CN115482928A

CN115482928A - Data prediction method, device, equipment and storage medium based on artificial intelligence

Info

Publication number: CN115482928A
Application number: CN202210817079.1A
Authority: CN
Inventors: 司世景; 王健宗; 邱与林
Original assignee: Ping An Technology Shenzhen Co Ltd
Current assignee: Ping An Technology Shenzhen Co Ltd
Priority date: 2022-07-12
Filing date: 2022-07-12
Publication date: 2022-12-16

Abstract

The invention relates to the field of artificial intelligence, and discloses a data prediction method, a device, equipment and a storage medium based on artificial intelligence, wherein the method comprises the following steps: acquiring a target prediction model consisting of a multilayer perceptron and a long-term and short-term memory network; acquiring health index data of a target user in a plurality of historical time periods, and converting each piece of health index data into a health index vector to obtain a plurality of health index vectors; inputting each health index vector into a multilayer perceptron to obtain a first predicted value corresponding to each historical time period; sequentially inputting the first predicted value corresponding to each historical time period into the long-short term memory network according to the time sequence, and predicting to obtain a target predicted value; determining the health condition of the target user in the current time period according to the target predicted value; in the invention, under the condition of not sacrificing the prediction capability of the model, the structure of the target prediction model is simpler, and the requirement of the target prediction model on computing power is reduced, thereby improving the practicability of the target prediction model.

Description

Data prediction method, device, equipment and storage medium based on artificial intelligence

Technical Field

The invention relates to the technical field of artificial intelligence, in particular to a data prediction method, a data prediction device, data prediction equipment and a storage medium based on artificial intelligence.

Background

Along with the improvement of living standard of people and cognition and intervention technical level of disease occurrence and development, people pay more and more attention to self health state, and the traditional mode of observing human health condition by abnormal manual work cannot meet the health monitoring requirement of people. Therefore, with the development of machine learning technology and the research of people on health conditions, a deep neural network model-based framework is often applied to the prediction of health conditions, so as to improve the artificial intelligence of the health condition prediction, thereby meeting the increasing health monitoring requirements of people.

However, in order to ensure the prediction accuracy of the model on the health condition, the traditional health condition prediction model has a complex and huge structure and more model parameters, and can be realized only by requiring extensive calculation force support when the prediction model is used for prediction, so that the practicability is poor.

Disclosure of Invention

The invention provides a data prediction method, a data prediction device, data prediction equipment and a storage medium based on artificial intelligence, which are used for solving the problems that a traditional health condition prediction model is complex and huge in structure, can be realized only by requiring a strong calculation force support and is poor in practicability.

Provided is a data prediction method based on artificial intelligence, comprising the following steps:

obtaining a target prediction model obtained by pre-training, wherein the target prediction model consists of a plurality of layers of perceptrons and a long-term and short-term memory network which are sequentially arranged;

acquiring health index data of a target user in a plurality of historical time periods, and converting each piece of health index data into a health index vector to obtain a plurality of health index vectors, wherein each health index vector corresponds to one historical time period;

inputting each health index vector into a multilayer perceptron, and predicting the health condition of corresponding historical time periods through the multilayer perceptron to obtain a first predicted value corresponding to each historical time period;

sequentially inputting the first predicted values corresponding to each historical time period into the long-short term memory network according to the time sequence, fitting the first predicted values corresponding to the multiple historical time periods through the long-short term memory network, and predicting to obtain target predicted values;

and determining the health condition of the target user in the current time period according to the target predicted value.

Provided is an artificial intelligence-based data prediction apparatus including:

the acquisition module is used for acquiring a target prediction model obtained by pre-training, and the target prediction model consists of a plurality of layers of perceptrons and a long-term and short-term memory network which are sequentially arranged;

the conversion module is used for acquiring health index data of a target user in a plurality of historical time periods, converting each health index data into a health index vector to obtain a plurality of health index vectors, wherein each health index vector corresponds to one historical time period;

the first prediction module is used for inputting each health index vector into the multilayer perceptron, and predicting the health condition of corresponding historical time periods through the multilayer perceptron to obtain a first prediction value corresponding to each historical time period;

the second prediction module is used for sequentially inputting the first prediction values corresponding to the historical time periods into the long-short term memory network according to the time sequence, fitting the first prediction values corresponding to the multiple historical time periods through the long-short term memory network and predicting the target prediction values;

and the determining module is used for determining the health condition of the target user in the current time period according to the target predicted value.

There is provided a computer device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, the processor when executing the computer program performing the steps of the artificial intelligence based data prediction method as described above.

A computer-readable storage medium is provided, in which a computer program is stored, wherein the computer program, when being executed by a processor, is adapted to carry out the steps of the artificial intelligence based data prediction method as described above.

In one scheme provided by the data prediction method, the data prediction device, the data prediction equipment and the storage medium based on the artificial intelligence, a target prediction model obtained by pre-training is obtained, and the target prediction model consists of a plurality of layers of perceptrons and a long-short term memory network which are sequentially arranged; acquiring health index data of a target user in a plurality of historical time periods, and converting each piece of health index data into a health index vector to obtain a plurality of health index vectors, wherein each health index vector corresponds to one historical time period; inputting each health index vector into a multilayer perceptron, and predicting the health condition of corresponding historical time periods through the multilayer perceptron to obtain a first predicted value corresponding to each historical time period; sequentially inputting the first predicted values corresponding to each historical time period into the long-short term memory network according to the time sequence, fitting the first predicted values corresponding to the multiple historical time periods through the long-short term memory network, and predicting to obtain target predicted values; determining the health condition of the target user in the current time period according to the target predicted value; according to the method, the health condition of the user in a future period of time is predicted through the target prediction model consisting of the multilayer perceptron and the long-term and short-term memory network, the structure of the target prediction model is simpler under the condition that the prediction capability of the model is not sacrificed, the requirement of the target prediction model on computing power is reduced, the target prediction model can be deployed to more devices to execute the prediction task, and therefore the practicability of the target prediction model is improved.

Drawings

In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings needed to be used in the description of the embodiments of the present invention will be briefly introduced below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art that other drawings can be obtained according to these drawings without inventive labor.

FIG. 1 is a schematic diagram of an application environment of an artificial intelligence based data prediction method according to an embodiment of the present invention;

FIG. 2 is a flow chart of an artificial intelligence based data prediction method according to an embodiment of the present invention;

FIG. 3 is a schematic flow chart of an artificial intelligence based data prediction method according to an embodiment of the present invention;

FIG. 4 is a schematic diagram of an implementation of step S62 in FIG. 3;

FIG. 5 is a flowchart illustrating an implementation of step S621 in FIG. 4;

FIG. 6 is a schematic diagram of an implementation of step S62 in FIG. 3;

FIG. 7 is a schematic diagram of another implementation flow after step S61 in FIG. 3;

FIG. 8 is a schematic diagram of an artificial intelligence data prediction apparatus in accordance with an embodiment of the present invention;

fig. 9 is a schematic structural diagram of a computer device according to an embodiment of the present invention.

Detailed Description

The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are some, not all, embodiments of the present invention. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.

The data prediction method based on artificial intelligence provided by the embodiment of the invention can be applied to the application environment shown in figure 1, wherein the terminal equipment is communicated with the server through a network. After receiving a prediction instruction sent by a user through terminal equipment, a server acquires a target prediction model obtained by pre-training, wherein the target prediction model consists of a plurality of layers of perceptrons and a long-short term memory network which are sequentially arranged, health index data of the target user in a plurality of historical time periods are acquired through the terminal equipment, the server converts each health index data into a health index vector to acquire a plurality of health index vectors, each health index vector corresponds to one historical time period, each health index vector is input into the multilayer perceptron, health condition prediction corresponding to the historical time periods is performed through the multilayer perceptron to acquire a first prediction value corresponding to each historical time period, the first prediction value corresponding to each historical time period is sequentially input into the long-short term memory network according to a time sequence, the first prediction values corresponding to the historical time periods are fitted through the long-short term memory network to acquire a target, finally, the health condition of the target user in the current time period is determined according to the target prediction value, the server executes a prediction task by using the target prediction model with a simpler structure, the requirement of the target prediction model on the computational load is reduced without sacrificing the prediction capability of the model, and the practicability of the target prediction is improved.

In other embodiments, the data prediction method based on artificial intelligence may also be executed by a terminal device, where the server pre-trains an obtained target prediction model, where the target prediction model is composed of multiple layers of perceptrons and a long-short term memory network arranged in sequence, and deploys the target prediction model to the terminal device, and when receiving a prediction instruction, the terminal device obtains the pre-trained target prediction model, obtains health index data of a target user in multiple historical time periods, converts each health index data into a health index vector, obtains multiple health index vectors, where each health index vector corresponds to one historical time period, inputs each health index vector into the multiple layers of perceptrons, performs health condition prediction of the corresponding historical time period by the multiple layers of perceptrons, obtains a first prediction value corresponding to each historical time period, sequentially inputs the first prediction value corresponding to each historical time period into the long-short term memory network according to a time sequence, fits the first prediction values corresponding to the multiple historical time periods by the long-short term memory network, obtains a target prediction value by prediction, and finally determines a health condition of the target user in a current time period according to the target prediction value; in the embodiment, the health condition of a user in a future period of time is predicted through the target prediction model consisting of the multilayer perceptron and the long-term and short-term memory network, the structure of the target prediction model is simpler under the condition of not sacrificing the prediction capability of the model, and the requirement of the target prediction model on computing power is reduced, so that the target prediction model can be deployed to more devices (such as terminal devices and edge devices) to execute a prediction task, the practicability of the target prediction model is improved, and the artificial intelligence of a prediction system is improved.

The target prediction model and the data such as the health index data of each historical time period are stored in a database of the terminal equipment or the server, so that when a prediction task is executed subsequently, relevant data can be directly obtained according to requirements, the data processing time is shortened, and the prediction efficiency is improved.

The database in this embodiment is stored in the blockchain network, and is used to store data used and generated in the artificial intelligence-based data prediction method, such as a target prediction model, health index data of each historical time period, a target prediction value, and relevant data of a user in a future time period, and the like. The block chain referred by the application is a novel application mode of computer technologies such as distributed data storage, point-to-point transmission, a consensus mechanism, an encryption algorithm and the like. A block chain (Blockchain), which is essentially a decentralized database, is a string of data blocks associated by using a cryptographic method, and each data block contains information of a batch of network transactions, which is used for verifying the validity (anti-counterfeiting) of the information and generating a next block. The blockchain may include a blockchain underlying platform, a platform product service layer, an application service layer, and the like. The database is deployed in the blockchain, so that the safety of data storage can be improved.

The terminal device may be, but is not limited to, various personal computers, notebook computers, smart phones, tablet computers, and portable wearable devices. The server may be implemented as a stand-alone server or as a server cluster consisting of a plurality of servers.

In an embodiment, as shown in fig. 2, a data prediction method based on artificial intelligence is provided, which is described by taking the terminal device in fig. 1 as an example, and includes the following steps:

s10: and obtaining a target prediction model obtained by pre-training, wherein the target prediction model consists of a plurality of layers of perceptrons and a long-term and short-term memory network which are sequentially arranged.

After receiving the prediction instruction, the terminal device needs to obtain a target prediction model obtained by pre-training, wherein the target prediction model is composed of a plurality of layers of perceptrons and a long-term and short-term memory network which are sequentially arranged. The target prediction model server trains the obtained prediction model in advance according to the health index data of the historical records of different users.

The target prediction model for predicting different symptoms adopts different health index data, namely the health index data is a health index related to the predicted symptoms; taking hepatitis prediction as an example, the adopted health index data comprises index data such as body temperature, aspartate aminotransferase, alanine aminotransferase and the like of a user. In this embodiment, the health index data including the index data of the user's body temperature, aspartate aminotransferase, alanine aminotransferase, etc. are only exemplary, and in other embodiments, the health index data may also be other health index data according to the difference of the predicted symptoms, which is not described herein.

S20: the method comprises the steps of obtaining health index data of a target user in multiple historical time periods, converting each piece of health index data into a health index vector, and obtaining multiple health index vectors.

After receiving the prediction instruction, the terminal device further needs to acquire health index data of the target user in a plurality of historical time periods, convert each health index data into a health index vector, and obtain a plurality of health index vectors, wherein each health index vector corresponds to one historical time period. The health index data can be user health status data which is stored in the terminal equipment and obtained by recording after the terminal equipment monitors the user in real time, and the user health status of the user is directly monitored through the terminal equipment, so that the health status can be predicted in real time according to the detected health index data, the health status of the user and the infection risk of corresponding symptoms can be known timely and accurately, and the prevention and treatment of diseases can be effectively assisted.

The acquisition mode of the health index data of the target user in a plurality of historical time periods is as follows: dividing historical health monitoring data of a user into historical monitoring data of different historical time periods according to a preset time period; the method comprises the steps of determining a plurality of health indexes corresponding to symptoms to be predicted, determining a historical monitoring value corresponding to each health index in monitoring data of each historical time period, and recording the historical monitoring values as health index data of the historical time period, wherein each health index data comprises the historical monitoring values corresponding to the health indexes, namely each health index data comprises a plurality of health index values, so that the health index data of the historical time periods are obtained.

Wherein, convert every health index data into health index vector, include: the method comprises the steps of inputting health index data of a target user in a historical time period into a preset word embedding model, carrying out word embedding processing on the health index data through the preset word embedding model to obtain a health index vector with a preset dimension, recording the health index vector as the health index vector corresponding to the time period of the user, traversing all the historical time periods, and obtaining a health sample vector corresponding to each historical time period.

The preset word embedding model can be a general neural network model or a model obtained by training according to corresponding symptom data in advance. The preset dimensionality is determined according to the number of the health index values in the historical health index data, and the larger the number of the health index values in the historical health index data is, the larger the preset dimensionality is. In addition, in order to ensure the validity of vector conversion, vector conversion can be performed according to the type of each health index value in the health index data. When the vector conversion is carried out on the health index data of a target user in each historical time period, the type (including qualitative indexes and quantitative indexes) of each health index value in the health index data needs to be determined firstly, all the health index values of which the types are the qualitative indexes are input into a preset word embedding model to be subjected to word embedding processing, an embedding vector is obtained, the word embedding dimension is determined according to the number of the qualitative indexes, the health index values of which the types are the quantitative indexes are input into an encoder to be subjected to vector coding, a coding vector is obtained, the embedding vector and the coding vector are spliced into one vector and recorded as the health index vector, different vector conversion operations are carried out according to the types of the health index values in the health index data, and finally the vector conversion vector is spliced into one health index vector.

S30: and inputting each health index vector into a multilayer perceptron, and predicting the health condition of the corresponding historical time period through the multilayer perceptron to obtain a first predicted value corresponding to each historical time period.

After a plurality of health index vectors are obtained, namely the health index vectors corresponding to a plurality of historical time periods of a user are obtained, the terminal equipment inputs each health index vector into the multilayer sensing machine, health condition prediction corresponding to the historical time periods is carried out through the multilayer sensing machine, namely, the infection probability of a certain symptom corresponding to the historical time periods is predicted through the multilayer sensing machine, and a first predicted value corresponding to each historical time period is obtained, wherein the first predicted value is the predicted infection probability of the user to the certain symptom in the corresponding historical time period.

S40: and sequentially inputting the first predicted value corresponding to each historical time period into the long-short term memory network according to the time sequence, fitting the first predicted values corresponding to the multiple historical time periods through the long-short term memory network, and predicting to obtain the target predicted value.

After the first predicted value corresponding to each historical time period is obtained, the first predicted value corresponding to each historical time period is sequentially input into the long-short term memory network according to the time sequence, the first predicted values corresponding to the multiple historical time periods are fitted through the long-short term memory network, and therefore a target predicted value is obtained through prediction, and the target predicted value is the predicted infection probability of a user to a certain symptom in the current time period.

S50: and determining the health condition of the target user in the current time period according to the target predicted value.

And after the first predicted values corresponding to the plurality of historical time periods are fitted through the long-short term memory network and the target predicted values are obtained through prediction, determining the health condition of the target user in the current time period according to the target predicted values. Because the target predicted value is the predicted infection probability of a user to a certain symptom in the current time period, the larger the target predicted value is, the higher the risk that the target user is infected with the corresponding symptom, and if the target predicted value is larger than a preset value (such as 0.5), it is indicated that the risk that the target user is infected with the corresponding symptom in the current time period is higher, and it is determined that the health condition of the target user in the current time period is not good at this moment, and the corresponding symptom is likely to be infected; if the target predicted value is less than or equal to the preset value (such as 0.5), it indicates that the risk of infection of the corresponding symptom of the target user in the current time period is low, and it is determined that the health condition of the target user in the current time period is normal.

After the health condition of the target user in the current time period is determined according to the target predicted value, alarm prompt information is generated according to the target predicted value, wherein the alarm prompt information comprises the target predicted value (namely the predicted infection probability of the user to a certain symptom in the current time period) and the health condition of the target user in the current time period, and the alarm prompt information is sent to the target user and/or related personnel (such as parents, doctors and the like) so that the target user, the related personnel can timely obtain the health condition of the target user, and corresponding measures are collected.

In the embodiment, a target prediction model obtained by pre-training is obtained, and the target prediction model consists of a plurality of layers of perceptrons and a long-term and short-term memory network which are sequentially arranged; the method comprises the steps of obtaining health index data of a target user in multiple historical time periods, converting each piece of health index data into a health index vector to obtain multiple health index vectors, wherein each health index vector corresponds to one historical time period; inputting each health index vector into a multilayer perceptron, and predicting the health condition of corresponding historical time periods through the multilayer perceptron to obtain a first predicted value corresponding to each historical time period; sequentially inputting the first predicted values corresponding to each historical time period into the long-short term memory network according to the time sequence, fitting the first predicted values corresponding to the multiple historical time periods through the long-short term memory network, and predicting to obtain target predicted values; determining the health condition of the target user in the current time period according to the target predicted value; according to the method, the health condition of the user in a future period is predicted through the target prediction model consisting of the multilayer perceptron and the long-term and short-term memory network, the structure of the target prediction model is simpler under the condition that the prediction capability of the model is not sacrificed, the requirement of the target prediction model on calculation power is reduced, the target prediction model can be deployed to more devices to execute prediction tasks, and therefore the practicability of the target prediction model is improved.

In an embodiment, before obtaining the target prediction model, the target prediction model needs to be obtained by training according to historical health index data of different users in different time periods. The training process of the target prediction model is as follows:

s01: acquiring historical health index data (namely historical health index data) of different users in different time periods;

before training the model, historical health index data of different users in different time periods are required to be acquired as sample data for model training. The preset time period is determined according to the symptom monitoring period, and the preset time period can be in a week unit, namely historical health index data of each user in the previous n weeks are obtained and serve as sample data. In other embodiments, the preset time period may be another time period, for example, the preset time period may be in units of days, that is, historical health index data of each user in the previous n days is obtained as sample data; the preset time period may also be in the unit of one month, which is not described herein. The different time periods are a plurality of consecutive time periods.

The method comprises the steps that historical health monitoring data of a user need to be divided into historical monitoring data of different time periods according to a preset time period; determining a plurality of health indexes corresponding to symptoms to be predicted, determining a historical monitoring value corresponding to each health index in monitoring data of each time period, and recording the historical monitoring values as health index data of the time period, wherein each health index data comprises the historical monitoring values corresponding to the plurality of health indexes, namely each health index data comprises a plurality of health index values, so that the health index data of the time periods are obtained.

For example, the health indexes corresponding to pneumonia are user body temperature, aspartate aminotransferase and alanine aminotransferase, and for each user, monitoring values of the health indexes such as user body temperature, aspartate aminotransferase and alanine aminotransferase are collected in a week unit and recorded as the health index data of the week, so that the health index data of the user in each week is obtained, and further the health index data of different users in each week is obtained.

S02: and performing vector conversion on the historical health index data of each user in each time period to obtain a health sample vector corresponding to each time period of the user, wherein each health sample vector corresponds to a symptom infection result label.

Then, inputting historical health index data of each user in a certain time period into a preset word embedding model, carrying out word embedding processing on the historical health index data through the preset word embedding model to obtain a health sample vector with a preset dimension, recording the health sample vector as a health sample vector corresponding to the user in the time period, and traversing all the time periods to obtain the health sample vector corresponding to each time period.

The preset word embedding model can be a general neural network model or a model obtained by training according to corresponding symptom data in advance. The preset dimensionality is determined according to the number of the health indexes in the historical health index data, and the larger the number of the health indexes in the historical health index data is, the larger the preset dimensionality is.

In order to ensure the effectiveness of vector conversion, vector conversion can be performed according to the types of the health index values in the historical health index data. When vector conversion is carried out on historical health index data of each user in each time period, the type (including qualitative indexes and quantitative indexes) of each health index value in the historical health index data needs to be determined, all health index values of which the types are the qualitative indexes are input into a preset word embedding model to be subjected to word embedding processing, a first vector is obtained, word embedding dimensionality is determined according to the number of the qualitative indexes, the health index values of which the types are the quantitative indexes are input into an encoder to be subjected to vector coding, a second vector is obtained, the first vector and the second vector are spliced into one vector and recorded as a health sample vector, different vector conversion operations are carried out according to the types of the health index values in the historical health index data, the first vector and the second vector are finally spliced into one health sample vector, and the effectiveness of vector conversion is improved on the basis of reducing data processing quantity.

S03: and training the preset model according to the health sample vectors corresponding to different users in each time period to obtain a target prediction model meeting the preset performance condition.

And performing vector conversion on the historical health index data of each user in each time period to obtain a health sample vector corresponding to each time period of the user, so as to obtain health sample vectors corresponding to different users in each time period, and then training a preset model according to the health sample vectors corresponding to different users in each time period to obtain a target prediction model meeting preset performance conditions. The preset model includes a Multilayer Perceptron (MLP) and a Long Short-Term Memory network (LSTM).

Wherein, training the preset model according to the health sample vectors corresponding to different users in each time period comprises: and (3) dividing the health sample vectors corresponding to different users in each time period into a training set and a test set according to a preset proportion (such as 6). Then, taking the minimum training loss as a training target, updating model parameters of a preset model by using a random gradient descent method, and outputting the current preset model as a model to be tested when the training loss is continuously reduced to be stable (namely the training loss fluctuation of a plurality of continuous training rounds is smaller than a preset difference value); taking a health sample vector corresponding to a user in each time period as an example, respectively inputting the health sample vector corresponding to the user in each time period into a multi-layer perceptron, predicting symptom infection probability of the health sample vector corresponding to each time period through the multi-layer perceptron to obtain first prediction probability corresponding to each time period, sequentially inputting the first prediction probability corresponding to each time period into a long-short term memory network according to a time sequence, fitting the first prediction probability corresponding to each time period through the long-short term memory network to obtain fitting prediction probability, taking the fitting prediction probability as the symptom infection probability of the user in the current time period, further determining training loss according to the fitting prediction probability, and determining whether the training loss meets a convergence condition (the convergence condition is that the training loss is minimum); and then, carrying out test training on the model to be tested by adopting a test set based on an early stop method, when the test loss is reduced but the model is not rebounded, determining that the performance of the model to be tested meets the preset performance condition, and outputting the model to be tested as a target prediction model. In other embodiments, the target prediction model may also be trained in other manners through conventional training, which is not described herein.

In this embodiment, the loss function when the target prediction model is obtained by training is as follows:

wherein, loss represents the Loss of eye training; y is _s,t+1 Represents the actual symptom infection result of the s < th > user in t +1 time periods, and when the symptom infection result is to infect the symptom, y _s,t+1 Is 1, when the symptom infection results in non-infection of the symptom, y _s,t+1 Is 0;

representing the predicted value of the preset model for the s-th user in t +1 time periods (the predicted value is the predicted probability of the user infecting the symptom); n denotes the total number of users and t denotes the total number of time segments for different time segments.

In the embodiment, the preset model comprises a multilayer perceptron (MLP) and a long-short term memory network (LSTM), in the model training process, one MLP is used for excavating the potential relation between the user health index conditions and the symptom infection probability in the previous t time periods, the simple MLP model avoids the occurrence of overfitting to a certain extent, and then the LSTM is used for fitting the influence of the symptom infection probability in the previous t time periods on the infection probability in the t +1 th week, so that the prediction is completed.

In an embodiment, as shown in fig. 3, after step S40, that is, after the first predicted values corresponding to a plurality of historical time periods are fitted through the long-short term memory network, and the target predicted value is obtained by prediction, the method specifically includes the following steps:

s61: when the health index data of the target user in any historical time period is monitored to change, whether parameter updating needs to be carried out on the target prediction model or not is determined according to the change condition of the health index data.

The health index data change comprises two conditions, one is that one or more health index values of the health index data in a certain historical time period of the user are wrong, the health index data in the historical time period are corrected according to an actual monitoring value, the health index data in the historical time period are caused, and at the moment, the health index data in one or more historical time periods may be changed; another is that along with research and technical development, health indicators corresponding to different symptoms may change, that is, the health indicators of the symptoms change, at this time, health indicator data of all historical time periods need to be updated, one or more original health indicator values in the health indicator data of all historical time periods are replaced with new health indicator values, and at this time, the health indicator data of all historical time periods change.

It should be understood that, with research and technical development, health indicators corresponding to different symptoms may change, and corresponding symptom infection results (whether symptoms are infected) may also change, and when health indicators change to cause changes in health indicator data, or health indicator data of a target user is incorrect and needs to be corrected, that is, when a data backfill phenomenon occurs, a traditional prediction model needs to be supported by a great deal of effort to correct a target predicted value by updated and corrected user health indicator data (backfill data for short).

In this embodiment, after the first predicted values corresponding to a plurality of historical time periods are fitted through the long and short term memory network and the target predicted values are obtained through prediction, the health conditions of the target user in the plurality of historical time periods need to be monitored in real time, including monitoring the change conditions of the health index data of each historical time period and the update conditions of the symptom infection results of the user in each historical time period, and then whether the target predicted values need to be corrected or not is determined according to the monitoring results, and whether the parameters of the target prediction model need to be updated or not is determined. When the health index data is not monitored to change and the symptom infection result is monitored to be updated, the predicted result of the target prediction and the model parameters are not influenced, so that the target predicted value does not need to be corrected, and the parameters of the target prediction model do not need to be updated.

When the health index data of the target user in any historical time period is monitored to be changed or the symptom infection result of the target user in any historical time period is monitored to be updated, the changed health index data and the symptom infection result both affect the target prediction value, but the symptom infection result does not affect the model parameter of the target prediction model, and the changed health index data does not necessarily affect the model parameter of the target prediction model, so that whether the parameter update of the target prediction model is needed or not needs to be determined further according to the change condition of the health index data.

Determining whether parameter updating needs to be carried out on the target prediction model according to the change condition of the health index data, wherein the parameter updating comprises the following steps: when the health index data of the target user in any historical time period is monitored to change, determining whether the number of the changed data in the health index data of the multiple historical time periods is larger than a preset number, wherein the preset number is an integer smaller than 4; if the number of the changed data in the plurality of health index data is larger than the preset number, determining that parameter updating needs to be carried out on the target prediction model; and if the number of the changed data in the plurality of health index data is less than or equal to the preset number, determining that parameter updating of the target prediction model is not needed. When the number of changed data in the health index data of the plurality of historical time periods is less than or equal to the preset number, that is, the health index data not exceeding the preset number (the preset number may be 1, 2 or 3) in the health index data of the plurality of historical time periods is changed, as the single or small number of health index feature changes and no method is provided for updating the target prediction model, effective mineable information is provided, so that the parameters of the target prediction model do not need to be updated, and only the target prediction value needs to be corrected according to the changed health index data; when the number of the changed data in the health index data of the plurality of historical time periods is greater than the preset number, that is, the health index data of the plurality of historical time periods with the number exceeding the preset number (for example, 1) is changed, the relationship between the health index characteristic and the target variable (that is, the symptom infection probability) may have been significantly changed, and sufficient information may be provided for parameter update of the target prediction model, so that the parameter update of the target prediction model is required.

In other embodiments, it may also be determined whether parameter updating needs to be performed on the target prediction model according to a change condition of the health index data in other manners, for example, it is determined whether the health index data changes, and if the health index changes, the health index data in all historical time periods may change, and at this time, a relationship between a health index feature and a target variable (i.e., a symptom infection probability) has significantly changed, so that sufficient information may be provided for parameter updating of the target prediction model, and thus parameter updating needs to be performed on the target prediction model; if the health index is not changed, only the transmission errors of the health index data in certain historical time periods are corrected, a small amount of health index characteristic changes and no method is provided for effective mineable information for updating the target prediction model, so that the parameters of the target prediction model do not need to be updated.

S62: and if the parameters of the target prediction model need to be updated, recording the changed health index data as updated health index data, and updating the parameters of the target prediction model by using the updated health index data to obtain an updated prediction model.

After determining whether the parameter of the target prediction model needs to be updated according to the change condition of the health index data, if determining that the parameter of the target prediction model needs to be updated, recording the changed health index data as updated health index data, and performing parameter update on the target prediction model by using the updated health index data to obtain an updated prediction model (namely the updated target prediction model).

S63: and predicting the updated health index data and the unchanged health index data by using an updated prediction model to obtain an updated target prediction value.

And after parameter updating is carried out on the target prediction model by utilizing the updated health index data to obtain an updated prediction model, the updated health index data and the unchanged health index data are predicted by adopting the updated prediction model to obtain an updated target prediction value. If the health index changes, the health index data of all historical time periods can be changed, all the historical time periods correspond to updated health index numbers, vector conversion is only needed to be performed on all the updated health index numbers corresponding to the historical time periods to obtain updated health index vectors corresponding to all the historical time periods, and then the updated health index vectors corresponding to all the historical time periods are respectively input into an updated prediction model to be predicted to obtain updated target prediction values; if the health index is not changed, but the sending errors of the health index data of a certain historical time period are corrected, and the health index data which is not changed exists, vector conversion is carried out on the updated health index number to obtain an updated health index vector corresponding to the historical time period, and then the updated health index vector and the health index vector corresponding to the health index data which is not changed are respectively input into an updated prediction model for prediction, so that an updated target prediction value is obtained. The processing procedure of the model for each vector is as described above, and is not described herein again.

In the embodiment, the target prediction model consists of a plurality of layers of perceptrons and a long-short term memory network, wherein the plurality of layers of perceptrons are used for processing health index data, the long-short term memory network is used for fitting the predicted values of the plurality of layers of perceptrons to different historical time periods, and the target prediction model has a simple structure, so that when a data backfill phenomenon exists, the re-prediction of real-time updated and corrected data can be efficiently and accurately realized without the support of a heavy calculation force, so that the target predicted value is corrected, the accuracy of a prediction result is improved, and accurate and effective assistance is provided for subsequent disease prediction and treatment.

In this embodiment, the situations of health index feature backfill (health condition backfill) and target variable backfill (symptom infection probability backfill) are considered at the same time, all data backfill problems which may occur in reality are actually considered and covered, and different predicted value correction measures are provided for different backfills: for data updated by a target user in real time, two conditions of health index data change (namely health index characteristic backfill) and symptom infection result update (namely disease condition backfill) are considered at the same time, if the health index data change, whether parameters of a target prediction model need to be updated or not is determined according to the actual change condition of the health index data, and then the updated data is corrected and predicted to obtain a more accurate symptom infection probability prediction value in the current time period; if the symptom infection result (namely the morbidity situation) changes, only the value of the symptom infection probability in the corresponding historical time period is changed, and the value and other unchanged symptom infection probability predicted values are input into the LSTM to be fitted to obtain an updated target predicted value, so that the prediction correction is completed. By monitoring the health condition data of the target user changing in real time, the target predicted value can be corrected according to different data change conditions, and therefore the accuracy of the target predicted value is improved.

In the embodiment, the first predicted values corresponding to a plurality of historical time periods are fitted through the long-short term memory network, after the target predicted value is obtained through prediction, when the change of the health index data of the target user in any historical time period is monitored, whether the parameter updating of the target prediction model is needed or not is determined according to the change condition of the health index data; if the target prediction model needs to be subjected to parameter updating, recording the changed health index data as updated health index data, and performing parameter updating on the target prediction model by using the updated health index data to obtain an updated prediction model; the updated health index data and the unchanged health index data are respectively predicted by adopting the updated prediction model to obtain an updated target predicted value, the target predicted value can be corrected according to different data change conditions, and the accuracy of the target predicted value is improved.

In an embodiment, as shown in fig. 4, in step S62, performing parameter update on the target prediction model by using the updated health index data to obtain an updated prediction model, specifically, the method includes the following steps:

s621: keeping the parameters of the long-term and short-term memory network constant, and updating the parameters of the multilayer perceptron according to the updated health index data to obtain updated parameters of the multilayer perceptron;

s622: and updating the parameters of the multilayer perceptron in the target prediction model into updated parameters to obtain an updated prediction model.

When the target prediction model is determined to need to be subjected to parameter updating according to the change condition of the health index data, the parameters of the long-term and short-term memory network are kept constant, the parameters of the multilayer perceptron are subjected to parameter updating according to the updated health index data to obtain the updated parameters of the multilayer perceptron, and then the parameters of the multilayer perceptron in the target prediction model are updated to the updated parameters to obtain the updated prediction model.

It should be understood that, because the prediction of the relationship prediction part of the symptom infection probability sequence (the sequence formed by the first predicted values corresponding to different historical time periods), that is, the long-term and short-term memory network LSTM part, is not affected by the change of the health index data, when the target prediction model is updated, the parameters of the long-term and short-term memory network can be kept constant, and do not participate in parameter updating, the parameters of the multi-layer perceptron MLP are updated only by using the updated health index data, the requirement on calculation power is reduced as much as possible without sacrificing the prediction accuracy, after the updated prediction model is obtained by updating, the updated health index data is input into the updated prediction model, and the infection probability predicted value (i.e., the first predicted value) corresponding to the historical time periods is obtained by predicting by the multi-layer perceptron MLP, so that the first predicted value corresponding to other health index data which does not change forms the corrected infection probability prediction sequence, and then the corrected infection probability predicted value of the current time period, that is obtained by inputting the corrected infection probability prediction sequence into the LSTM, that is the updated target predicted value.

In the embodiment, parameters of the long-term and short-term memory network are kept constant, the parameters of the multilayer perceptron are updated according to the updated health index data to obtain updated parameters of the multilayer perceptron, then the parameters of the multilayer perceptron in the target prediction model are updated to the updated parameters to obtain an updated prediction model, the step of updating the parameters of the target prediction model by using the updated health index data to obtain the updated prediction model is detailed, when the parameters of the model are updated, all the parameters are not required to be updated, only the parameter part of the MLP of the multilayer perceptron is updated, on the basis of guaranteeing the prediction precision, the calculation force requirement of model updating is reduced, namely when a plurality of health index data characteristics are backfilled, only part of the parameters of the model are required to be trained on the original basis, and efficient correction is realized while the original excavated characteristic information is utilized.

In an embodiment, as shown in fig. 5, step S621 is to update parameters of the multi-layer sensor according to the updated health index data to obtain updated parameters of the multi-layer sensor, and specifically includes the following steps:

s6211: respectively inputting the updated health index data and the unchanged health index data into a multilayer perceptron to obtain a plurality of corresponding perception predicted values;

s6212: inputting the corresponding updated first predicted values into the long-term and short-term memory network in sequence according to the time sequence for fitting to obtain fitting predicted values;

s6213: and determining a loss function based on the fitting predicted value, and updating parameters of the multilayer perceptron by adopting a back propagation algorithm until the parameters of the multilayer perceptron are recorded and updated when the loss function is minimum.

When the condition that the parameters of the target prediction model need to be updated is determined according to the change condition of the health index data, the parameters of the long-short term memory network LSTM are kept fixed, the updated health index data and the unchanged health index data are respectively input into the multilayer perceptron to obtain a plurality of corresponding perception predicted values, then the corresponding updated first predicted values are sequentially input into the long-short term memory network according to the time sequence to be fitted to obtain fitted predicted values, then a loss function is determined based on the fitted predicted values, the parameters of the multilayer perceptron are updated by adopting a back propagation algorithm, and the parameters of the multilayer perceptron are recorded and updated until the loss function is minimum.

In the embodiment, the updated health index data and the unchanged health index data are respectively input into the multilayer perceptron to obtain a plurality of corresponding perception predicted values, then the corresponding updated first predicted values are sequentially input into the long-term and short-term memory network according to the time sequence for fitting to obtain the fitting predicted values, the loss function is determined based on the fitting predicted values, the parameters of the multilayer perceptron are updated by adopting a back propagation algorithm, the parameters of the multilayer perceptron are recorded and updated until the loss function is minimum, the specific process of updating the parameters of the multilayer perceptron according to the updated health index data to obtain the updated parameters of the multilayer perceptron is determined, the parameters are updated by the back propagation algorithm, and the parameters of the multilayer perceptron can be quickly updated on the basis of ensuring the updating precision.

In other embodiments, when the parameters of the multi-layer perceptron are updated, the parameters of the multi-layer perceptron in the target prediction model can be retrained in a traditional model training mode until the multi-layer perceptron converges, and the parameters during convergence are output as the updated parameters of the multi-layer perceptron.

In an embodiment, as shown in fig. 6, after step S61, that is, after determining whether parameter updating needs to be performed on the target prediction model according to a change condition of the health indicator data, the method specifically includes the following steps:

s64: if the parameters of the target prediction model do not need to be updated, vector conversion is carried out on the changed health index data to obtain an updated health index vector;

s65: inputting the updated health index vector into a multilayer perceptron for prediction to obtain a first predicted value corresponding to the updated health index vector;

s66: and sequentially inputting the first predicted value corresponding to the updated health index vector and the first predicted value corresponding to the unchanged health index vector into the long-term and short-term memory network according to the time sequence for fitting prediction to obtain an updated target predicted value.

After determining whether parameter updating is needed to be carried out on the target prediction model according to the change condition of the health index data, if parameter updating is not needed to be carried out on the target prediction model, the changed health index data is less, the parameters of the target prediction model are not influenced or are slightly influenced, the secondary parameters of the target prediction model are not updated at the moment, and the target prediction value can be corrected directly according to the changed health index data. Wherein, the correction process includes: performing vector conversion on the changed health index data to obtain an updated health index vector, and inputting the updated health index vector into a multilayer perceptron for prediction to obtain a first predicted value corresponding to the updated health index vector (namely the first predicted value of the updated health index vector corresponding to the historical time period); and sequentially inputting the first predicted value corresponding to the updated health index vector and the first predicted value corresponding to the health index vector which does not change (namely the first predicted value corresponding to the historical time period which does not change) into the long-short term memory network according to the time sequence for fitting prediction to obtain the updated target predicted value.

For example, if the health index data of the ith user in the ith historical time period changes, the corresponding first predicted value w _si (indicating the first predicted value for the ith user during the ith historical period) is changed, at which time w is changed _si Is bonded to be w' _si (indicating the first predicted value of the updated health indicator vector corresponding to the ith user in the ith historical time period), so that only the input of the MLP needs to be provided with the original first predicted value sequence (w) _s1 ,w _s2 ,w _s3 ,...w _si ,...,w _st ) Modified as (w) _s1 ,w _s2 ,w _s3 ,...w′ _si ,...,w _st ) Then, prediction is performed through MLP + LSTM, and a corrected prediction result, that is, an updated target prediction value, can be obtained.

In this embodiment, after determining whether parameter updating is required to be performed on the target prediction model according to the change condition of the health index data, if parameter updating is not required to be performed on the target prediction model, vector conversion is performed on the changed health index data to obtain an updated health index vector; inputting the updated health index vector into a multilayer perceptron for prediction to obtain a first predicted value corresponding to the updated health index vector; sequentially inputting a first predicted value corresponding to the updated health index vector and a first predicted value corresponding to the unchanged health index vector into a long-term and short-term memory network according to a time sequence for fitting prediction to obtain an updated target predicted value, performing vector conversion on the updated health index number to obtain an updated health index vector corresponding to the historical time period, and then respectively inputting the updated health index vector and the health index vector corresponding to the unchanged health index data into an updated prediction model for prediction to obtain an updated target predicted value; when data of a target user are updated and refilled, the target prediction model does not need to be updated, the updated target prediction model is used for correcting the target prediction value, when a small amount of health index data are refilled, the corrected target prediction value can be obtained only by changing the input condition of the target prediction model, model parameter updating is not needed on the basis of ensuring that the target prediction value is corrected, data processing amount is reduced, and the calculation force support of model updating is reduced.

In an embodiment, as shown in fig. 7, after step S40, that is, after the first predicted values corresponding to a plurality of historical time periods are fitted through the long-short term memory network, and the target predicted value is predicted, the method specifically includes the following steps:

s71: and when the condition infection result of the target user in any historical time period is monitored to be updated, updating the first predicted value corresponding to the health index vector according to the condition infection result of the historical time period.

S72: and sequentially inputting the updated first predicted value and other non-updated first predicted values into the long-term and short-term memory network according to the time sequence for fitting prediction to obtain an updated target predicted value.

After the first predicted values corresponding to a plurality of historical time periods are fitted through the long-short term memory network and the target predicted values are obtained through prediction, the health conditions of the target user in the plurality of historical time periods need to be monitored in real time, when the symptom infection results (namely the morbidity) of the target user in any historical time period are monitored to be updated, the target prediction model is not affected by parameters, only the symptom infection probability value in the corresponding historical time period is changed, namely the first predicted value corresponding to the health index vector needs to be updated according to the symptom infection results in the historical time periods, then the updated first predicted value and other first predicted values which are not updated are sequentially input into the long-short term memory network according to the time sequence for fitting prediction, and the updated target predicted value is obtained.

For example, the first predicted value corresponding to the original symptom infection result of the s-th user in the i-th historical time period is

After the symptom infection result branch line is updated, the corresponding first predicted value of the ith user in the ith historical time period is corrected to be y' _si Then only the LSTM input is required

Is modified into

And fitting through the LSTM to obtain a corrected prediction result, wherein t is the total number of the plurality of historical times.

In this embodiment, the situations of health index feature backfill (health condition backfill) and target variable backfill (symptom infection probability backfill) are considered at the same time, all data backfill problems which may occur in reality are actually considered and covered, and different predicted value correction measures are provided for different backfills: for data updated by a target user in real time, two conditions of health index data change (namely health index characteristic backfill) and symptom infection result update (namely disease condition backfill) are considered at the same time, if the health index data change, parameters of LSTM in a target prediction model are fixed, and parameter parts of MLP are updated, so that the updated data are corrected and predicted to obtain a more accurate symptom infection probability prediction value in the current time period; if the symptom infection result (namely the morbidity situation) changes, only the value of the symptom infection probability in the corresponding historical time period is changed, and the value of the symptom infection probability is input into the LSTM to be fitted with the predicted value of the symptom infection probability which is not changed to obtain an updated target predicted value, so that the prediction correction is completed. The target predicted value can be corrected according to different data change conditions by monitoring the health condition data of the target user changing in real time, so that the accuracy of the target predicted value is improved.

In the embodiment, after the first predicted values corresponding to a plurality of historical time periods are fitted through the long-short term memory network and are predicted to obtain the target predicted values, when the symptom infection results of the target users in any historical time period are monitored to be updated, the first predicted values corresponding to the health index vectors are updated according to the symptom infection results of the historical time periods, the updated first predicted values and other non-updated first predicted values are sequentially input into the long-short term memory network according to the time sequence to be fitted and predicted to obtain the updated target predicted values, the target predicted values are corrected by the updated target predicted models without repeatedly updating the target predicted models, when the target variables (namely the symptom infection results) are updated and refilled, the corrected target predicted values can be obtained only by changing the input condition of the LSTM in the target predicted models, and on the basis of guaranteeing the accuracy of the target predicted values, key model updating is not needed, and data processing is reduced.

It should be understood that, the sequence numbers of the steps in the foregoing embodiments do not imply an execution sequence, and the execution sequence of each process should be determined by functions and internal logic of the process, and should not limit the implementation process of the embodiments of the present invention in any way.

In one embodiment, an artificial intelligence based data prediction apparatus is provided, and the artificial intelligence based data prediction apparatus corresponds to the artificial intelligence based data prediction method in the foregoing embodiment one to one. As shown in fig. 8, the artificial intelligence based data prediction apparatus includes an acquisition module 801, a conversion module 802, a first prediction module 803, a second prediction module 804, and a determination module 805. The functional modules are explained in detail as follows:

an obtaining module 801, configured to obtain a target prediction model obtained through pre-training, where the target prediction model is composed of multiple layers of perceptrons and a long-term and short-term memory network arranged in sequence;

the conversion module 802 is configured to obtain health index data of a target user in multiple historical time periods, and convert each health index data into a health index vector to obtain multiple health index vectors, where each health index vector corresponds to one historical time period;

the first prediction module 803 is configured to input each health indicator vector into the multi-layer sensing machine, and perform health condition prediction of corresponding historical time periods through the multi-layer sensing machine to obtain a first prediction value corresponding to each historical time period;

the second prediction module 804 is used for sequentially inputting the first prediction values corresponding to the historical time periods into the long-short term memory network according to the time sequence, fitting the first prediction values corresponding to the multiple historical time periods through the long-short term memory network, and predicting the target prediction values;

and a determining module 805, configured to determine a health condition of the target user in the current time period according to the target predicted value.

Further, the artificial intelligence-based data prediction apparatus includes a monitoring module 806, an updating module 807 and a third prediction module 808, and after fitting first prediction values corresponding to a plurality of historical time periods through a long-short term memory network and predicting to obtain target prediction values, the apparatus further includes:

the monitoring module 806 is configured to determine whether parameter updating is required to be performed on the target prediction model according to a change condition of the health index data when it is monitored that the health index data of the target user in any historical time period changes;

an updating module 807, configured to record the changed health index data as updated health index data if the parameter of the target prediction model needs to be updated, and perform parameter updating on the target prediction model by using the updated health index data to obtain an updated prediction model;

and the third prediction module 808 is configured to predict the updated health index data and the unchanged health index data by using the updated prediction model, so as to obtain an updated target prediction value.

Further, the updating module 807 is specifically configured to:

keeping the parameters of the long-term and short-term memory network constant, and updating the parameters of the multilayer perceptron according to the updated health index data to obtain updated parameters of the multilayer perceptron;

and updating the parameters of the multilayer perceptron in the target prediction model into updated parameters to obtain an updated prediction model.

Further, the updating module 807 is specifically configured to:

respectively inputting the updated health index data and the unchanged health index data into a multilayer perceptron to obtain a plurality of corresponding perception predicted values;

inputting the corresponding updated first predicted values into the long-term and short-term memory network in sequence according to the time sequence for fitting to obtain fitting predicted values;

and determining a loss function based on the fitting predicted value, and updating parameters of the multilayer perceptron by adopting a back propagation algorithm until the parameters of the multilayer perceptron are recorded and updated when the loss function is minimum.

Further, after determining whether parameter update of the target prediction model is needed according to a change condition of the health indicator data, the third prediction module 808 is further configured to:

if the parameters of the target prediction model do not need to be updated, vector conversion is carried out on the changed health index data to obtain an updated health index vector;

inputting the updated health index vector into a multilayer perceptron for prediction to obtain a first predicted value corresponding to the updated health index vector;

and sequentially inputting the first predicted value corresponding to the updated health index vector and the first predicted value corresponding to the unchanged health index vector into the long-term and short-term memory network according to the time sequence for fitting prediction to obtain an updated target predicted value.

Further, after the first predicted values corresponding to the multiple historical time periods are fitted through the long-short term memory network and the target predicted values are obtained through prediction, the third prediction module 808 is further configured to:

when the condition infection result of the target user in any historical time period is monitored to be updated, updating a first predicted value corresponding to the historical time period according to the condition infection result of the historical time period;

and sequentially inputting the updated first predicted value and other non-updated first predicted values into the long-term and short-term memory network according to the time sequence for fitting prediction to obtain an updated target predicted value.

For specific limitations of the artificial intelligence based data prediction apparatus, reference may be made to the above limitations of the artificial intelligence based data prediction method, which will not be described herein again. The modules in the artificial intelligence based data prediction device can be wholly or partially realized by software, hardware and a combination thereof. The modules can be embedded in a hardware form or independent from a processor in the computer device, and can also be stored in a memory in the computer device in a software form, so that the processor can call and execute operations corresponding to the modules.

In one embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as shown in fig. 9. The computer device includes a processor, a memory, a network interface, and a database connected by a system bus. Wherein the processor of the computer device is configured to provide computing and control capabilities. The memory of the computer device comprises a nonvolatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the operation of an operating system and computer programs in the non-volatile storage medium. The database of the computer device is used for storing data generated and used by the data prediction method based on artificial intelligence. The network interface of the computer device is used for communicating with an external terminal through a network connection. The computer program is executed by a processor to implement an artificial intelligence based data prediction method.

In one embodiment, there is provided a computer device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, the processor implementing the following steps when executing the computer program:

the method comprises the steps of obtaining health index data of a target user in multiple historical time periods, converting each piece of health index data into a health index vector to obtain multiple health index vectors, wherein each health index vector corresponds to one historical time period;

In one embodiment, a computer-readable storage medium is provided, having a computer program stored thereon, which when executed by a processor, performs the steps of:

It will be understood by those skilled in the art that all or part of the processes of the methods of the embodiments described above may be implemented by hardware instructions of a computer program, which may be stored in a non-volatile computer-readable storage medium, and when executed, may include the processes of the embodiments of the methods described above. Any reference to memory, storage, database, or other medium used in the embodiments provided herein may include non-volatile and/or volatile memory, among others. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically Programmable ROM (EPROM), electrically Erasable Programmable ROM (EEPROM), or flash memory. Volatile memory can include Random Access Memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms such as Static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double Data Rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous Link DRAM (SLDRAM), rambus (Rambus) direct RAM (RDRAM), direct Rambus Dynamic RAM (DRDRAM), and Rambus Dynamic RAM (RDRAM), among others.

It will be apparent to those skilled in the art that, for convenience and brevity of description, only the above-mentioned division of the functional units and modules is illustrated, and in practical applications, the above-mentioned function distribution may be performed by different functional units and modules according to needs, that is, the internal structure of the apparatus is divided into different functional units or modules to perform all or part of the above-mentioned functions.

The above-mentioned embodiments are only used to illustrate the technical solution of the present invention, and not to limit the same; although the present invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some technical features may be equivalently replaced; such modifications and substitutions do not substantially depart from the spirit and scope of the embodiments of the present invention, and are intended to be included within the scope of the present invention.

Claims

1. A data prediction method based on artificial intelligence is characterized by comprising the following steps:

acquiring health index data of a target user in a plurality of historical time periods, converting each piece of health index data into a health index vector to obtain a plurality of health index vectors, wherein each health index vector corresponds to one historical time period;

inputting each health index vector into the multilayer perceptron, and predicting the health condition of the corresponding historical time period through the multilayer perceptron to obtain a first predicted value corresponding to each historical time period;

sequentially inputting the first predicted value corresponding to each historical time period into the long-short term memory network according to the time sequence, fitting the first predicted values corresponding to the multiple historical time periods through the long-short term memory network, and predicting to obtain a target predicted value;

2. The artificial intelligence based data prediction method of claim 1, wherein the fitting of the first predicted values corresponding to a plurality of the historical time periods through the long-short term memory network, after a target predicted value is predicted, further comprises:

when the change of the health index data of the target user in any historical time period is monitored, determining whether the parameter updating of the target prediction model is needed or not according to the change condition of the health index data;

if the target prediction model needs to be subjected to parameter updating, recording the changed health index data as updated health index data, and performing parameter updating on the target prediction model by using the updated health index data to obtain an updated prediction model;

and predicting the updated health index data and the unchanged health index data by adopting the updated prediction model to obtain the updated target prediction value.

3. The artificial intelligence based data prediction method of claim 2, wherein the parameter updating the target prediction model using the updated health index data to obtain an updated prediction model comprises:

keeping the parameters of the long and short term memory network fixed, and updating the parameters of the multilayer perceptron according to the updated health index data to obtain updated parameters of the multilayer perceptron;

and updating the multilayer perceptron parameters in the target prediction model into the updated parameters to obtain the updated prediction model.

4. The artificial intelligence based data prediction method of claim 3, wherein the updating the parameters of the multi-layer perceptron according to the updated health index data to obtain the updated parameters of the multi-layer perceptron comprises:

respectively inputting the updated health index data and the unchanged health index data into the multilayer perceptron to obtain a plurality of corresponding perception predicted values;

sequentially inputting the corresponding updated first predicted values into the long-term and short-term memory network according to a time sequence for fitting to obtain fitting predicted values;

and determining a loss function based on the fitting predicted value, and updating the parameters of the multilayer perceptron by adopting a back propagation algorithm until the parameters of the multilayer perceptron are recorded as the updated parameters when the loss function is minimum.

5. The artificial intelligence based data prediction method of claim 2, wherein after determining whether parameter updates to the target prediction model are needed according to the changes in the health indicator data, the method further comprises:

if the parameters of the target prediction model do not need to be updated, performing vector conversion on the changed health index data to obtain an updated health index vector;

inputting the updated health index vector into the multilayer perceptron to predict to obtain a first predicted value corresponding to the updated health index vector;

and sequentially inputting the first predicted value corresponding to the updated health index vector and the first predicted value corresponding to the health index vector which does not change into the long-short term memory network according to a time sequence for fitting prediction to obtain the updated target predicted value.

6. The artificial intelligence based data prediction method of any one of claims 1-5, wherein the fitting of the first predicted values corresponding to a plurality of the historical time periods through the long-short term memory network, after a target predicted value is predicted, the method further comprises:

when the condition infection result of the target user in any historical time period is monitored to be updated, updating the first predicted value corresponding to the historical time period according to the condition infection result of the historical time period;

and sequentially inputting the updated first predicted value and other first predicted values which are not updated into the long-short term memory network according to a time sequence for fitting prediction to obtain the updated target predicted value.

7. An artificial intelligence-based data prediction apparatus, comprising:

the system comprises an acquisition module, a storage module and a processing module, wherein the acquisition module is used for acquiring a target prediction model obtained by pre-training, and the target prediction model consists of a plurality of layers of perceptrons and a long-term and short-term memory network which are sequentially arranged;

the conversion module is used for acquiring health index data of a target user in a plurality of historical time periods, converting each piece of health index data into a health index vector to obtain a plurality of health index vectors, wherein each health index vector corresponds to one historical time period;

the first prediction module is used for inputting each health index vector into the multilayer perceptron, and performing health condition prediction corresponding to the historical time periods through the multilayer perceptron to obtain a first prediction value corresponding to each historical time period;

the second prediction module is used for sequentially inputting the first prediction values corresponding to the historical time periods into the long-short term memory network according to the time sequence, fitting the first prediction values corresponding to the historical time periods through the long-short term memory network and predicting target prediction values;

8. The artificial intelligence based data prediction apparatus according to claim 7, wherein the fitting, by the long-short term memory network, the first predicted values corresponding to a plurality of the historical time periods, and after a target predicted value is predicted, further comprises:

the monitoring module is used for determining whether parameter updating needs to be carried out on the target prediction model according to the change condition of the health index data when the change of the health index data of the target user in any historical time period is monitored;

the updating module is used for recording the changed health index data as updated health index data if the parameter of the target prediction model needs to be updated, and updating the parameter of the target prediction model by using the updated health index data to obtain an updated prediction model;

and the third prediction module is used for predicting the updated health index data and the unchanged health index data by adopting the updated prediction model to obtain the updated target prediction value.

9. A computer device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, characterized in that the processor implements the steps of the artificial intelligence based data prediction method according to any one of claims 1 to 6 when executing the computer program.

10. A computer-readable storage medium, in which a computer program is stored which, when being executed by a processor, carries out the steps of the artificial intelligence based data prediction method according to any one of claims 1 to 6.