bulletin boards as the number of affected people increases. Etc.) are increasing. Therefore, disease can be predicted based on disease monitoring data and associated public opinion data.

The opinion data may include the number of searches for a particular word. For example, the number of searches for a specific word by a predetermined search engine (for example, the number of searches per day for a specific word by a predetermined search engine in a specific area) can be statistic.

The opinion data may also include the number of opinion information about a particular word contained on a particular opinion website (eg, news, forums, blogs, bulletin boards, etc.).

The specific word is a word related to a predicted disease, for example, if the specific word is a word related to a symptom of a disease, and the predicted disease is influenza, the specific word. May include sudden onset, high fever, chills, headache, weakness, throat irritation, myalgia, dry cough, etc. Also, if the predicted disease is hand-foot-and-mouth disease, the particular word may include mouth pain, anorexia nervosa, low-grade fever, small cold sores on the hands, small ulcers on the mouth, and the like.

The opinion data is the same as the time zone corresponding to the disease monitoring data, and the opinion data is the same as the statistical cycle of the disease monitoring (for example, daily, weekly). For example, if the disease monitoring data is the number of daily consultations from January to February 2018, the opinion data is the daily opinion data from January to February 2018 (for example, a specific word is searched per day). Number of times). Further, if the disease monitoring data is the number of weekly consultations from January to December 2017, the opinion data is the weekly opinion data from January to December 2017 (for example, the number of times a specific word is searched per week). ).

The pretreatment unit 304 is used to preprocess the disease monitoring data, weather data, and public opinion data.

Preprocessing for disease monitoring data, weather data and public opinion data may include anomalous data processing. The reason why the abnormality data processing is performed on the disease monitoring data, the weather data, and the public opinion data is to correct the abnormality data in the disease monitoring data, the weather data, and the public opinion data, and to improve the reliability and accuracy of the disease prediction. ..

The anomaly data processing may include complementing missing values in the disease monitoring data, weather data and public opinion data. The missing value can be complemented by the average or median of the data before and after the missing value, or the missing value can be complemented by a method called regression fitting.

Furthermore, the anomaly data processing can also include correcting outliers in the disease monitoring data, weather data and public opinion data. The outlier is a numerical value that greatly deviates from other data. The outliers can be corrected by an interpolation method.

Preprocessing for disease monitoring data, weather data and public opinion data may also include conversion of the data format to said disease monitoring data, weather data and public opinion data. For example, by performing standardization processing on disease monitoring data, weather data, and public opinion data, the disease monitoring data, weather data, and public opinion data are provided with a standard format of consistency and used as input data of an LSTM model. Suitable.

The construction unit 305 is used to construct a multi-layer long short-term memory recurrent neural network (Long Short-term Memory Recurrent Neural Network) model, that is, a multi-layer RSTM model. The multi-layer LSTM model includes two LSTM cell layers and one fully connected layer, and the first LSTM cell layer is an input composed of input data (for example, the disease monitoring data, weather data and public opinion data). (Data) used to build features and acquire first hidden layer units, the second layer LSTM cell layer combines the first hidden layer units and the second hidden layer unit. The fully connected layer is used to obtain a prediction result (eg, disease prediction result) based on the second hidden layer unit, and the LSTM cell layer is used to obtain an oblivion gate and an input, respectively. The storage state of the LSTM cell layer is controlled by the oblivion gate, the input gate, and the output gate, including the gate and the output gate.

The multi-layer LSTM model used in the method includes two LSTM cell layers and one fully connected layer, the first LSTM cell layer being input data (eg, said disease monitoring data, weather data and public opinion data). The LSTM cell layer of the second layer is a combination of the first hidden layer units, and is used to construct a feature for the input data composed of the first hidden layer unit. (Ii) Used to obtain a hidden layer unit. The fully connected layer obtains a predicted value based on the second hidden layer unit. The first non-display layer unit is a local feature, while the second non-display layer unit is an overall feature. That is, the LSTM cell layer of the first layer is used to extract local data, while the LSTM cell layer of the second layer is used to obtain the overall characteristics according to the local data. The fully connected layer is used to obtain a prediction result (for example, a disease prediction result) based on the overall characteristics.

The LSTM cell layer includes a forgetting gate, an input gate, and an output gate, and the storage state of the LSTM cell layer is controlled by the forgetting gate, the input gate, and the output gate. The input gate determines whether to receive the input at the current time, and the output gate determines whether to output the storage state.

In one embodiment, it is possible to calculate the forgetting gate _{f t} of LSTM cell layer, the input gate _{i t,} the output gate _{o t,} the memory state _{c t} and hidden layer units _{h t} as follows.

Among _{them, W f, U f, b} f is a parameter of the forgetting _{gate, W i, U i, b} i is the parameter of the input _{gate, W o, U o, b} o is a parameter of an output gate, W _c , U _c , and b _c are parameters of the storage unit.

In another embodiment, it is possible to calculate the forgetting gate _{f t} of LSTM cell layer, the input gate _{i t,} the output gate _{o t,} the memory state _{c t} and hidden layer units _{h t} as follows.

The optimization unit 306 acquires training data and verification data from the preprocessed disease monitoring data, weather data, and public opinion data, and performs training and performance verification on the multi-layer LSTM model using the training data and verification data. , Used to obtain an optimized multi-layer RSTM model.

Time-series data may be cut out from the preprocessed disease monitoring data, weather data, and public opinion data to form the training data and the verification data.

The input data of the multi-layer RSTM model is a vector having a predetermined dimension (for example, 1000 dimensions). In the cut time series data, a vector of a predetermined dimension is constructed from the preprocessed disease monitoring data, weather data, and public opinion data corresponding to each time point, and the vector corresponding to each time point is sequentially put into the multilayer RSTM model in chronological order. It can be input and used for training and verification of the multi-layer RSTM model.

For example, from the preprocessed disease monitoring data, weather data, and public opinion data, the first time series data for training the multi-layer RSTM model is cut out. In the cut out first temporary series data, the first vector of a predetermined dimension is constructed from the preprocessed disease monitoring data, weather data, and public opinion data corresponding to each time point, and the first vector corresponding to each time point is arranged in chronological order. It is used to sequentially input data into the multi-layer LSTM model and train the multi-layer LSTM model. From the preprocessed disease monitoring data, weather data and public opinion data, second time series data for verification against the multi-layer RSTM model is cut out. In the cut second time series data, a second vector of a predetermined dimension is constructed from the preprocessed disease monitoring data, weather data, and public opinion data corresponding to each time point, and the second vector corresponding to each time point is arranged in chronological order. It is used to sequentially input data to the multi-layer LSTM model and perform verification on the multi-layer LSTM model.

When training the multi-layer LSTM model, the loss function of the multi-layer LSTM model may be defined as a mean square error, and by adjusting the parameters of the multi-layer LSTM model, the mean square error can be minimized. The value is obtained. The RMSprop algorithm can be used during the training process. RMSprop is an improved stochastic gradient descent algorithm. Since the mean square error and the RMSprop algorithm are conventional techniques, description thereof will be omitted here.

The prediction unit 307 acquires the disease monitoring data, the weather data, and the public opinion data before the prediction time point from the preprocessed disease monitoring data, the weather data, and the public opinion data, and the disease monitoring data, the weather data, and the public opinion before the prediction time point. The data is input to the optimized multi-layer RSTM model to obtain disease monitoring results at the time of prediction.

The acquired disease monitoring data, weather data, and public opinion data before the prediction time point are time series data. In the acquired disease monitoring data, weather data, and public opinion data before the predicted time point, a third vector of a predetermined dimension is constructed from the preprocessed disease monitoring data, weather data, and public opinion data corresponding to each time point, and at each time point. The corresponding third vector can be sequentially input into the multi-layer RSTM model in chronological order to predict the disease at the time of prediction.

At the time of disease prediction, from the initial time point to obtaining the predicted value at the predetermined time point, the optimized multi-layer RSTM model combines the input data at the current time point and the hidden layer unit at the previous time point layer by layer. By obtaining each hidden layer unit at the current time point, obtaining the predicted value at the current time point based on the hidden layer unit at the current time point, and constantly performing recursive processing in chronological order, the following Get the hidden layer unit and predicted value at the time of.

Example 3 predicts morbidity data by a multi-layer RSTM model. The LSTM model can directly extract knowledge from the data, construct a feature vector useful for prediction, and improve the prediction accuracy. Furthermore, compared to the conventional RNN model, the LSTM model solves the problem that the gradient disappears during the long-term dependence caused when the amount of time series data is too large. Then, in Example 3, weather data and public opinion data are used for disease prediction as influential factors to improve the accuracy of disease prediction.

Example 4
FIG. 4 is a detailed configuration diagram of the second acquisition unit (that is, 302 in FIG. 3) of the disease prediction device according to the fourth embodiment of the present application.

The second acquisition unit 302 can capture the weather data by an internet worm via an API interface opened on the weather information website. As shown in FIG. 4, the second acquire unit 302 includes a generation subunit 3021, a request subunit 3022, an analysis subunit 3023, a judgment subunit 3024, a capture subunit 3025, and a storage subunit 3026. It may be.

The generation subunit 3021 is used to generate a seed URL and a subsequent URL for the API interface of the weather information website.

The seed URL is the basis and premise for the Internet worm to do all the work. The seed URL may be one or a plurality. The structural features of the URL of the weather information website can be analyzed and subsequent URLs can be obtained based on the structural features of the URL.

The request subunit 3022 is used to send an HTTP request to the API interface of the weather information website and request access to the API interface.

The GET method can send an HTTP request to the API interface of the weather information website. When the weather information website agrees to acquire the weather data provided by the website, it notifies that the operation for acquiring the weather data is to be performed by returning an HTTP response.

The analysis subunit 3023 is used to examine the data content by analyzing and identifying the data content provided by the weather information website.

Since the weather information website provides the data content in a specific format, it is necessary to analyze and identify the data content in the specific format provided by the weather information website in order to examine the data content. For example, the data format provided by the API interface of the weather information website is JSON. JSON is a data exchange format that uses grammatical customs similar to C language. The data content is examined by analyzing and identifying the data content in the JSON format.

The determination subunit 3024 is used to determine whether or not the data content is a predetermined information content.

In order to acquire specific weather data, it must be determined whether or not the data content is a predetermined information content. If the data content is not the predetermined information content, the data content is discarded, and if not, the process proceeds to the next step.

The capture subunit 3025 is used to capture the data content when the data content is a predetermined information content.

The ultimate goal of data capture is to capture and localize the data content in the network. When capturing the data content in the JSON format, the state space search can be performed by the depth-first search strategy.

The storage subunit 3026 is used to locally store the captured data content as the weather data.

A database can be created in a computer device, and the weather data can be stored in the database.

As the number of web pages increases dramatically, traditional Internet worms will download a large number of irrelevant web pages. The second acquisition unit 302 avoids downloading irrelevant web pages by capturing the weather data by an internet worm via an API interface opened on the weather information website, and streamlines the weather data. It is possible to improve the disease prediction efficiency.

Example 5
FIG. 5 is a schematic view of the computer device according to the fifth embodiment of the present application. The computer device 1 includes a memory 20, a processor 30, and a computer-readable command 40 such as a disease prediction program stored in the memory 20 and executed by the processor 30. When the processor 30 executes the computer-readable command 40, the steps in the embodiment of the disease prediction method are realized, for example, steps 101-107 shown in FIG. Alternatively, when the processor 30 executes the computer-readable command 40, the function of each module / unit in the embodiment of the device is realized, for example, units 301-307 in FIG.

Illustratively, the computer-readable command 40 is divided into one or more modules / units, the one or more modules / units are stored in the memory 20 and executed by the processor 30. Therefore, the present invention is realized. The one or more modules / units may be a series of computer-readable command segments capable of achieving a particular function, which describes the process of executing the computer-readable command 40 in the computer apparatus 1. Is for. For example, the computer-readable command 40 can be divided into units 301, 302, 303, 304, 305, 306, and 307 in FIG. 3, and for the functions of each unit, refer to Example 3.

The computer device 1 may be a computing device such as a desktop computer, a laptop computer, a palmtop computer, and a cloud server. As will be appreciated by those skilled in the art, the schematic view 5 is merely an example of a computer device 1, does not limit the computer device 1, and has more or less members than those shown in the figure. , Or a part of the members, or a combination of different members. For example, the computer device 1 may further include an input / output device, a network access device, a bus, and the like.

The processor 30 may be a central processing unit (CPU), another general-purpose processor, a digital signal processor (DSP), an integrated circuit for a specific application (Application Special Integrated Circuit, ASIC). ), A field programmable gate array (Field-Programmable Gate Array, FPGA), or other programmable logic device, discrete gate, or transistor logic device, discrete hardware component, and the like. The general-purpose processor may be a microprocessor, or the processor 30 may be any conventional processor or the like, and the processor 30 is a control center of the computer device 1 and is a computer device via various interfaces and lines. 1 Connect each part of the whole.

The memory 20 can be used to store the computer-readable commands 40 and / or modules / units, and the processor 30 stores computer-readable commands and / or modules / units stored in the memory 20. Each function of the computer device 1 is realized by operating or executing the data and calling the data stored in the memory 20. The memory 20 may mainly include a program storage area and a data storage area, wherein the program storage area includes an operating system and an application required for at least one function (for example, a voice reproduction function, an image reproduction function, etc.). Etc. can be memorized. Data (for example, audio data, telephone directory, etc.) created according to the use of the computer device 1 can be stored in the data storage area. Further, the memory 20 may include a high-speed random access memory, a non-volatile memory such as a hard disk, a memory, and a plug-in hard disk, a SmartMedia Card (SMC), and a secure digital (Secure Digital, SD). ) Cards, Flash Cards, at least one disk storage device, flash storage devices, or other volatile solid state storage devices.

The module / unit integrated in the computer device 1 is realized in the form of a software functional unit, and can be stored in a non-volatile readable storage medium when sold or used as an independent product. Based on this understanding, the present application can implement all or part of the process in the methods of the above embodiments, and can also command the relevant hardware to complete with computer-readable commands, said computer. The readable command is stored in a non-volatile readable storage medium, and when the computer readable command is executed by a processor, the steps of the embodiments of each of the above methods can be realized. The non-volatile readable storage medium is any entity or device capable of carrying the computer-readable command code, a recording medium, a U disk, a mobile hard disk, a magnetic disk, an optical disk, a computer memory, a read-only memory (ROM, Read-Only Memory), and the like. Random access memory (RAM, Random Access Memory), electric carrier signals, telecommunications signals, software distribution media, and the like may be included.

It is also clear that the word "contains" does not exclude other units or steps, nor does it exclude the singular or the plural. The plurality of units or computer devices described in the computer device claims can also be realized through software or hardware by the same unit or computer device. Words such as first and second do not represent any particular order, but rather names.

Finally, it should be noted that the above embodiments are used only to illustrate the technical solutions of the present invention and are not limited thereto. The invention of the present application will be described in detail with reference to preferred embodiments, but those skilled in the art will modify or equivalent the technical solution of the present invention without departing from the spirit and scope of the technical solution of the present invention. It should be understood that it can be replaced with.

Claims (20)

The disease monitoring data is acquired, and the disease monitoring data is a time-series data step and
A step of acquiring weather data related to the disease monitoring data, and the weather data is time-series data corresponding to the disease monitoring data,
The public opinion data related to the disease monitoring data is acquired, and the public opinion data is a step that is time-series data corresponding to the disease monitoring data.
A step of preprocessing the disease monitoring data, the weather data, and the public opinion data, and
Multi-layer long short-term memory recurrent neural network model, that is, the steps to build a multi-layer RSTM model,
Training data and verification data are acquired from the preprocessed disease monitoring data, weather data, and public opinion data, and the training data and the verification data are used to perform training and performance verification on the multilayer LSTM model, which is optimal. Steps to obtain a multi-layered RSTM model
The preprocessed disease monitoring data, the weather data, and the public opinion data are used to acquire the disease monitoring data, the weather data, and the public opinion data before the prediction time point, and the disease monitoring data, the weather data, and the public opinion data before the prediction time point are obtained. Steps to input into an optimized multi-layer LSTM model and obtain disease monitoring results at the time of prediction,
A method for predicting a disease, which comprises. The steps to capture weather data from a web page are
Steps to generate seed URLs and subsequent URLs for the weather information website API interface,
A step of sending an HTTP request to the API interface of the weather information website and requesting access to the API interface.
The steps of analyzing and identifying the data content provided by the weather information website and examining the data content,
A step of determining whether or not the data content is a predetermined information content, and
When the data content is a predetermined information content, the step of capturing the data content and
The step of locally storing the captured data contents as the weather data, and
The method according to claim 1, wherein the method is provided. The method according to claim 1, wherein the public opinion data includes the number of searches for a specific word or the number of public opinion information about a specific word contained in a specific public opinion website. The step of preprocessing the disease monitoring data, the weather data, and the public opinion data is
Steps to supplement missing values to the disease monitoring data, the weather data, and the public opinion data,
Steps to correct outliers in the disease monitoring data, the weather data and the public opinion data, and
The step of converting the data format to the disease monitoring data, the weather data, and the public opinion data, and
The method according to claim 1, wherein the method is provided. The method according to any one of claims 1-4, wherein the weather data includes humidity, temperature, atmospheric pressure, precipitation, vapor pressure, wind speed, wind direction, and sunshine duration. The multi-layer LSTM model includes two LSTM cell layers and one fully connected layer, and the first LSTM cell layer builds features on the input data to acquire the first hidden layer unit. The second LSTM cell layer is used to combine the first non-display layer units to obtain a second non-display layer unit, and the fully connected layer is the second non-display layer unit. Each LSTM cell layer includes an oblivion gate, an input gate, and an output gate, respectively, and the storage state of the LSTM cell layer is obtained by the oblivion gate, the input gate, and the output gate. The method according to any one of claims 1-4, wherein is controlled. The invention according to any one of claims 1-4, wherein the loss function used in the process of training the multi-layer RSTM model is a mean square error, and the algorithm used is an RMSprop algorithm. Method. Disease monitoring data is acquired, and the disease monitoring data is the first acquisition unit, which is time-series data, and
The weather data related to the disease monitoring data is acquired, and the weather data includes a second acquisition unit which is time-series data corresponding to the disease monitoring data.
The public opinion data related to the disease monitoring data is acquired, and the public opinion data includes a third acquisition unit which is time-series data corresponding to the disease monitoring data.
A preprocessing unit that preprocesses the disease monitoring data, the weather data, and the public opinion data, and
A multi-layer long short-term memory recurrent neural network model, that is, a construction unit for constructing a multi-layer RSTM model,
Training data and verification data are acquired from the preprocessed disease monitoring data, weather data, and public opinion data, and training and performance verification are performed on the multi-layer LSTM model using the training data and the verification data. , An optimization unit that provides an optimized multi-layer RSTM model,
The preprocessed disease monitoring data, the weather data, and the public opinion data are used to acquire the disease monitoring data, the weather data, and the public opinion data before the prediction time point, and the disease monitoring data, the weather data, and the public opinion data before the prediction time point are obtained. A prediction unit that inputs to an optimized multi-layer RSTM model and obtains disease monitoring results at the time of prediction.
A disease prediction device comprising. It ’s a computer device,
The computer device comprises a memory and a processor, which stores at least one computer-readable command, and the processor executes at least one computer-readable command.
The disease monitoring data is acquired, and the disease monitoring data is a time-series data step and
A step of acquiring weather data related to the disease monitoring data, and the weather data is time-series data corresponding to the disease monitoring data,
The public opinion data related to the disease monitoring data is acquired, and the public opinion data is a step that is time-series data corresponding to the disease monitoring data.
A step of preprocessing the disease monitoring data, the weather data, and the public opinion data, and
Multi-layer long short-term memory recurrent neural network model, that is, the steps to build a multi-layer RSTM model,
Training data and verification data are acquired from the preprocessed disease monitoring data, weather data, and public opinion data, and the training data and verification data are used to perform training and performance verification on the multilayer RSTM model for optimization. Steps to obtain a multi-layered RSTM model
The preprocessed disease monitoring data, the weather data, and the public opinion data are used to acquire the disease monitoring data, the weather data, and the public opinion data before the prediction time point, and the disease monitoring data, the weather data, and the public opinion data before the prediction time point are obtained. Steps to input into an optimized multi-layer LSTM model and obtain disease monitoring results at the time of prediction,
A computer device characterized by realizing. The steps to capture weather data from the web page
Steps to generate seed URLs and subsequent URLs for the weather information website API interface,
A step of sending an HTTP request to the API interface of the weather information website and requesting access to the API interface.
The steps of analyzing and identifying the data content provided by the weather information website and examining the data content,
A step of determining whether or not the data content is a predetermined information content, and
When the data content is a predetermined information content, the step of capturing the data content and
The step of locally storing the captured data contents as the weather data, and
9. The computer device according to claim 9. The computer device according to claim 9, wherein the public opinion data includes the number of searches for a specific word or the number of public opinion information about a specific word contained in a specific public opinion website. The step of preprocessing the disease monitoring data, the weather data, and the public opinion data is
Steps to supplement missing values to the disease monitoring data, the weather data, and the public opinion data,
Steps to correct outliers in the disease monitoring data, the weather data and the public opinion data, and
The step of converting the format of the disease monitoring data, the weather data, and the public opinion data, and
9. The computer device according to claim 9. The computer according to any one of claims 9-12, wherein the weather data includes humidity, temperature, atmospheric pressure, precipitation, vapor pressure, wind speed, wind direction, and sunshine duration. apparatus. The multi-layer LSTM model includes two LSTM cell layers and one fully connected layer, for the first LSTM cell layer to build features for input data and acquire a first hidden layer unit. The second layer of the LSTM cell layer is used to combine the first non-display layer units to obtain a second non-display layer unit, and the fully connected layer is the second non-display layer unit. The LSTM cell layer includes an oblivion gate, an input gate, and an output gate, respectively, and the oblivion gate, the input gate, and the output gate cause the storage state of the LSTM cell layer to be stored. The computer device according to any one of claims 9-12, characterized in that it is controlled. It is a non-volatile readable storage medium
The non-volatile readable storage medium stores at least one computer readable command.
The at least one computer-readable command is determined by the processor.
The disease monitoring data is acquired, and the disease monitoring data is a time-series data step and
A step of acquiring weather data related to the disease monitoring data, and the weather data is time-series data corresponding to the disease monitoring data,
The public opinion data related to the disease monitoring data is acquired, and the public opinion data is a step that is time-series data corresponding to the disease monitoring data.
A step of preprocessing the disease monitoring data, the weather data, and the public opinion data, and
Multi-layer long short-term memory recurrent neural network model, that is, the steps to build a multi-layer RSTM model,
Training data and verification data are acquired from the preprocessed disease monitoring data, weather data, and public opinion data, and the training data and the verification data are used to perform training and performance verification on the multilayer LSTM model, which is optimal. Steps to obtain a multi-layered RSTM model
The preprocessed disease monitoring data, the weather data, and the public opinion data are used to acquire the disease monitoring data, the weather data, and the public opinion data before the prediction time point, and the disease monitoring data, the weather data, and the public opinion data before the prediction time point are obtained. Steps to input into an optimized multi-layer LSTM model and obtain disease monitoring results at the time of prediction,
A non-volatile readable storage medium characterized by the realization of. The steps to capture weather data from a web page are
Steps to generate seed URLs and subsequent URLs for the weather information website API interface,
A step of sending an HTTP request to the API interface of the weather information website and requesting access to the API interface.
The steps of analyzing and identifying the data content provided by the weather information website and examining the data content,
A step of determining whether or not the data content is a predetermined information content, and
When the data content is a predetermined information content, the step of capturing the data content and
The step of locally storing the captured data contents as the weather data, and
15. The storage medium according to claim 15, wherein the storage medium is provided with. The storage medium according to claim 15, wherein the public opinion data includes the number of searches for a specific word or the number of public opinion information about a specific word contained in a specific public opinion website. The preprocessing performed on the disease monitoring data, the weather data, and the public opinion data is
Steps to supplement missing values to the disease monitoring data, the weather data, and the public opinion data,
Steps to correct outliers in the disease monitoring data, the weather data and the public opinion data, and
The step of converting the data format to the disease monitoring data, the weather data, and the public opinion data, and
15. The storage medium according to claim 15, wherein the storage medium is provided with. The storage medium according to any one of claims 15-18, wherein the weather data includes humidity, air temperature, atmospheric pressure, precipitation, vapor pressure, wind speed, wind direction, and sunshine time. .. The multi-layer LSTM model includes two LSTM cell layers and one fully connected layer, for the first LSTM cell layer to build features for input data and acquire a first hidden layer unit. The second layer of the LSTM cell layer is used to combine the first non-display layer units to obtain a second non-display layer unit, and the fully connected layer is the second non-display layer unit. The LSTM cell layer includes an oblivion gate, an input gate, and an output gate, respectively, and the memory state of the LSTM cell layer is determined by the oblivion gate, the input gate, and the output gate, respectively. The storage medium according to any one of claims 15-18, characterized in that it is controlled.

Images