WO2021022933A1

WO2021022933A1 - Method and device for multitask prediction, electronic device, and storage medium

Info

Publication number: WO2021022933A1
Application number: PCT/CN2020/098233
Authority: WO
Inventors: 王涛; 朱葛
Original assignee: 平安科技（深圳）有限公司
Priority date: 2019-08-06
Filing date: 2020-06-24
Publication date: 2021-02-11
Also published as: CN110619423B; CN110619423A

Abstract

A method and device for multitask prediction, an electronic device, and a storage medium. The method: when a prediction instruction is received, acquiring current scenario data (S10); determining, on the basis of the current scenario data, a target task corresponding to the current scenario data (S11); determining whether the target task is a prediction task appearing for the first time (S12); when the target task is a prediction task appearing for the first time, acquiring target data related to the target task (S13); proportionally splitting the target data to produce a first dataset and a second dataset (S14); preprocessing the first dataset to produce data characteristics (S15); inputting the data characteristics into at least one pretrained model to produce at least one prediction result (S16); employing a long short-term memory algorithm to train the at least one prediction result to produce a target model (S17); and inputting the second dataset into the target model to produce a target result (S18). Not only can a prediction be made as required via the target model, but a time series prediction can also be made on the basis of the prediction task.

Description

Multitask prediction method, device, electronic equipment and storage medium

This application claims the priority of the Chinese patent application filed with the Chinese Patent Office on August 6, 2019, with the application number 201910722718.4 and the invention title "Multitasking prediction method, device, electronic equipment and storage medium", the entire content of which is incorporated by reference Incorporated in this application.

Technical field

This application relates to the technical field of intelligent decision-making, and in particular to a multi-task prediction method, device, electronic equipment and storage medium.

Background technique

With the rapid development of artificial intelligence, computer technology has facilitated people's lives in all walks of life, and it is no exception in terms of predicting specific scenarios. However, in the existing technical solutions, multiple scenarios are predicted. The inventor realizes that a model needs to be trained for each scenario, resulting in low prediction efficiency. Therefore, how to train a prediction for multiple scenarios The model has become an urgent problem to be solved. In addition, every time the same task is predicted, it is still necessary to obtain data prediction from the Internet, which also reduces the efficiency of prediction.

Summary of the invention

In view of the above, it is necessary to provide a multi-task prediction method, device, electronic device, and storage medium, which can predict on-demand through the target model, and can also perform sequential prediction based on the prediction task.

A multi-task prediction method, the method includes:

When receiving the prediction instruction, obtain the current scene data;

Determine the target task corresponding to the current scene data according to the current scene data;

Judging whether the target task is a predicted task that appears for the first time;

When the target task is a predicted task that appears for the first time, acquiring target data related to the target task;

Split the target data in proportion to obtain a first data set and a second data set;

Preprocessing the first data set to obtain data characteristics;

Input the data feature into at least one pre-trained model to obtain at least one prediction result;

Training the at least one prediction result using a long and short-term memory algorithm to obtain a target model;

The second data set is input into the target model to obtain a target result.

A multi-task prediction device, the device includes:

The acquiring unit is used to acquire current scene data when the prediction instruction is received;

A determining unit, configured to determine a target task corresponding to the current scene data according to the current scene data;

A judging unit for judging whether the target task is a prediction task that appears for the first time;

The acquiring unit is further configured to acquire target data related to the target task when the target task is a predicted task that appears for the first time;

A splitting unit, configured to split the target data in proportion to obtain a first data set and a second data set;

A preprocessing unit, configured to preprocess the first data set to obtain data characteristics;

The input unit is used to input the data feature into at least one pre-trained model to obtain at least one prediction result;

A training unit, configured to train the at least one prediction result by using a long and short-term memory algorithm to obtain a target model;

The input unit is also used to input the second data set into the target model to obtain a target result.

An electronic device, which includes:

The memory stores at least one computer readable instruction; and

The processor executes at least one computer-readable instruction stored in the memory to implement the following steps:

When receiving the prediction instruction, obtain the current scene data;

Preprocessing the first data set to obtain data characteristics;

The second data set is input into the target model to obtain a target result.

A computer-readable storage medium in which at least one computer-readable instruction is stored, and the at least one computer-readable instruction is executed by a processor in an electronic device to implement the following steps:

When receiving the prediction instruction, obtain the current scene data;

Preprocessing the first data set to obtain data characteristics;

The second data set is input into the target model to obtain a target result.

It can be seen from the above technical solutions that the present application can be applied to the field of intelligent decision-making of artificial intelligence, not only can predict on-demand through the target model, but also can make sequential predictions based on the prediction task.

Description of the drawings

Fig. 1 is a flowchart of a preferred embodiment of the multi-task prediction method of the present application.

Fig. 2 is a functional block diagram of a preferred embodiment of the multi-task prediction device of the present application.

3 is a schematic structural diagram of an electronic device according to a preferred embodiment of the multi-task prediction method according to the present application.

detailed description

In order to make the objectives, technical solutions, and advantages of the present application clearer, the present application will be described in detail below with reference to the drawings and specific embodiments.

As shown in FIG. 1, it is a flowchart of a preferred embodiment of the multi-task prediction method of the present application. According to different needs, the order of the steps in the flowchart can be changed, and some steps can be omitted.

The multi-task prediction method is applied to one or more electronic devices. The electronic device is a device that can automatically perform numerical calculation and/or information processing in accordance with pre-set or stored instructions. Its hardware includes but not Limited to microprocessors, application specific integrated circuits (ASICs), programmable gate arrays (Field-Programmable Gate Arrays, FPGAs), digital processors (Digital Signal Processors, DSPs), embedded devices, etc.

The electronic device may be any electronic product that can interact with a user with a human machine, such as a personal computer, a tablet computer, a smart phone, a personal digital assistant (PDA), a game console, an interactive network television ( Internet Protocol Television, IPTV), smart wearable devices, etc.

The electronic device may also include a network device and/or user equipment. Wherein, the network device includes, but is not limited to, a single network server, a server group composed of multiple network servers, or a cloud composed of a large number of hosts or network servers based on Cloud Computing.

The network where the electronic device is located includes but is not limited to the Internet, wide area network, metropolitan area network, local area network, virtual private network (Virtual Private Network, VPN), etc.

S10: Acquire current scene data when a prediction instruction is received.

In at least one embodiment of the present application, the current scenario data may include, but is not limited to: stock trend scenario, product sales volume scenario, disease incidence scenario, etc.

In at least one embodiment of the present application, the prediction instruction may be triggered by the user, or may be automatically triggered when certain conditions are met, which is not limited in the present application.

Wherein, the meeting certain conditions includes, but is not limited to: meeting a preset time, etc.

The preset time may include a determined time point, or include a time period, etc., for example: the preset time may be 7 o'clock in the morning every day.

S11: Determine a target task corresponding to the current scene data according to the current scene data.

In at least one embodiment of the present application, the electronic device determining the target task to which the current scene data belongs according to the current scene data includes:

The electronic device matches the current scene data with pre-configured scene data, and determines a task corresponding to the matched scene data as the target task.

For example, the target task may include, but is not limited to: predicting the sales volume of A product, predicting the stock trend of X stock, predicting the incidence of D disease, etc.

Through the foregoing implementation manners, the target task can be quickly and accurately identified, so that it is convenient to determine whether the target task is a predicted task that appears for the first time.

S12: Determine whether the target task is a predicted task that appears for the first time.

In at least one embodiment of the present application, the electronic device determining whether the target task is a predicted task that appears for the first time includes:

The electronic device detects the target task, and when it is detected that the target task has not been trained before a preset time point, the electronic device determines that the target task is a prediction task that appears for the first time, and when the target task is detected The task is out of training before the preset time point, and the electronic device determines that the target task is not a predicted task that appears for the first time. Wherein, the preset time point may include the time when the prediction task is received, which is not limited in this application.

Through the foregoing implementation manners, it can be accurately determined whether the target task is a predicted task that appears for the first time, which facilitates subsequent processing of the target task.

S13: When the target task is a predicted task that appears for the first time, obtain target data related to the target task.

In at least one embodiment of the present application, the electronic device acquiring target data related to the target task includes, but is not limited to, one or a combination of the following methods:

(1) The electronic device uses Web crawler technology to obtain target data related to the target task from the Internet.

Wherein, the Internet may include any website that supports access, such as Baidu, Google, Tencent, Weibo, etc.

Further, the web crawler technology (also known as web spider, web robot) is a way to automatically grab information on the World Wide Web according to certain rules.

For example: when the target task is to predict the stock trend of X stock, then the target data is the trend of X stock in the past preset time period; when the target task is to predict the sales volume of A product, then the target The data is the sales volume of A product in the past preset time period.

Through the foregoing implementation manners, more target data can be obtained, thereby improving the training effect of the target model and reducing the training error of the target model.

(2) The electronic device receives target data related to the target task uploaded by the user.

Through the foregoing implementation manners, accurate target data can be obtained, which is beneficial to obtaining a more accurate target model later.

S14: Split the target data in proportion to obtain a first data set and a second data set.

In at least one embodiment of the present application, the electronic device splits the target data in proportion to obtain the first data set and the second data set.

Specifically, the electronic device determines a preset proportion of target data in the target data as the first data set, wherein the first data set is used to train at least one model, and further, the electronic device The target data except for the first data set in the target data is determined as the second data set, wherein the second data set is used as the input data of the target model. Wherein, the preset ratio is not limited, and may be 0.8, 0.6, etc.

S15: Perform preprocessing on the first data set to obtain data characteristics.

In at least one embodiment of the present application, the electronic device preprocessing the first data set to obtain data characteristics includes:

The electronic device performs deviation detection on the first data set to obtain deviation data. Further, the electronic device deletes the deviation data to obtain the data characteristics.

In at least one embodiment of the present application, the electronic device uses a density-based outlier detection method to perform deviation detection on the first data set to obtain deviation data.

Specifically, the electronic device uses a relative density detection technology to divide the first data set into several objects, calculates the density of each object separately, and obtains the outlier score of each object. Further, the electronic The device calculates the neighborhood average density of each object, and when the outlier score of the object is less than the neighborhood average density corresponding to the object, the object is determined as deviation data.

Further, the electronic device deletes the deviation data from the first data set to obtain the data characteristics.

Through the foregoing implementation manners, the deviation data can be accurately obtained and eliminated, which is beneficial to the subsequent accurate training of the target model, thereby improving the accuracy of the target model.

Of course, the method of preprocessing the first data set is not limited in this application as long as it is legal and reasonable.

S16. Input the data feature into at least one pre-trained model to obtain at least one prediction result.

In at least one embodiment of the present application, the electronic device trains the at least one model before inputting the data features into the at least one pre-trained model to obtain at least one prediction result.

Specifically, training the at least one model by the electronic device includes, but is not limited to, one or a combination of the following methods:

(1) The electronic device acquires a first training set related to the target task, wherein the first training set does not intersect with the first data set; further, the electronic device uses neural network algorithm training In the first training set, the at least one model is obtained.

Specifically, the electronic device performs normalization processing on the first training set, and further, the electronic device constructs a network using the first training set after the normalization processing to obtain the first network, and the electronic device Training the first network by using a preset learning rate to obtain the at least one model.

It should be noted that the learning rate of the at least one model obtained by training may be the same or different. When a learning rate is configured in advance, the learning rate of the at least one model is infinitely close to the learning rate, and when multiple learning rates are configured, the relationship between the at least one model and the multiple learning rates The configuration can be customized (for example, the electronic device is configured with a learning rate A and a learning rate B, and it is determined that there are 3 models that need to be trained based on the learning rate A, and it is determined that 4 models need to be trained based on the learning rate B, Then the at least one model is the 7 models trained above).

Through the foregoing implementation manners, a more accurate model can be obtained, and the accuracy of subsequent training of the target model can be improved.

(2) The electronic device acquires a first training set related to the target task, wherein the first training set does not intersect with the first data set; further, the electronic device uses a linear regression algorithm for training In the first training set, the at least one model is obtained.

Specifically, the electronic device constructs a model based on the first training set to obtain a prediction function. Further, the electronic device uses a gradient descent algorithm to reduce the error of the prediction function, and obtains a prediction function with an error less than a threshold, which is The at least one model.

Wherein, the threshold is set in advance and can be: 0.2, etc., which is not limited in this application.

Through the foregoing implementation manners, the electronic device can quickly obtain a model, which improves the speed of subsequent training of the target model.

In at least one embodiment of the present application, the electronic device inputs the data feature into the at least one pre-trained model to obtain at least one prediction result.

Specifically, the electronic device inputs the data characteristics into each of the at least one model to obtain at least one first result of each model, and further, the electronic device integrates at least one of each model The first result is the at least one prediction result.

Through the foregoing implementation manners, multiple prediction results can be obtained, providing a training basis for training the target model.

S17. Train the at least one prediction result using a long- and short-term memory algorithm to obtain a target model.

The Long Short-Term Memory (LSTM) algorithm includes three network layers, where the three network layers are an input gate layer, a forget gate layer, and an output gate layer.

In at least one embodiment of the present application, the electronic device training the at least one prediction result by using a long- and short-term memory algorithm to obtain a target model includes:

The electronic device inputs the at least one prediction result into the forgetting gate layer to perform forgetting processing to obtain second training data. Further, the electronic device uses a cross-validation method to divide the second training data into a second training set And a second verification set, input the second training set to the input gate layer for training to obtain a secondary learner, and adjust the secondary learner according to the second verification set to obtain a target model.

In at least one embodiment of the present application, the electronic device uses a cross-validation method to divide the second training data into a second training set and a second verification set, which specifically includes:

The electronic device randomly divides the second training data into at least one data packet according to a preset number, determines any one of the at least one data packet as the second verification set, and determines the remaining data packets For the second training set, repeat the above steps until all data packets are used as the second verification set in turn.

For example: the electronic device divides the second training data into three data packets, namely data packet E, data packet F, and data packet G, and determines the data packet E as the verification set, and the data packet F and data packet G are determined to be the second training set. Secondly, the data packet F is determined as the verification set, and the data packet E and the data packet G are determined as the second training set. Finally, the data packet G is determined to be the verification set, and the data packet E and the data packet F are determined to be the second training set.

Through the foregoing implementation manner, the second training data is divided by a cross-validation method, so that all of the second training data participates in training and verification, thereby improving the fitness of training the target model.

In at least one embodiment of the present application, the electronic device adjusting the secondary learner according to the second verification set to obtain the target model includes:

The electronic device uses a hyperparameter grid search method to obtain optimal hyperparameter points from the second verification set, and further, the electronic device adjusts the secondary learner through the optimal hyperparameter points, Obtain the target model.

Specifically, the electronic device splits the second verification set according to a fixed step to obtain a target subset, traverses the parameters of the two end points on the target subset, and passes the parameter verification point of the two end points. The secondary learner obtains the learning rate of each parameter, determines the parameter with the best learning rate as the first hyperparameter point, and in the neighborhood of the first hyperparameter point, reduces the step size and continues Traverse until the step length is the preset step length, that is, the obtained hyperparameter point is the optimal hyperparameter point, and further, the electronic device adjusts the secondary learning according to the optimal hyperparameter point Device to obtain the target model.

Among them, this application does not limit the preset step length.

Through the foregoing implementation manners, a more accurate target model can be obtained, and further accurate target results can be obtained.

In at least one embodiment of the present application, since the long and short-term memory algorithm has the advantage of time series, the target model trained by the long- and short-term memory algorithm also has a certain time sequence. Through the above-mentioned embodiments, The sequential target model is quickly obtained, which is convenient for subsequent sequential prediction of the prediction task.

In at least one embodiment of the present application, after determining whether the target task is a predicted task that appears for the first time, the method further includes:

When it is determined that the target task is not the predicted task that appears for the first time, the electronic device obtains target data when the target task first appears, and inputs the target data into the target model to obtain a target result.

Through the foregoing implementation manners, after judging that the target task is not a prediction task that appears for the first time, the target model can be directly used for prediction, which can avoid repeated training of the target model, thereby improving the efficiency of prediction.

S18: Input the second data set into the target model to obtain a target result.

In at least one embodiment of the present application, after the second data set is input into the target model and the target result is obtained, the method further includes:

The electronic device detects whether the target result is abnormal, and when it detects that the target result is abnormal, generates alarm information, and sends the alarm information to a terminal device of a designated contact.

Wherein, the alarm information may include target tasks, target results, and predicted time points.

Further, the designated contact person may include the user who triggered the prediction task, and the like.

Through the foregoing implementation manners, when the target result is abnormal, the target result can be alerted in advance and reminded in time, which is beneficial for the user to take precautionary measures in advance.

For example: when the target task is to predict the stock trend of X stock, and the target result is the stock trend of X stock in the future preset time period, the electronic device detects that the stock trend of X stock exists in the next week Risk, further, the electronic device generates the alarm information, and sends the alarm information to the terminal device of the designated contact.

When the target task is to predict the sales volume of product A, and the target result is the sales volume of product A in the next month, the electronic device detects that the sales volume of product A is less than a threshold, and further, The electronic device generates the alarm information, and sends the alarm information to the terminal device of the designated contact.

Wherein, the threshold may be a preset sales volume, which is not limited in this application.

It can be seen from the above technical solutions that this application can be applied to the field of intelligent decision-making in artificial intelligence. When a prediction instruction is received, current scene data can be obtained, and the target task to which the current scene data belongs can be determined according to the current scene data. , Judge whether the target task is a predicted task that appears for the first time, and when the target task is a predicted task that appears for the first time, obtain target data related to the target task, divide the target data in proportion to obtain the first data The first data set is preprocessed to obtain the data feature, and the data feature is input into at least one pre-trained model to obtain at least one prediction result. The long and short-term memory algorithm is used to train the The at least one prediction result is used to obtain a target model, and the second data set is input to the target model to obtain the target result. Not only can the target model be predicted on-demand, but also time-series predictions can be made according to the prediction task .

As shown in FIG. 2, it is a functional block diagram of a preferred embodiment of the multi-task prediction device of the present application. The multi-task prediction device 11 includes an acquisition unit 110, a determination unit 111, a judgment unit 112, a split unit 113, a preprocessing unit 114, an input unit 115, a training unit 116, a generation unit 117, a sending unit 118, and a detection unit 119. The module/unit referred to in this application refers to a series of computer program segments that can be executed by the processor 13 and can complete fixed functions, and are stored in the memory 12. In this embodiment, the functions of each module/unit will be described in detail in subsequent embodiments.

When a prediction instruction is received, the obtaining unit 110 obtains current scene data.

The determining unit 111 determines the target task corresponding to the current scene data according to the current scene data.

In at least one embodiment of the present application, the determining unit 111 determining the target task to which the current scene data belongs according to the current scene data includes:

The determining unit 111 matches the current scene data with pre-configured scene data, and determines a task corresponding to the matched scene data as the target task.

The judging unit 112 judges whether the target task is a predicted task that appears for the first time.

In at least one embodiment of the present application, the judging unit 112 judging whether the target task is a predicted task that appears for the first time includes:

The judgment unit 112 detects the target task. When it is detected that the target task has not been trained before a preset time point, the judgment unit 112 determines that the target task is a prediction task that appears for the first time. If the target task is out of training before the preset time point, the judgment unit 112 determines that the target task is not a prediction task that appears for the first time. Wherein, the preset time point may include the time when the prediction task is received, which is not limited in this application.

When the target task is a predicted task that appears for the first time, the obtaining unit 110 obtains target data related to the target task.

In at least one embodiment of the present application, the acquiring unit 110 acquiring target data related to the target task includes, but is not limited to, one or a combination of the following methods:

(1) The obtaining unit 110 uses web crawler technology to obtain target data related to the target task from the Internet.

(2) The acquiring unit 110 receives target data related to the target task uploaded by the user.

The splitting unit 113 splits the target data in proportion to obtain the first data set and the second data set.

In at least one embodiment of the present application, the splitting unit 113 splits the target data in proportion to obtain a first data set and a second data set.

Specifically, the splitting unit 113 determines a preset ratio of target data in the target data as the first data set, where the first data set is used to train at least one model, and further, the The splitting unit 113 determines the target data except for the first data set in the target data as the second data set, where the second data set is used as the input data of the target model. Wherein, the preset ratio is not limited, and may be 0.8, 0.6, etc.

The preprocessing unit 114 preprocesses the first data set to obtain data characteristics.

In at least one embodiment of the present application, the preprocessing unit 114 preprocessing the first data set to obtain data characteristics includes:

The preprocessing unit 114 performs deviation detection on the first data set to obtain deviation data. Further, the preprocessing unit 114 deletes the deviation data to obtain the data characteristics.

In at least one embodiment of the present application, the preprocessing unit 114 adopts a density-based outlier detection method to perform deviation detection on the first data set to obtain deviation data.

Specifically, the preprocessing unit 114 uses relative density detection technology to divide the first data set into several objects, calculates the density of each object separately, and then obtains the outlier score of each object. Further, The preprocessing unit 114 calculates the average density of the neighborhood of each object, and when the outlier score of the object is less than the average density of the neighborhood corresponding to the object, the object is determined as deviation data.

Further, the preprocessing unit 114 deletes the deviation data from the first data set to obtain the data characteristics.

The input unit 115 inputs the data feature into at least one pre-trained model to obtain at least one prediction result.

In at least one embodiment of the present application, before inputting the data features into at least one pre-trained model and obtaining at least one prediction result, train the at least one model.

Specifically, training the at least one model includes, but is not limited to, one or a combination of the following methods:

(1) The acquiring unit 110 acquires a first training set related to the target task, wherein the first training set does not intersect with the first data set; further, the training unit 116 uses neural network algorithm training In the first training set, the at least one model is obtained.

Specifically, the training unit 116 performs normalization processing on the first training set, and further, the training unit 116 uses the normalized first training set to construct a network to obtain the first network. The training unit 116 uses a preset learning rate to train the first network to obtain the at least one model.

It should be noted that the learning rate of the at least one model obtained by training may be the same or different. When a learning rate is configured in advance, the learning rate of the at least one model is infinitely close to the learning rate, and when multiple learning rates are configured, the relationship between the at least one model and the multiple learning rates The configuration can be customized (for example, the training unit 116 is configured with learning rate A and learning rate B, and it is determined that there are 3 models that need to be trained based on the learning rate A, and it is determined that there are 4 models that need to be trained based on the learning rate B , The at least one model is the 7 models trained above).

(2) The acquiring unit 110 acquires a first training set related to the target task, wherein the first training set and the first data set are not intersected; further, the training unit 116 adopts linear regression The algorithm trains the first training set to obtain the at least one model.

Specifically, the training unit 116 constructs a model based on the first training set to obtain a prediction function. Further, the training unit 116 uses a gradient descent algorithm to reduce the error of the prediction function to obtain a prediction function with an error less than a threshold, That is the at least one model.

Through the foregoing implementation manner, the training unit 116 can quickly obtain a model, which improves the speed of subsequent training of the target model.

In at least one embodiment of the present application, the input unit 115 inputs the data feature into the at least one pre-trained model to obtain at least one prediction result.

Specifically, the input unit 115 inputs the data features into each of the at least one model to obtain at least one first result of each model. Further, the input unit 115 integrates the data of each model At least one first result, the at least one prediction result is obtained.

The training unit 116 uses a long and short-term memory algorithm to train the at least one prediction result to obtain a target model.

In at least one embodiment of the present application, the training unit 116 adopts a long and short-term memory algorithm to train the at least one prediction result, and obtaining a target model includes:

The training unit 116 inputs the at least one prediction result to the forgetting gate layer to perform forgetting processing to obtain second training data. Further, the training unit 116 uses a cross-validation method to divide the second training data into second training data. The training set and the second verification set, the second training set is input to the input gate layer for training to obtain a secondary learner, and the secondary learner is adjusted according to the second verification set to obtain a target model.

In at least one embodiment of the present application, the training unit 116 uses a cross-validation method to divide the second training data into a second training set and a second verification set, which specifically includes:

The training unit 116 randomly divides the second training data into at least one data packet according to a preset number, determines any one of the at least one data packet as the second verification set, and the remaining data packets It is determined as the second training set, and the above steps are repeated until all data packets are sequentially used as the second verification set.

For example: the training unit 116 divides the second training data into three data packets, namely data packet E, data packet F, and data packet G, and determines the data packet E as the verification set, and the data The packet F and the data packet G are determined as the second training set. Secondly, the data packet F is determined as the verification set, and the data packet E and the data packet G are determined as the second training set. Finally, the data packet G is determined to be the verification set, and the data packet E and the data packet F are determined to be the second training set.

In at least one embodiment of the present application, the training unit 116 adjusting the secondary learner according to the second verification set to obtain a target model includes:

The training unit 116 uses a hyperparameter grid search method to obtain the optimal hyperparameter points from the second verification set. Further, the training unit 116 performs an operation on the secondary learner through the optimal hyperparameter points. Adjust to obtain the target model.

Specifically, the training unit 116 splits the second verification set according to a fixed step size to obtain a target subset, traverses the parameters of the two ends of the target subset, and passes the parameter verification of the two ends. The secondary learner obtains the learning rate of each parameter, determines the parameter with the best learning rate as the first hyperparameter point, and reduces the step size in the neighborhood of the first hyperparameter point Continue to traverse until the step length is the preset step length, that is, the obtained hyperparameter point is the optimal hyperparameter point, and further, the training unit 116 adjusts the secondary hyperparameter point according to the optimal hyperparameter point. Level learner to obtain the target model.

Among them, this application does not limit the preset step length.

When it is determined that the target task is not the predicted task that appears for the first time, the obtaining unit 110 obtains the target data when the target task first appears, and further, the input unit 115 inputs the target data to the target model In, get the target result.

The input unit 115 inputs the second data set into the target model to obtain a target result.

The detection unit 119 detects whether the target result is abnormal, and when it is detected that the target result is abnormal, the generating unit 117 generates alarm information. Further, the sending unit 118 sends the alarm information to the terminal device of the designated contact.

For example: when the target task is to predict the stock trend of X stock, and the target result is the stock trend of X stock in the future preset time period, when it is detected that the stock trend of X stock in the next week is at risk, further Preferably, the generating unit 117 generates the alarm information, and further, the sending unit 118 sends the alarm information to the terminal device of the designated contact.

When the target task is to predict the sales volume of product A, and the target result is the sales volume of product A in the next month, when it is detected that the sales volume of product A is less than a threshold, further, the generating unit 117 generates the alarm information, and further, the sending unit 118 sends the alarm information to the terminal device of the designated contact.

As shown in FIG. 3, it is a schematic structural diagram of an electronic device according to a preferred embodiment of the multi-task prediction method of the present application.

The electronic device 1 is a device that can automatically perform numerical calculation and/or information processing according to pre-set or stored instructions. Its hardware includes, but is not limited to, a microprocessor, an Application Specific Integrated Circuit (ASIC) ), programmable gate array (Field-Programmable Gate Array, FPGA), digital processor (Digital Signal Processor, DSP), embedded equipment, etc.

The electronic device 1 can also be, but is not limited to, any electronic product that can interact with the user through a keyboard, a mouse, a remote control, a touch panel, or a voice control device, for example, a personal computer, a tablet computer, or a smart phone. , Personal Digital Assistant (PDA), game consoles, interactive network TV (Internet Protocol Television, IPTV), smart wearable devices, etc.

The electronic device 1 may also be a computing device such as a desktop computer, a notebook, a palmtop computer, and a cloud server.

The network where the electronic device 1 is located includes, but is not limited to, the Internet, a wide area network, a metropolitan area network, a local area network, a virtual private network (Virtual Private Network, VPN), etc.

In an embodiment of the present application, the electronic device 1 includes, but is not limited to, a memory 12, a processor 13, and a computer program stored in the memory 12 and running on the processor 13, such as Multi-task prediction program.

Those skilled in the art can understand that the schematic diagram is only an example of the electronic device 1 and does not constitute a limitation on the electronic device 1. It may include more or less components than those shown in the figure, or a combination of certain components, or different components. Components, for example, the electronic device 1 may also include input and output devices, network access devices, buses, and the like.

The processor 13 may be a central processing unit (Central Processing Unit, CPU), other general-purpose processors, digital signal processors (Digital Signal Processors, DSP), application specific integrated circuits (ASICs), Ready-made programmable gate array (Field-Programmable Gate Array, FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor, etc. The processor 13 is the computing core and control center of the electronic device 1 and connects the entire electronic device with various interfaces and lines. Each part of 1, and executes the operating system of the electronic device 1, and various installed applications, program codes, etc.

The processor 13 executes the operating system of the electronic device 1 and various installed applications. The processor 13 executes the application program to implement the steps in the foregoing embodiments of the multi-task prediction method, such as steps S10, S11, S12, S13, S14, S15, S16, S17, and S18 shown in FIG. 1.

Alternatively, when the processor 13 executes the computer program, the function of each module/unit in the foregoing device embodiments is implemented, for example: when a prediction instruction is received, current scene data is acquired; according to the current scene data, all The target task corresponding to the current scene data; determine whether the target task is a predicted task that appears for the first time; when the target task is a predicted task that appears for the first time, obtain target data related to the target task; split according to proportions The target data obtains a first data set and a second data set; preprocessing the first data set to obtain data features; inputting the data features into at least one pre-trained model to obtain at least one prediction Result; training the at least one prediction result using a long and short-term memory algorithm to obtain a target model; inputting the second data set into the target model to obtain a target result.

Exemplarily, the computer program may be divided into one or more modules/units, and the one or more modules/units are stored in the memory 12 and executed by the processor 13 to complete this Application. The one or more modules/units may be a series of computer program instruction segments capable of completing specific functions, and the instruction segments are used to describe the execution process of the computer program in the electronic device 1. For example, the computer program can be divided into an acquisition unit 110, a determination unit 111, a judgment unit 112, a split unit 113, a preprocessing unit 114, an input unit 115, a training unit 116, a generation unit 117, a transmission unit 118, and a detection unit. 119.

The memory 12 may be used to store the computer program and/or module, and the processor 13 runs or executes the computer program and/or module stored in the memory 12 and calls the data stored in the memory 12, Various functions of the electronic device 1 are realized. The memory 12 may mainly include a storage program area and a storage data area. The storage program area may store an operating system, an application program required by at least one function (such as a sound playback function, an image playback function, etc.), etc.; the storage data area may Store data (such as audio data, phone book, etc.) created based on the use of mobile phones. In addition, the memory 12 may include high-speed random access memory, and may also include computer memory, such as a hard disk, a memory, a plug-in hard disk, a Smart Media Card (SMC), a Secure Digital (SD) card, and a flash memory. Card (Flash Card), at least one magnetic disk storage device, flash memory device, or other volatile/non-volatile storage device.

The memory 12 may be an external memory and/or an internal memory of the electronic device 1. Further, the memory 12 may be a circuit with a storage function without a physical form in an integrated circuit, such as RAM (Random-Access Memory, random access memory), FIFO (First In First Out), etc. Alternatively, the memory 12 may also be a memory in physical form, such as a memory stick, a TF card (Trans-flash Card), and so on.

If the integrated module/unit of the electronic device 1 is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer readable storage medium. The computer-readable storage medium may be non-volatile or volatile. Based on this understanding, this application implements all or part of the processes in the above-mentioned embodiments and methods, and can also be completed by instructing relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium. When the program is executed by the processor, the steps of the foregoing method embodiments can be implemented.

Wherein, the computer program includes computer program code, and the computer program code may be in the form of source code, object code, executable file, or some intermediate forms. The computer-readable medium may include: any entity or device capable of carrying the computer program code, recording medium, U disk, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM, Read-Only Memory) , Random Access Memory (RAM, Random Access Memory), etc.

With reference to FIG. 1, the memory 12 in the electronic device 1 stores multiple instructions to implement a multi-task prediction method, and the processor 13 can execute the multiple instructions to realize: when a prediction instruction is received, Acquire current scene data; determine the target task corresponding to the current scene data according to the current scene data; determine whether the target task is a prediction task that appears for the first time; when the target task is a prediction task that appears for the first time, obtain Target data related to the target task; split the target data in proportion to obtain a first data set and a second data set; preprocess the first data set to obtain data characteristics; combine the data characteristics Input into at least one pre-trained model to obtain at least one prediction result; train the at least one prediction result using a long and short-term memory algorithm to obtain a target model; input the second data set into the target model to obtain a target result.

Specifically, for the specific implementation method of the processor 13 for the foregoing instructions, reference may be made to the description of the relevant steps in the embodiment corresponding to FIG. 1, which is not repeated here.

In the several embodiments provided in this application, it should be understood that the disclosed system, device, and method may be implemented in other ways. For example, the device embodiments described above are only illustrative. For example, the division of the modules is only a logical function division, and there may be other division methods in actual implementation.

The modules described as separate components may or may not be physically separated, and the components displayed as modules may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the objectives of the solutions of the embodiments.

In addition, the functional modules in the various embodiments of the present application may be integrated into one processing unit, or each unit may exist alone physically, or two or more units may be integrated into one unit. The above-mentioned integrated unit can be implemented in the form of hardware or in the form of hardware plus software functional modules.

For those skilled in the art, it is obvious that the present application is not limited to the details of the foregoing exemplary embodiments, and the present application can be implemented in other specific forms without departing from the spirit or basic characteristics of the application.

Therefore, no matter from which point of view, the embodiments should be regarded as exemplary and non-limiting. The scope of this application is defined by the appended claims rather than the above description, and therefore it is intended to fall into the claims. All changes in the meaning and scope of the equivalent elements of are included in this application. Any associated diagram marks in the claims should not be regarded as limiting the claims involved.

In addition, it is obvious that the word "including" does not exclude other units or steps, and the singular does not exclude the plural. Multiple units or devices stated in the system claims can also be implemented by one unit or device through software or hardware. The second class words are used to indicate names, and do not indicate any specific order.

Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the application and not to limit them. Although the application has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the application can be Make modifications or equivalent replacements without departing from the spirit and scope of the technical solution of the present application.

Claims

A multi-task prediction method, wherein the method includes:

When receiving the prediction instruction, obtain the current scene data;

Determine the target task corresponding to the current scene data according to the current scene data;

Judging whether the target task is a predicted task that appears for the first time;

When the target task is a predicted task that appears for the first time, acquiring target data related to the target task;

Split the target data in proportion to obtain a first data set and a second data set;

Preprocessing the first data set to obtain data characteristics;

Input the data feature into at least one pre-trained model to obtain at least one prediction result;

Training the at least one prediction result using a long and short-term memory algorithm to obtain a target model;

The second data set is input into the target model to obtain a target result.
The multi-task prediction method according to claim 1, wherein said obtaining target data related to said target task comprises one or a combination of the following methods:

Use web crawler technology to obtain target data related to the target task from the Internet; and/or

Receive target data related to the target task uploaded by the user.
The multi-task prediction method according to claim 1, wherein said preprocessing said first data set to obtain data features comprises:

Performing deviation detection on the first data set to obtain deviation data;

Delete the deviation data to obtain the data characteristics.
The multi-task prediction method of claim 1, wherein, before inputting the data features into at least one pre-trained model and obtaining at least one prediction result, the method further comprises:

Acquiring a first training set related to the target task, wherein the first training set and the first data set do not intersect;

The neural network algorithm and/or linear regression algorithm are used to train the first training set to obtain the at least one model.
8. The multi-task prediction method according to claim 1, wherein the training of the at least one prediction result using a long and short-term memory algorithm to obtain a target model comprises:

Inputting the at least one prediction result to the forgetting gate layer for forgetting processing to obtain second training data;

Dividing the second training data into a second training set and a second verification set by using a cross-validation method;

Input the second training set to the input gate layer for training to obtain a secondary learner;

According to the second verification set, the secondary learner is adjusted to obtain the target model.
The multi-task prediction method according to claim 1, wherein after determining whether the target task is a prediction task that appears for the first time, the method further comprises:

When it is judged that the target task is not the predicted task that appears for the first time, acquiring the target data when the target task first appears;

Input the target data into the target model to obtain the target result.
The multi-task prediction method of claim 1, wherein after the second data set is input into the target model and the target result is obtained, the method further comprises:

Detecting whether the target result is abnormal;

When an abnormality in the target result is detected, an alarm message is generated;

Send the alarm information to the terminal device of the designated contact.
A multi-task prediction device, wherein the device includes:

The acquiring unit is used to acquire current scene data when the prediction instruction is received;

A determining unit, configured to determine a target task corresponding to the current scene data according to the current scene data;

A judging unit for judging whether the target task is a prediction task that appears for the first time;

The acquiring unit is further configured to acquire target data related to the target task when the target task is a predicted task that appears for the first time;

A splitting unit, configured to split the target data in proportion to obtain a first data set and a second data set;

A preprocessing unit, configured to preprocess the first data set to obtain data characteristics;

The input unit is used to input the data feature into at least one pre-trained model to obtain at least one prediction result;

A training unit, configured to train the at least one prediction result by using a long and short-term memory algorithm to obtain a target model;

The input unit is also used to input the second data set into the target model to obtain a target result.
An electronic device, wherein the electronic device includes:

The memory stores at least one computer readable instruction; and

The processor executes at least one computer-readable instruction stored in the memory to implement the following steps:

When receiving the prediction instruction, obtain the current scene data;

Determine the target task corresponding to the current scene data according to the current scene data;

Judging whether the target task is a predicted task that appears for the first time;

When the target task is a predicted task that appears for the first time, acquiring target data related to the target task;

Split the target data in proportion to obtain a first data set and a second data set;

Preprocessing the first data set to obtain data characteristics;

Input the data feature into at least one pre-trained model to obtain at least one prediction result;

Training the at least one prediction result using a long and short-term memory algorithm to obtain a target model;

The second data set is input into the target model to obtain a target result.
9. The electronic device according to claim 9, wherein the execution of at least one computer-readable instruction by the processor to achieve the acquisition of target data related to the target task comprises one or a combination of the following methods:

Use web crawler technology to obtain target data related to the target task from the Internet; and/or

Receive target data related to the target task uploaded by the user.
9. The electronic device of claim 9, wherein the processor executes at least one computer readable instruction to implement the preprocessing of the first data set to obtain the data characteristics, comprising the following steps:

Performing deviation detection on the first data set to obtain deviation data;

Delete the deviation data to obtain the data characteristics.
The electronic device of claim 9, wherein, before the data feature is input into at least one pre-trained model to obtain at least one prediction result, the processor executes at least one computer-readable instruction to further implement The following steps:

Acquiring a first training set related to the target task, wherein the first training set and the first data set do not intersect;

The neural network algorithm and/or linear regression algorithm are used to train the first training set to obtain the at least one model.
9. The electronic device according to claim 9, wherein the processor executes at least one computer-readable instruction to implement the training of the at least one prediction result using a long- and short-term memory algorithm, and when the target model is obtained, the method comprises the following steps:

Inputting the at least one prediction result to the forgetting gate layer for forgetting processing to obtain second training data;

Dividing the second training data into a second training set and a second verification set by using a cross-validation method;

Input the second training set to the input gate layer for training to obtain a secondary learner;

According to the second verification set, the secondary learner is adjusted to obtain the target model.
9. The electronic device of claim 9, wherein after determining whether the target task is a predicted task that appears for the first time, the processor executes at least one computer-readable instruction to further implement the following steps:

When it is judged that the target task is not the predicted task that appears for the first time, obtain the target data when the target task first appears;

Input the target data into the target model to obtain the target result.
The electronic device according to claim 9, wherein, after the second data set is input into the target model and the target result is obtained, the processor executes at least one computer-readable instruction to further implement the following steps :

Detecting whether the target result is abnormal;

When an abnormality in the target result is detected, an alarm message is generated;

Send the alarm information to the terminal device of the designated contact.
A computer-readable storage medium, wherein: the computer-readable storage medium stores at least one computer-readable instruction, and the at least one computer-readable instruction is executed by a processor in an electronic device to implement the following steps:

When receiving the prediction instruction, obtain the current scene data;

Determine the target task corresponding to the current scene data according to the current scene data;

Judging whether the target task is a predicted task that appears for the first time;

When the target task is a predicted task that appears for the first time, acquiring target data related to the target task;

Split the target data in proportion to obtain a first data set and a second data set;

Preprocessing the first data set to obtain data characteristics;

Input the data feature into at least one pre-trained model to obtain at least one prediction result;

Training the at least one prediction result using a long and short-term memory algorithm to obtain a target model;

The second data set is input into the target model to obtain a target result.
The storage medium according to claim 16, wherein, when the at least one computer-readable instruction is executed by a processor to realize the obtaining of target data related to the target task, the method includes one or a combination of the following: :

Use web crawler technology to obtain target data related to the target task from the Internet; and/or

Receive target data related to the target task uploaded by the user.
15. The storage medium of claim 16, wherein the at least one computer-readable instruction is executed by a processor to perform the preprocessing of the first data set to obtain the data characteristics, comprising the following steps:

Performing deviation detection on the first data set to obtain deviation data;

Delete the deviation data to obtain the data characteristics.
The storage medium of claim 16, wherein, before the data feature is input into at least one pre-trained model to obtain at least one prediction result, the at least one computer readable instruction is executed by the processor for Implement the following steps:

Acquiring a first training set related to the target task, wherein the first training set and the first data set do not intersect;

The neural network algorithm and/or linear regression algorithm are used to train the first training set to obtain the at least one model.
15. The storage medium according to claim 16, wherein the at least one computer readable instruction is executed by a processor to implement the training of the at least one prediction result using a long and short-term memory algorithm to obtain a target model, comprising the following steps:

Inputting the at least one prediction result to the forgetting gate layer for forgetting processing to obtain second training data;

Dividing the second training data into a second training set and a second verification set by using a cross-validation method;

Input the second training set to the input gate layer for training to obtain a secondary learner;

According to the second verification set, the secondary learner is adjusted to obtain the target model.
15. The storage medium of claim 16, wherein after determining whether the target task is a predicted task that appears for the first time, the at least one computer readable instruction is executed by the processor to further implement the following steps:

When it is judged that the target task is not the predicted task that appears for the first time, acquiring the target data when the target task first appears;

Input the target data into the target model to obtain the target result.
The storage medium according to claim 16, wherein, after the second data set is input into the target model and the target result is obtained, the at least one computer readable instruction is executed by the processor to further implement the following step:

Detecting whether the target result is abnormal;

When an abnormality in the target result is detected, an alarm message is generated;

Send the alarm information to the terminal device of the designated contact.