CN-121997041-A - Variable prediction method, device, equipment and storage medium

CN121997041ACN 121997041 ACN121997041 ACN 121997041ACN-121997041-A

Abstract

The disclosure provides a variable prediction method, a variable prediction device, variable prediction equipment and a storage medium, wherein the variable prediction method comprises the steps of obtaining multi-factor data, constructing a first data set based on the multi-factor data, determining a corresponding first prediction model group based on the first data set, performing model training on the first data set based on the first prediction model group to generate a first prediction result set, performing model training on a second prediction model based on the first prediction result set and the first data set, and outputting a second prediction result. Through the innovative multi-teacher knowledge distillation framework, the attention mechanism-based environment factor dynamic fusion strategy and the efficient edge-cloud collaborative computing framework, the existing defects of the mechanism model are fundamentally overcome, and a general, efficient and robust solution is provided for multi-factor time sequence variable prediction in a complex environment.

Inventors

ZHANG ZHENCHANG
YANG LIANG
CHEN LEI
XIAO HONGYUAN
LIN YE
HE SONGTAO
CHEN YONGXIANG
CHEN HAIQIANG

Assignees

福建农林大学

Dates

Publication Date: 20260508
Application Date: 20251204

Claims (10)

1. A method of variable prediction, the method comprising: Acquiring multi-factor data, and constructing a first data set based on the multi-factor data; Determining a corresponding first set of predictive models based on the first dataset; Model training is carried out on the first data set based on the first prediction model group, and a first prediction result set is generated; and training a second prediction model based on the first prediction result set and the first data set, and outputting a second prediction result.
2. The method of claim 1, wherein the determining a corresponding first set of predictive models based on the first set of data comprises: The multi-factor data corresponds to at least one data factor in the first data set, and corresponding at least one prediction model is determined based on the at least one data factor; The first set of prediction models is composed based on the at least one prediction model.
3. The method of claim 1, wherein the constructing a first data set based on the multi-factor data comprises: Setting a window and a prediction step length, and performing sliding window processing on the multi-factor data based on the window and the prediction step length to obtain multi-sample data; based on the first time as a reference, aligning the multi-sample data into a unified time axis and space grid based on an interpolation algorithm, and performing space-time alignment processing; And carrying out normalization and standardization processing on the space-time aligned multi-sample data, and embedding a mode identifier into each channel in the multi-sample data to obtain the first data set in response to the completion of the normalization and standardization processing.
4. The method of claim 2, wherein the training a second predictive model based on the first set of predictive results and the first set of data, outputting a second predictive result, comprises: constructing a first loss function based on the first prediction result set and the predicted value of the second prediction model, wherein the second prediction model performs optimization learning based on the first loss function; calculating the contribution weight of each prediction model in the at least one prediction model corresponding to the first prediction result set by using an attention mechanism; the second model after optimization learning is calculated based on the first data set and the contribution weight, and the second prediction result is output; and the linear transformation result of each layer of network of the second model is mapped in a nonlinear way based on a nonlinear enhancement activation function.
5. The method according to claim 1, wherein the method further comprises: The second prediction model regularly acquires the first data set to perform result prediction, outputs the second prediction result, and generates alarm information when the second prediction result is larger than a first threshold value; acquiring intermediate features of the first prediction model and the second prediction model at fixed time, performing periodic rolling prediction based on the intermediate features, and updating training parameters of the second prediction model; In response to a network outage, the first predictive model and the second predictive model are run based on the historic cached data.
6. A variable prediction apparatus, the apparatus comprising: a data acquisition unit for acquiring multi-factor data, and constructing a first data set based on the multi-factor data; the first prediction model unit is used for determining a corresponding first prediction model group based on the first data set, carrying out model training on the first data set based on the first prediction model group, and generating a first prediction result set; And the second prediction model unit is used for training a second prediction model based on the first prediction result set and the first data set and outputting a second prediction result.
7. The apparatus of claim 6, wherein the device comprises a plurality of sensors, The first prediction model unit is further used for determining at least one corresponding prediction model based on at least one data factor in the first data set corresponding to the multi-factor data; The second prediction model unit is further used for constructing a first loss function based on the first prediction result set and the predicted value of the second prediction model, the second prediction model is used for optimizing and learning based on the first loss function, the attention mechanism is used for calculating the contribution weight of each prediction model in the at least one prediction model corresponding to the first prediction result set, the second model after optimizing and learning is used for calculating based on the first data set and the contribution weight, the second prediction result is output, and the linear transformation result of each layer of network of the second model is subjected to nonlinear mapping based on a nonlinear enhancement activation function.
8. The apparatus of claim 6, wherein the device comprises a plurality of sensors, The data acquisition unit is also used for setting a window and a prediction step length, carrying out sliding window processing on the multi-factor data based on the window and the prediction step length to obtain multi-sample data, aligning the multi-sample data to a unified time axis and a space grid based on an interpolation algorithm based on a first time as a reference, carrying out space-time alignment processing, carrying out normalization and standardization processing on the multi-sample data subjected to space-time alignment, and embedding a modal identifier into each channel in the multi-sample data to obtain the first data set in response to the completion of the normalization and standardization processing; The second model prediction unit is further used for obtaining the first data set at regular time to conduct result prediction, outputting the second prediction result, generating alarm information when the second prediction result is larger than a first threshold value, obtaining intermediate features of the first prediction model and the second prediction model at regular time, conducting periodic rolling prediction based on the intermediate features, updating training parameters of the second prediction model, responding to network interruption, and operating the first prediction model and the second prediction model based on the history cache data.
9. An electronic device, comprising: At least one processor, and A memory communicatively coupled to the at least one processor, wherein, The memory stores instructions executable by the at least one processor to enable the at least one processor to perform the method of any one of claims 1-5.
10. A non-transitory computer readable storage medium storing computer instructions for causing a computer to perform the method of any one of claims 1-5.

Description

Variable prediction method, device, equipment and storage medium Technical Field The present disclosure relates to the field of data processing, and in particular, to a variable prediction method, apparatus, device, and storage medium. Background In the scenes of ocean, industry, agriculture, smart city and the like, accurate prediction of multi-factor driven time sequence variables (such as stormy waves, red tide indexes, storage temperature, air conditioner energy consumption, traffic flow, equipment vibration or any environment monitoring indexes) is important to decision making and control. For example, sea scenes comprise real-time prediction of sea wave height and red tide index (sea temperature, chlorophyll, air pressure and the like), and warehouse scenes comprise 24 h predictions of internal temperature and air conditioner energy consumption in advance (which are influenced by multivariable coupling of external temperature and humidity, internal heat source power, air circulation condition and the like). However, the existing numerical model, statistical model and machine learning method generally have the defects that nonlinear coupling and time-varying correlation among variables are ignored, so that dynamic evolution of a target variable cannot be completely described, the data integrity is highly sensitive, missing values or noise obviously reduce prediction accuracy and stability, the generalization capability is limited, and the method is difficult to adapt to new working conditions or variable combination changes which are not covered by training data. Disclosure of Invention The present disclosure provides a variable prediction method, apparatus, device, and storage medium, to at least solve the above technical problems in the prior art. According to a first aspect of the present disclosure, there is provided a variable prediction method, the method comprising: Acquiring multi-factor data, and constructing a first data set based on the multi-factor data; Determining a corresponding first set of predictive models based on the first dataset; Model training is carried out on the first data set based on the first prediction model group, and a first prediction result set is generated; and training a second prediction model based on the first prediction result set and the first data set, and outputting a second prediction result. In an embodiment, the determining a corresponding first set of prediction models based on the first data set includes: The multi-factor data corresponds to at least one data factor in the first data set, and corresponding at least one prediction model is determined based on the at least one data factor; The first set of prediction models is composed based on the at least one prediction model. In an embodiment, the constructing the first data set based on the multi-factor data includes: Setting a window and a prediction step length, and performing sliding window processing on the multi-factor data based on the window and the prediction step length to obtain multi-sample data; based on the first time as a reference, aligning the multi-sample data into a unified time axis and space grid based on an interpolation algorithm, and performing space-time alignment processing; And carrying out normalization and standardization processing on the space-time aligned multi-sample data, and embedding a mode identifier into each channel in the multi-sample data to obtain the first data set in response to the completion of the normalization and standardization processing. In an embodiment, the training the second prediction model based on the first prediction result set and the first data set, and outputting the second prediction result includes: constructing a first loss function based on the first prediction result set and the predicted value of the second prediction model, wherein the second prediction model performs optimization learning based on the first loss function; calculating the contribution weight of each prediction model in the at least one prediction model corresponding to the first prediction result set by using an attention mechanism; the second model after optimization learning is calculated based on the first data set and the contribution weight, and the second prediction result is output; and the linear transformation result of each layer of network of the second model is mapped in a nonlinear way based on a nonlinear enhancement activation function. In an embodiment, the method further comprises: The second prediction model regularly acquires the first data set to perform result prediction, outputs the second prediction result, and generates alarm information when the second prediction result is larger than a first threshold value; acquiring intermediate features of the first prediction model and the second prediction model at fixed time, performing periodic rolling prediction based on the intermediate features, and updating training parameters of the second predicti