WO2021174881A1 - Multi-dimensional information combination prediction method, apparatus, computer device, and medium - Google Patents

Abstract

Provided are a multi-dimensional information combination prediction method, apparatus, computer device, and medium, relating to the technical field of artificial intelligence. Comprised are: receiving a prediction request sent by a user terminal (S101); extracting a time series feature corresponding to time series information in the prediction request, and invoking a preset time series model to calculate the time series feature to obtain a time series prediction result (S104); extracting non-time-series features corresponding to the non-time-series information in the prediction request, and invoking a preset non-time-series model to calculate the non-time-series feature to obtain a non-time-series prediction result (S106); calculating the time series prediction result and the non-time-series prediction result by means of a preset prediction model to obtain prediction information (S107). The invention also relates to blockchain technology; information can be stored in blockchain nodes. The invention achieves multi-dimensional prediction of prediction requests, improving the comprehensiveness of the prediction reference dimensions, and thus increasing the accuracy of prediction.

Description

Multi-dimensional information combination prediction method, device, computer equipment and medium

This application claims the priority of a Chinese patent application filed on September 4, 2020, with the application number CN 202010920980.2 and titled "Multi-dimensional information combination prediction method, device, computer equipment and medium", the entire content of which is incorporated by reference In this application.

Technical field

This application relates to the field of artificial intelligence intelligent decision-making technology, and relates to neural network technology in the field of intelligent decision-making technology, and in particular to a combination prediction method, device, computer equipment and medium of multi-dimensional information.

Background technique

A predictive model refers to a computer model that predicts the future behavior or state of an object based on the quantitative relationship between things described in mathematical language or formulas. It reveals the inherent regularity of things to a certain extent, and it is used when predicting. As a direct basis for calculating the predicted value.

Current prediction models usually use single-dimensional data to predict the target object. Generally, time series information (such as dynamic information) is used to predict the target object in a time series dimension (can be regarded as a dynamic dimension), or non-time series information is used. (Such as static information) predict the target object in a non-time-series dimension (which can be regarded as a static dimension). The inventor realizes that the method of using a single dimension to predict the target object cannot comprehensively consider the target object, resulting in low prediction accuracy.

Summary of the invention

The purpose of this application is to provide a combination prediction method, device, computer equipment and medium of multi-dimensional information, which are used to solve the problem of low prediction accuracy due to the inability to comprehensively consider the target object in the prior art; this application It can be applied to smart medical scenarios to promote the construction of smart cities.

In order to achieve the above objective, this application provides a combined prediction method of multi-dimensional information, including:

Receive the prediction request sent by the client;

Extracting the timing feature corresponding to the timing information in the prediction request, calling a preset timing model to calculate the timing feature to obtain a timing prediction result;

Extracting the non-time-series features corresponding to the non-time-series information in the prediction request, and call a preset non-time-series model to calculate the non-time-series features to obtain a non-time-series prediction result;

The prediction information is obtained by calculating the time series prediction result and the non-time series prediction result through a preset prediction model.

In order to achieve the above objective, this application also provides a multi-dimensional information combined prediction device, including:

The input module is used to receive the prediction request sent by the user terminal;

The time series prediction module is used to extract the time series feature corresponding to the time series information in the prediction request, call a preset time series model to calculate the time series feature to obtain a time series prediction result;

The non-time-series prediction module is configured to extract the non-time-series features corresponding to the non-time-series information in the prediction request, and call a preset non-time-series model to calculate the non-time-series features to obtain a non-time-series prediction result;

The comprehensive prediction module is used to calculate the time series prediction result and the non-time series prediction result through a preset prediction model to obtain prediction information.

In order to achieve the above objective, this application also provides a computer device, which includes a memory, a processor, and a computer program stored in the memory and running on the processor. The processor of the computer device executes the computer program. The combined prediction method of multi-dimensional information, the combined prediction method of multi-dimensional information includes:

Receive the prediction request sent by the client;

Extracting the timing feature corresponding to the timing information in the prediction request, calling a preset timing model to calculate the timing feature to obtain a timing prediction result;

Extracting the non-time-series features corresponding to the non-time-series information in the prediction request, and call a preset non-time-series model to calculate the non-time-series features to obtain a non-time-series prediction result;

The prediction information is obtained by calculating the time series prediction result and the non-time series prediction result through a preset prediction model.

In order to achieve the above objective, the present application also provides a computer-readable storage medium with a computer program stored on the readable storage medium, and the computer program stored in the readable storage medium realizes multi-dimensional information when executed by a processor. The combined prediction method of multi-dimensional information includes:

Receive the prediction request sent by the client;

Extracting the timing feature corresponding to the timing information in the prediction request, calling a preset timing model to calculate the timing feature to obtain a timing prediction result;

Extracting the non-time-series features corresponding to the non-time-series information in the prediction request, and call a preset non-time-series model to calculate the non-time-series features to obtain a non-time-series prediction result;

The prediction information is obtained by calculating the time series prediction result and the non-time series prediction result through a preset prediction model.

The combined prediction method, device, computer equipment, and medium of multi-dimensional information provided in this application calculate the time series characteristics corresponding to the time series information through the time series model, so as to accurately and quickly obtain the time series prediction result of the object corresponding to the time series information in the dimension of the time series information. Calculate the non-time-series features corresponding to the non-time-series information through the non-time-series model, so as to accurately and quickly obtain the non-time-series prediction results of the objects corresponding to the non-time-series information in the dimension of the non-time-series information. The time-series forecast result obtains forecast information that takes into account both time-series information and non-time-series information, realizes multi-dimensional forecasting of forecast requests, improves the comprehensiveness of the forecast reference dimensions, and thereby improves the accuracy of the forecast.

Description of the drawings

FIG. 1 is a flowchart of Embodiment 1 of a combined prediction method for multi-dimensional information according to this application;

2 is a schematic diagram of the environmental application of the multi-dimensional information combination prediction method in Embodiment 2 of the multi-dimensional information combination prediction method of this application;

FIG. 3 is a specific method flow chart of the combined prediction method of multi-dimensional information in the second embodiment of the combined prediction method of multi-dimensional information according to the present application;

4 is a flowchart of the time sequence characterization process in the second embodiment of the combined prediction method for multi-dimensional information according to the present application;

FIG. 5 is a flowchart of non-time-series characterization processing in Embodiment 2 of the combined prediction method for multi-dimensional information of the present application;

6 is a schematic diagram of program modules of Embodiment 3 of a combined prediction apparatus for multi-dimensional information of this application;

FIG. 7 is a schematic diagram of the hardware structure of the computer device in the fourth embodiment of the computer device of this application.

Detailed ways

In order to make the purpose, technical solutions, and advantages of this application clearer and clearer, the following further describes the application in detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described here are only used to explain the application, and are not used to limit the application. Based on the embodiments in this application, all other embodiments obtained by those of ordinary skill in the art without creative work shall fall within the protection scope of this application.

The following examples are now provided:

Example one:

Please refer to FIG. 1. The combined prediction method of multi-dimensional information in this embodiment includes:

S101: Receive a prediction request sent by the user terminal;

S104: Extract the time series feature corresponding to the time series information in the prediction request, call a preset time series model to calculate the time series feature to obtain a time series prediction result;

S106: Extract the non-time-series feature corresponding to the non-time-series information in the prediction request, and call a preset non-time-series model to calculate the non-time-series feature to obtain a non-time-series prediction result;

S107: Calculate the time series prediction result and the non-time series prediction result through a preset prediction model to obtain prediction information.

In this embodiment, the prediction request has object information, time series information, and non-time series information. The object information is the type of the object corresponding to the prediction request; for example, the type may be the name of the disease obtained by the object (such as a patient), such as diabetes, heart disease, and myocardial infarction. The time sequence information is the state data accumulated over time by the object corresponding to the prediction request, and the state data reflects the behavior and shape of the object. The state data, where the behavior can be the prescription medication obtained by the patient, and the form can be the patient’s symptoms and key inspection values.

Calculate the time series characteristics corresponding to the time series information through the time series model to accurately and quickly obtain the time series prediction results of the objects corresponding to the time series information in the dimension of the time series information. In the dimension of information, the non-time-series prediction results of the objects corresponding to the non-time-series information are obtained accurately and quickly, and the time-series prediction results and non-time-series prediction results are obtained through the prediction model, and the prediction information that takes into account both the time-series information and the non-time-series information is obtained. The forecast request is forecasted in multiple dimensions, which improves the comprehensiveness of the forecast reference dimensions, and thus improves the accuracy of the forecast.

This application can be applied in smart medical scenarios to promote the construction of smart cities.

Embodiment two:

This embodiment is a specific application scenario of the foregoing Embodiment 1. Through this embodiment, the method provided by this application can be described more clearly and specifically.

Next, in a server running the combined prediction method with multi-dimensional information, the time series feature is calculated to obtain the time series prediction result, and the non-time series feature is calculated to obtain the non-time series prediction result, and then the time series prediction result and the non-time series prediction result are calculated. The prediction information obtained from the prediction result is taken as an example to illustrate the method provided in this embodiment in detail. It should be noted that this embodiment is only exemplary, and does not limit the protection scope of the embodiment of this application.

Fig. 2 schematically shows an environmental application diagram of the multi-dimensional information combination prediction method according to the second embodiment of the present application.

In an exemplary embodiment, the authentication server 2 where the multi-dimensional information combination prediction method is located is connected to the user terminal 4 through the network 3, and the server 2 may provide services through one or more networks 3, and the network 3 may include various Network equipment, such as routers, switches, multiplexers, hubs, modems, bridges, repeaters, firewalls, proxy devices and/or etc. The network 3 may include physical links, such as coaxial cable links, twisted pair cable links, optical fiber links, combinations thereof, and/or the like. The network 3 may include a wireless link, such as a cellular link, a satellite link, a Wi-Fi link, and/or the like; the user terminal 4 may be a computer device such as a smart phone, a tablet computer, a notebook computer, and a desktop computer.

FIG. 3 is a specific method flowchart of a multi-dimensional information combined prediction method provided by an embodiment of the present application. The method specifically includes steps S201 to S207.

S201: Receive a prediction request sent by a user terminal; where the prediction request has object information, time sequence information, and non-time sequence information.

The object information is the type of the object corresponding to the prediction request; for example, the type may be the name of the disease obtained by the object (such as a patient), such as diabetes, heart disease, and myocardial infarction.

The time sequence information is the state data accumulated over time by the object corresponding to the prediction request, and the state data reflects the behavior and shape of the object. The state data, among which the behavior can be the prescription medication obtained by the patient, and the shape can be the patient’s symptoms and key inspection values.

For example, suppose that the patient has five visits due to his illness, and the time sequence information given by the physician for the diagnosis of the patient is obtained five times, as shown in the following table:

Order of visits	Prescription medication	Patient symptoms	Critical inspection check value
the first time	Medicine A	Headache, nausea	Heart rate: M1 Blood pressure: N1
the second time	Medicine A, Medicine B	Headache	Heart rate: M2 Blood pressure: N2
the third time	Medicine B	Mild headache, nausea	Heart rate: M3 Blood pressure: N3
the fourth time	C medicine	Mild headache	Heart rate: M4 Blood pressure: N4

The non-chronological information is the attribute data of the object corresponding to the prediction request reflected in the prediction request; for example, the attribute data may be the basic personal information of the object (such as the patient), such as identity information, social information, Living habits, etc., where the identity information may include: gender, age, and ethnicity; the social information may include: education level, occupation; and living habits may include: smoking status.

For example: get the attribute data of Zhang San (that is, the object), as shown in the following table:

gender	age	nationality	education level	profession	Marital status	Smoking status
male	28	Chinese	Undergraduate	engineer	unmarried	do not smoke

S202: Extract the object information in the prediction request, and obtain a time series model and a non-time series model corresponding to the object information from a preset model set.

In order to ensure targeted calculations and predictions for different types of prediction requests, so as to improve the prediction accuracy of the objects corresponding to the prediction requests, this step uses the object information in the prediction request to obtain data from the preset model set. The time series model and the non-time series model corresponding to the object information, wherein the time series model is used to predict the behavior or state of the object (for example, compliance, recovery possibility, etc.) from the dimension of the time series information Time-series prediction information, where the non-time-series model is used to predict the behavior or state (for example, compliance, recovery possibility, etc.) of the object from the dimension of the non-time-series information to obtain non-time-series prediction information.

Exemplarily, if the extracted object information is diabetes, then from the model set, the time series model and the non-time series model trained through the training sample corresponding to the object information (for example, the training sample of the object information is diabetes) , Used to predict the forecast request.

S203: Perform a time sequence characterization process on the time sequence information in the prediction request to obtain a time sequence feature.

In order to facilitate the calculation of the time series information through the neural network model to obtain the required time series prediction results, this step is to perform the time series characterization process on the time series information to obtain the time series characteristics displayed in the form of vectors, so that the neural network model can learn the time series through the time series characteristics The content of the information, and predict the time series prediction results based on the content.

In a preferred embodiment, referring to FIG. 4, the sequence characterization process includes the following steps:

S31: Split the timing information according to a preset split rule to obtain at least one timing sub-information.

In this step, the splitting rule is a method of splitting information preset by the user. In this embodiment, the splitting rule is to treat the information corresponding to the same medical visit order as an independent time sequence sub-information. By splitting the timing information, based on the above example, the following timing sub-information will be obtained:

The first sequence sub-information is as follows:

Order of visits	Prescription medication	Patient symptoms	Critical inspection check value
the first time	Medicine A	Headache, nausea	Heart rate: M1 Blood pressure: N1

The second sequence sub-information is as follows:

Order of visits	Prescription medication	Patient symptoms	Critical inspection check value
the second time	Medicine A, Medicine B	Headache	Heart rate: M2 Blood pressure: N2

The third sequence sub-information is as follows:

Order of visits	Prescription medication	Patient symptoms	Critical inspection check value
the third time	Medicine B	Mild headache, nausea	Heart rate: M3 Blood pressure: N3

The fourth sequence sub-information is as follows:

Order of visits	Prescription medication	Patient symptoms	Critical inspection check value
the fourth time	C medicine	Mild headache	Heart rate: M4 Blood pressure: N4

S32: Determine whether the number of obtained timing sub-information reaches a preset timing threshold;

In this step, the timing threshold can be set by the user, for example: 5.

S33: If yes, summarize the sequence sub-information to obtain a sub-information set.

S34: If not, create zero-padded sub-information and summarize it with the timing sub-information to obtain a sub-information set, so that the number of zero-padded sub-information and timing sub-information in the sub-information set reaches the timing threshold.

In order to ensure that the obtained timing features can be calculated by the preset timing model and finally output more accurate timing prediction results, this step provides zero-padded sub-information when the timing sub-information does not reach the timing threshold to ensure that the timing model can be normal Run to ensure the accuracy of the output timing prediction results.

Exemplarily, based on the above example, the number of timing sub-information obtained is 4, and the timing threshold is 5. Therefore, it is necessary to create a zero-padded sub-information as follows:

Order of visits	Prescription medication	Patient symptoms	Critical inspection check value
0	0	0	0

S35: Extract the timing sub-information in the sub-information set, and identify the vector value corresponding to each data item in the timing sub-information through a preset timing vector table, and according to the data item in the timing sub-information Arrange the vector values corresponding to each of the data items to obtain the time sequence sub-features corresponding to the time sequence sub-information; wherein, the data item is an indivisible minimum unit in the time sequence sub-information.

In this step, the time series vector table is formulated by the user according to needs, and has a data information table with the mapping relationship between each data item and the vector value in the time series sub-information.

Exemplarily, taking the first sequence information as an example based on the above solution:

Order of visits	Prescription medication	Patient symptoms	Critical inspection check value
the first time	Medicine A	Headache, nausea	Heart rate: M1 Blood pressure: N1

If in the time series vector table, the vector value corresponding to the first time is 0.1, the vector value corresponding to drug A is 0.2, the vector value of the combination of headache and nausea is 0.3, the vector value corresponding to heart rate: M1 and blood pressure: N1 Is 0.1, then the time sequence word vector obtained is:

{0.1,0.2,0.3,0.1}

Exemplarily, taking the second sequence information as an example based on the above solution:

Order of visits	Prescription medication	Patient symptoms	Critical inspection check value
the second time	Medicine A, Medicine B	Headache	Heart rate: M2 Blood pressure: N2

The second time the corresponding vector value is 0.2, the vector value corresponding to drug A and drug B is 0.25, the vector value of the combination of headache is 0.2, and the vector value corresponding to heart rate: M2 and blood pressure: N2 is 0.15, then you get The time series word vector is:

{0.2,0.25,0.2,0.15}

S36: According to the position of the timing sub-information in the timing information, arrange the timing sub-features to obtain the timing feature corresponding to the timing information.

Exemplarily, based on the above examples, the first to fourth timing information's timing sub-features and the zero-padded sub-feature are obtained, and the following timing features are obtained:

S204: Extract the time series feature corresponding to the time series information in the prediction request, and call a preset time series model to calculate the time series feature to obtain a time series prediction result.

In order to quickly and accurately predict the time series information to obtain the corresponding time series prediction results, this step calculates the time series characteristics corresponding to the time series information through the preset time series model, so as to accurately and quickly obtain the object corresponding to the time series information in the dimension of the time series information. Time series forecast results.

In this embodiment, the LSTM (Long Short Term Memory) model is used as the time series model, and the initial LSTM model is trained through preset time series samples to obtain the time series model, and the time series samples include vectors representing the content of the time series samples. The time series feature sample and the compliance value reflecting the object corresponding to the time series sample, wherein the time series feature sample is used as the input vector of the initial LSTM model, the compliance value is used as the training target of the initial LSTM model, and the compliance value The value is expressed as a decimal number less than or equal to 1 between 0-1 to reflect the degree of compliance of the patient to the doctor. Therefore, in this step, the patient's time series information is calculated to predict the patient's compliance value, that is, the time series prediction result.

It should be noted that LSTM (Long Short-Term Memory) is a long and short-term memory network, a time recursive neural network, suitable for processing and predicting important events with relatively long intervals and delays in a time series, due to the training of the LSTM model The process and operating principle belong to the prior art, and those skilled in the art can train and use the LSTM model through the prior art. Therefore, the training process and operating principle of the LSTM model will not be described in detail here. The compliance (Patient compliance/Treatment compliance) is also called compliance, which refers to the treatment of the subject (i.e., the patient) according to the doctor's prescription and the execution of behavior consistent with the doctor's order. The compliance value refers to the subject's treatment in accordance with the regulations, And the quantified degree of performing the behavior consistent with the doctor's order, for example: 0.5, which means that the patient's treatment to the doctor, and the degree of completion of the execution of the doctor's order is 50%.

S205: Perform non-sequential characterization processing on the non-sequential information in the prediction request to obtain non-sequential features.

In order to facilitate the calculation of non-time-series information through the neural network model to obtain the required non-time-series prediction results, this step uses non-time-series characterization processing on the non-time-series information to obtain non-time-series features in the form of vectors, so that the neural network model can pass The non-sequential feature learns the content of the non-sequential information, and predicts based on the content to obtain the non-sequential prediction result.

In a preferred embodiment, referring to FIG. 5, the non-sequential characterization processing includes the following steps:

S51: Identify the vector value corresponding to each data item in the non-sequential information through a preset non-sequential vector table, where the data item is an indivisible minimum unit in the non-sequential sub-information.

In this step, the non-sequential vector table is formulated by the user according to needs, and has a data information table with the mapping relationship between each data item and the vector value in the non-sequential sub-information.

S52: According to the position of the data item in the non-sequential information, arrange the vector values corresponding to each data item and summarize to obtain the non-sequential feature.

Exemplarily, based on the above example, the non-sequential information is as follows:

gender	age	nationality	education level	profession	Marital status	Smoking status
male	28	Chinese	Undergraduate	engineer	unmarried	do not smoke

If in the non-time series vector table, the vector value corresponding to gender male is 1, the vector value corresponding to age 28 is 0.28, the vector value corresponding to ethnicity is Han, the vector value is 0, education level is undergraduate, the vector value is 0.5, occupation is corresponding to engineer The vector value is 0.03, the vector value corresponding to the marital status being unmarried is 0.1, and the vector value corresponding to the smoking status being non-smoking is 1. Then, the non-chronological characteristics obtained are as follows:

{1,0.28,0,0.5,0.03,0.1,1}

S206: Extract the non-time-series feature corresponding to the non-time-series information in the prediction request, and call a preset non-time-series model to calculate the non-time-series feature to obtain a non-time-series prediction result.

In order to achieve fast and accurate prediction of non-sequential information to obtain the corresponding non-sequential prediction results, this step calculates the non-sequential features corresponding to the non-sequential information through the preset non-sequential model, so as to accurately and quickly obtain the non-sequential information in the dimension The non-time-series prediction result of the object corresponding to the non-time-series information.

In this embodiment, a deep neural network composed of multiple fully-connected layers (FC) is used as the non-sequential model, and the deep neural network is trained through preset non-sequential samples to obtain the results. The non-chronological model;

The non-time-series samples include non-time-series feature samples that represent the content of the non-time-series samples in the form of vectors, and a compliance value reflecting the object corresponding to the non-time-series samples, wherein the non-time-series feature samples are used as the initial deep neural network model Input vector, the adherence value is used as the training target of the initial deep neural network model, and the adherence value is expressed as any decimal between 0-1 and less than or equal to 1, to reflect the degree of compliance of the patient to the doctor . Therefore, in this step, the patient's non-sequential information is calculated to predict the patient's compliance value, that is, the non-sequential prediction result.

It should be noted that fully connected layers (FC) play the role of "classifier" in the convolutional neural network technology, which is used to map the obtained "distributed feature representation" to the sample label space. . Therefore, by constructing a deep neural network composed of multiple fully-connected layers (FC) to classify the non-chronological features, it will quickly and accurately predict the patient's compliance based on the non-chronological information. The effect of performance value (ie: non-chronological prediction result).

Because the training process and operating principle of the deep neural network composed of multiple fully connected layers belong to the prior art, those skilled in the art can train and use the deep neural network through the existing technology. Therefore, the training process of the deep neural network And the operating principle will not be repeated here. The compliance (Patient compliance/Treatment compliance) is also called compliance, which refers to the treatment of the subject (i.e., the patient) according to the doctor's prescription and the execution of behavior consistent with the doctor's order. The compliance value refers to the subject's treatment in accordance with the regulations, And the quantified degree of performing the behavior consistent with the doctor's order, for example: 0.5, which means that the patient's treatment to the doctor, and the degree of completion of the execution of the doctor's order is 50%.

S207: Calculate the time series prediction result and the non-time series prediction result through a preset prediction model to obtain prediction information.

In order to ensure that accurate prediction judgments can be made on the objects corresponding to the prediction request from the two dimensions of time series information and non-time series information, this step uses the prediction model to compare the time series prediction results and non-time series prediction results, and obtains that both time series information and time series information are considered. The prediction information of non-time series information improves the accuracy of the prediction.

In a preferred embodiment, the objective function of the prediction model is a geometric mean function, as follows:

Among them, M is the prediction information, Scoe1 is the time series prediction result, and Scoe2 is the non-time series prediction result.

Since the prediction evaluation of the object is performed from two dimensions of the time series prediction result and the non-time series prediction result, not only the influence of the time series prediction result and the non-time series prediction result itself on the final prediction information needs to be considered, but also the time series prediction The difference between the result and the non-time-series forecast result affects the final forecast information, and the use of the geometric mean function just satisfies the above two influence considerations.

For example: the time series prediction result of object A is 0.4, the non-time series prediction result is 0.4; the time series prediction result of object B is 0.1, and the non-time series prediction result is 0.9; if the average method is used, the prediction information of object A is 0.4, and the object The prediction information of B is 0.5. That is to say, object B, which has a large difference between the time series prediction result and the non-time series prediction result and is unstable, is more compliant than object A, which is obviously unreasonable in reality. Using the prediction model in this step, it will be obtained that the prediction information of object A is 0.4 and the prediction information of object B is 0.3, which is obviously more in line with reality, thereby improving the accuracy of prediction.

At the same time, compared to the method of ensuring the accuracy of the final prediction information by setting the weights of the time series prediction results and the non-time series prediction results, it is not only unable to consider the impact of the large difference between the time series prediction results and the non-time series prediction results on the final prediction information, but also The setting of the weight causes the prediction model to overfit the time series prediction results and the non-time series prediction results with small differences, and the problem of underfitting the time series prediction results and the non-time series prediction results with large differences occurs.

Preferably, after calculating the time series prediction result and the non-time series prediction result to obtain the prediction information through a preset prediction model, the method includes:

Upload the prediction information to the blockchain.

It should be noted that the corresponding summary information is obtained based on the prediction information. Specifically, the summary information is obtained by hashing the prediction information, for example, obtained by using the sha256s algorithm. Uploading summary information to the blockchain can ensure its security and fairness and transparency to users. The user equipment can download the summary information from the blockchain in order to verify whether the predicted information has been tampered with. The blockchain referred to in this example is a new application mode of computer technology such as distributed data storage, point-to-point transmission, consensus mechanism, and encryption algorithm. Blockchain, essentially a decentralized database, is a series of data blocks associated with cryptographic methods. Each data block contains a batch of network transaction information for verification. The validity of the information (anti-counterfeiting) and the generation of the next block. The blockchain can include the underlying platform of the blockchain, the platform product service layer, and the application service layer.

Embodiment three:

Please refer to FIG. 6, a multi-dimensional information combination prediction apparatus 1 of this embodiment includes:

The input module 11 is used to receive the prediction request sent by the user terminal;

The time series prediction module 14 is used to extract the time series feature corresponding to the time series information in the prediction request, and call a preset time series model to calculate the time series feature to obtain a time series prediction result;

The non-time-series prediction module 16 is configured to extract the non-time-series feature corresponding to the non-time-series information in the prediction request, and call a preset non-time-series model to calculate the non-time-series feature to obtain a non-time-series prediction result;

The comprehensive prediction module 17 is configured to calculate the time series prediction result and the non-time series prediction result through a preset prediction model to obtain prediction information.

Optionally, the multi-dimensional information combination prediction device 1 further includes:

The model selection module 12 is configured to extract the object information in the prediction request, and obtain a time series model and a non-time series model corresponding to the object information from a preset model set.

Optionally, the multi-dimensional information combination prediction device 1 further includes:

The timing processing module 13 is used to perform timing characterization processing on the timing information in the prediction request to obtain timing characteristics.

Optionally, the multi-dimensional information combination prediction device 1 further includes:

The non-sequential processing module 15 is configured to perform non-sequential characterization processing on the non-sequential information in the prediction request to obtain non-sequential features.

The technical solution is applied to the field of intelligent decision-making of artificial intelligence, and it is constructed to call the time series model constructed based on the neural network to calculate the time series feature to obtain the time series prediction result, and to call the non-time series model constructed based on the neural network to calculate the non-time series feature Obtain a non-time-series prediction result, and calculate the time-series prediction result and the non-time-series prediction result through a prediction model to obtain a multi-dimensional prediction model of prediction information.

Embodiment four:

In order to achieve the above objective, this application also provides a computer device 5. The components of the multi-dimensional information combination prediction device 1 of the third embodiment can be dispersed in different computer devices. The computer device 5 can be a smart phone that executes a program, Tablet computers, notebook computers, desktop computers, rack servers, blade servers, tower servers or cabinet servers (including independent servers, or server clusters composed of multiple application servers), etc. The computer device in this embodiment at least includes but is not limited to: a memory 51 and a processor 52 that can be communicatively connected to each other through a system bus, as shown in FIG. 7. It should be pointed out that FIG. 7 only shows a computer device with components, but it should be understood that it is not required to implement all the illustrated components, and more or fewer components may be implemented instead.

In this embodiment, the memory 51 (ie, readable storage medium) includes flash memory, hard disk, multimedia card, card-type memory (for example, SD or DX memory, etc.), random access memory (RAM), static random access memory (SRAM), Read only memory (ROM), electrically erasable programmable read only memory (EEPROM), programmable read only memory (PROM), magnetic memory, magnetic disk, optical disk, etc. In some embodiments, the memory 51 may be an internal storage unit of a computer device, such as a hard disk or memory of the computer device. In other embodiments, the memory 51 may also be an external storage device of the computer device, such as a plug-in hard disk equipped on the computer device, a smart memory card (Smart Media Card, SMC), and a Secure Digital (SD). Card, Flash Card, etc. Of course, the memory 51 may also include both the internal storage unit of the computer device and its external storage device. In this embodiment, the memory 51 is generally used to store the operating system and various application software installed in the computer equipment, such as the program code of the multi-dimensional information combination prediction device of the third embodiment, and so on. In addition, the memory 51 can also be used to temporarily store various types of data that have been output or will be output.

In some embodiments, the processor 52 may be a central processing unit (Central Processing Unit, CPU), a controller, a microcontroller, a microprocessor, or other data processing chips. The processor 52 is generally used to control the overall operation of the computer equipment. In this embodiment, the processor 52 is used to run program codes or process data stored in the memory 51, for example, to run a combined prediction device for multi-dimensional information, to implement the multi-dimensional information combined prediction method of the first and second embodiments. The combined prediction method of multi-dimensional information includes:

Receive the prediction request sent by the client;

Extracting the timing feature corresponding to the timing information in the prediction request, calling a preset timing model to calculate the timing feature to obtain a timing prediction result;

Extracting the non-time-series features corresponding to the non-time-series information in the prediction request, and call a preset non-time-series model to calculate the non-time-series features to obtain a non-time-series prediction result;

The prediction information is obtained by calculating the time series prediction result and the non-time series prediction result through a preset prediction model.

Embodiment five:

To achieve the above objective, this application also provides a computer-readable storage medium. In this embodiment, the computer-readable storage medium may be volatile or non-volatile, such as flash memory, hard disk, multimedia card, Card-type memory (for example, SD or DX memory, etc.), random access memory (RAM), static random access memory (SRAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), programmable Read-only memory (PROM), magnetic memory, magnetic disks, optical disks, servers, App application malls, etc., have computer programs stored thereon, and corresponding functions are realized when the programs are executed by the processor 52. The computer-readable storage medium of this embodiment is used to store the combined prediction device of multi-dimensional information. When executed by the processor 52, the combined prediction method of multi-dimensional information of Embodiment 1 and Embodiment 2 is realized. The combination of the multi-dimensional information Forecasting methods include:

Receive the prediction request sent by the client;

Extracting the timing feature corresponding to the timing information in the prediction request, calling a preset timing model to calculate the timing feature to obtain a timing prediction result;

Extracting the non-time-series features corresponding to the non-time-series information in the prediction request, and call a preset non-time-series model to calculate the non-time-series features to obtain a non-time-series prediction result;

The prediction information is obtained by calculating the time series prediction result and the non-time series prediction result through a preset prediction model.

The serial numbers of the foregoing embodiments of the present application are for description only, and do not represent the superiority or inferiority of the embodiments.

Through the description of the above implementation manners, those skilled in the art can clearly understand that the above-mentioned embodiment method can be implemented by means of software plus the necessary general hardware platform, of course, it can also be implemented by hardware, but in many cases the former is better.的实施方式。

The above are only the preferred embodiments of the application, and do not limit the scope of the patent for this application. Any equivalent structure or equivalent process transformation made using the content of the description and drawings of the application, or directly or indirectly applied to other related technical fields , The same reason is included in the scope of patent protection of this application.

Claims (20)

1. A combined forecasting method of multi-dimensional information, including:

   Receive the prediction request sent by the client;

   Extracting the timing feature corresponding to the timing information in the prediction request, calling a preset timing model to calculate the timing feature to obtain a timing prediction result;

   Extracting the non-time-series features corresponding to the non-time-series information in the prediction request, and call a preset non-time-series model to calculate the non-time-series features to obtain a non-time-series prediction result;

   The prediction information is obtained by calculating the time series prediction result and the non-time series prediction result through a preset prediction model.
2. The combined prediction method of multi-dimensional information according to claim 1, wherein after receiving the prediction request sent by the user terminal, the method comprises:

   The object information in the prediction request is extracted, and the time series model and the non-time series model corresponding to the object information are obtained from a preset model set.
3. The method for combined prediction of multi-dimensional information according to claim 1, wherein before extracting the time sequence feature corresponding to the time sequence information in the prediction request, the method comprises:

   Performing sequence characterization processing on the sequence information in the prediction request to obtain sequence characteristics;

   Before extracting the non-time-series feature corresponding to the non-time-series information in the prediction request, the method includes:

   Non-sequential characterization processing is performed on the non-sequential information in the prediction request to obtain non-sequential features.
4. The method for combined prediction of multi-dimensional information according to claim 3, wherein the time sequence characterization processing includes the following steps:

   Splitting the timing information according to a preset splitting rule to obtain at least one timing sub-information;

   Judging whether the number of obtained timing sub-information reaches a preset timing threshold;

   If yes, sum up the time sequence sub-information to obtain a sub-information set;

   If not, create zero-padded sub-information and summarize it with the timing sub-information to obtain a sub-information set, so that the number of zero-padded sub-information and timing sub-information in the sub-information set reaches the timing threshold;

   Extract the timing sub-information in the sub-information set, and identify the vector value corresponding to each data item in the timing sub-information through a preset timing vector table, according to the position of the data item in the timing sub-information Arranging the vector values corresponding to each of the data items to obtain the time sequence sub-features corresponding to the time sequence sub-information; wherein, the data item is an indivisible minimum unit in the time sequence sub-information;

   According to the position of the timing sub-information in the timing information, the timing sub-features are arranged to obtain the timing characteristics corresponding to the timing information.
5. The combined prediction method of multi-dimensional information according to claim 3, wherein the non-time-series characterization processing includes the following steps:

   Identify the vector value corresponding to each data item in the non-sequential information through a preset non-sequential vector table, where the data item is an indivisible minimum unit in the non-sequential sub-information;

   According to the position of the data item in the non-sequential information, the vector values corresponding to each of the data items are arranged and summarized to obtain the non-sequential feature.
6. The method for combined prediction of multi-dimensional information according to claim 1, wherein an LSTM model is used as a time series model, and the time series model is obtained by training the initial LSTM model with preset time series samples;

   A deep neural network composed of multiple fully connected layers is used as the non-sequential model, and the non-sequential model is obtained by training the deep neural network through preset non-sequential samples.
7. The combined prediction method of multi-dimensional information according to claim 1, wherein the objective function of the prediction model is a geometric mean function, as follows:

   

   Among them, M is the prediction information, Scoe1 is the time series prediction result, and Scoe2 is the non-time series prediction result;

   After calculating the time series prediction result and the non-time series prediction result through the preset prediction model to obtain the prediction information, it includes:

   Upload the prediction information to the blockchain.
8. A combined prediction device for multi-dimensional information, which includes:

   The input module is used to receive the prediction request sent by the user terminal;

   The time series prediction module is used to extract the time series feature corresponding to the time series information in the prediction request, call a preset time series model to calculate the time series feature to obtain a time series prediction result;

   The non-time-series prediction module is configured to extract the non-time-series features corresponding to the non-time-series information in the prediction request, and call a preset non-time-series model to calculate the non-time-series features to obtain a non-time-series prediction result;

   The comprehensive prediction module is used to calculate the time series prediction result and the non-time series prediction result through a preset prediction model to obtain prediction information.
9. A computer device comprising a memory, a processor, and a computer program stored on the memory and capable of running on the processor, wherein the processor of the computer device implements a combination prediction method of multi-dimensional information when the computer program is executed , The combined prediction method of multi-dimensional information includes:

   Receive the prediction request sent by the client;

   Extracting the timing feature corresponding to the timing information in the prediction request, calling a preset timing model to calculate the timing feature to obtain a timing prediction result;

   Extracting the non-time-series features corresponding to the non-time-series information in the prediction request, and call a preset non-time-series model to calculate the non-time-series features to obtain a non-time-series prediction result;

   The prediction information is obtained by calculating the time series prediction result and the non-time series prediction result through a preset prediction model.
10. The computer device according to claim 9, wherein after receiving the prediction request sent by the user terminal, the method comprises:

    The object information in the prediction request is extracted, and the time series model and the non-time series model corresponding to the object information are obtained from a preset model set.
11. 9. The computer device according to claim 9, wherein before extracting the timing feature corresponding to the timing information in the prediction request, it comprises:

    Performing sequence characterization processing on the sequence information in the prediction request to obtain sequence characteristics;

    Before extracting the non-time-series feature corresponding to the non-time-series information in the prediction request, the method includes:

    Performing non-time-series characterization processing on the non-time-series information in the prediction request to obtain non-time-series features;

    The time sequence characterization process includes the following steps:

    Splitting the timing information according to a preset splitting rule to obtain at least one timing sub-information;

    Judging whether the number of obtained timing sub-information reaches a preset timing threshold;

    If yes, sum up the time sequence sub-information to obtain a sub-information set;

    If not, create zero-padded sub-information and summarize it with the timing sub-information to obtain a sub-information set, so that the number of zero-padded sub-information and timing sub-information in the sub-information set reaches the timing threshold;

    Extract the timing sub-information in the sub-information set, and identify the vector value corresponding to each data item in the timing sub-information through a preset timing vector table, according to the position of the data item in the timing sub-information Arranging the vector values corresponding to each of the data items to obtain the time sequence sub-features corresponding to the time sequence sub-information; wherein, the data item is an indivisible minimum unit in the time sequence sub-information;

    According to the position of the timing sub-information in the timing information, the timing sub-features are arranged to obtain the timing characteristics corresponding to the timing information.
12. The computer device according to claim 11, wherein the non-sequential characterization processing includes the following steps:

    Identify the vector value corresponding to each data item in the non-sequential information through a preset non-sequential vector table, where the data item is an indivisible minimum unit in the non-sequential sub-information;

    According to the position of the data item in the non-sequential information, the vector values corresponding to each of the data items are arranged and summarized to obtain the non-sequential feature.
13. The computer device according to claim 9, wherein an LSTM model is used as the time series model, and the time series model is obtained by training the initial LSTM model with preset time series samples;

    A deep neural network composed of multiple fully connected layers is used as the non-sequential model, and the non-sequential model is obtained by training the deep neural network through preset non-sequential samples.
14. The computer device according to claim 9, wherein the objective function of the prediction model is a geometric mean function, as follows:

    

    Among them, M is the prediction information, Scoe1 is the time series prediction result, and Scoe2 is the non-time series prediction result;

    After calculating the time series prediction result and the non-time series prediction result through the preset prediction model to obtain the prediction information, it includes:

    Upload the prediction information to the blockchain.
15. A computer-readable storage medium having a computer program stored on the readable storage medium, wherein the computer program stored in the readable storage medium is executed by a processor to implement a combined prediction method of multi-dimensional information, the Combination forecasting methods of multi-dimensional information include:

    Receive the prediction request sent by the client;

    Extracting the timing feature corresponding to the timing information in the prediction request, calling a preset timing model to calculate the timing feature to obtain a timing prediction result;

    Extracting the non-time-series features corresponding to the non-time-series information in the prediction request, and call a preset non-time-series model to calculate the non-time-series features to obtain a non-time-series prediction result;

    The prediction information is obtained by calculating the time series prediction result and the non-time series prediction result through a preset prediction model.
16. The computer-readable storage medium according to claim 15, wherein after receiving the prediction request sent by the user terminal, the method comprises:

    The object information in the prediction request is extracted, and the time series model and the non-time series model corresponding to the object information are obtained from a preset model set.
17. 15. The computer-readable storage medium according to claim 15, wherein before extracting the timing feature corresponding to the timing information in the prediction request, the method comprises:

    Performing sequence characterization processing on the sequence information in the prediction request to obtain sequence characteristics;

    Before extracting the non-time-series feature corresponding to the non-time-series information in the prediction request, the method includes:

    Performing non-time-series characterization processing on the non-time-series information in the prediction request to obtain non-time-series features;

    The time sequence characterization process includes the following steps:

    Splitting the timing information according to a preset splitting rule to obtain at least one timing sub-information;

    Judging whether the number of obtained timing sub-information reaches a preset timing threshold;

    If yes, sum up the time sequence sub-information to obtain a sub-information set;

    If not, create zero-padded sub-information and summarize it with the timing sub-information to obtain a sub-information set, so that the number of zero-padded sub-information and timing sub-information in the sub-information set reaches the timing threshold;

    Extract the timing sub-information in the sub-information set, and identify the vector value corresponding to each data item in the timing sub-information through a preset timing vector table, according to the position of the data item in the timing sub-information Arranging the vector values corresponding to each of the data items to obtain the time sequence sub-features corresponding to the time sequence sub-information; wherein, the data item is an indivisible minimum unit in the time sequence sub-information;

    According to the position of the timing sub-information in the timing information, the timing sub-features are arranged to obtain the timing characteristics corresponding to the timing information.
18. 18. The computer-readable storage medium of claim 17, wherein the non-sequential characterization process comprises the following steps:

    Identifying the vector value corresponding to each data item in the non-sequential information through a preset non-sequential vector table, where the data item is an indivisible minimum unit in the non-sequential sub-information;

    According to the position of the data item in the non-sequential information, the vector values corresponding to each of the data items are arranged and summarized to obtain the non-sequential feature.
19. The computer-readable storage medium according to claim 15, wherein an LSTM model is used as the time series model, and the time series model is obtained by training the initial LSTM model with preset time series samples;

    A deep neural network composed of multiple fully connected layers is used as the non-sequential model, and the non-sequential model is obtained by training the deep neural network through preset non-sequential samples.
20. The computer-readable storage medium according to claim 15, wherein the objective function of the prediction model is a geometric mean function, as follows:

    

    Among them, M is the prediction information, Scoe1 is the time series prediction result, and Scoe2 is the non-time series prediction result;

    After calculating the time series prediction result and the non-time series prediction result through the preset prediction model to obtain the prediction information, it includes:

    Upload the prediction information to the blockchain.

Images