CN115879569A - IoT observation data online learning method and system - Google Patents
IoT observation data online learning method and system Download PDFInfo
- Publication number
- CN115879569A CN115879569A CN202310214205.9A CN202310214205A CN115879569A CN 115879569 A CN115879569 A CN 115879569A CN 202310214205 A CN202310214205 A CN 202310214205A CN 115879569 A CN115879569 A CN 115879569A
- Authority
- CN
- China
- Prior art keywords
- data
- online
- data stream
- difference
- observation data
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 57
- 238000013136 deep learning model Methods 0.000 claims abstract description 41
- 230000008569 process Effects 0.000 claims abstract description 19
- 238000012545 processing Methods 0.000 claims abstract description 13
- 230000004927 fusion Effects 0.000 claims description 21
- 230000006870 function Effects 0.000 claims description 12
- 239000013598 vector Substances 0.000 description 9
- 238000010586 diagram Methods 0.000 description 6
- 238000005070 sampling Methods 0.000 description 6
- 238000012549 training Methods 0.000 description 6
- 238000013135 deep learning Methods 0.000 description 5
- 238000012544 monitoring process Methods 0.000 description 5
- 230000007246 mechanism Effects 0.000 description 4
- 238000004422 calculation algorithm Methods 0.000 description 3
- 230000008859 change Effects 0.000 description 3
- 238000013461 design Methods 0.000 description 3
- 230000000694 effects Effects 0.000 description 3
- 238000013473 artificial intelligence Methods 0.000 description 2
- 238000013528 artificial neural network Methods 0.000 description 2
- 230000015556 catabolic process Effects 0.000 description 2
- 230000007547 defect Effects 0.000 description 2
- 238000006731 degradation reaction Methods 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 230000006872 improvement Effects 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 230000000644 propagated effect Effects 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 238000012795 verification Methods 0.000 description 2
- 230000002411 adverse Effects 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000015572 biosynthetic process Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 238000004590 computer program Methods 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 230000001186 cumulative effect Effects 0.000 description 1
- 238000007405 data analysis Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000011478 gradient descent method Methods 0.000 description 1
- 238000007689 inspection Methods 0.000 description 1
- 230000007774 longterm Effects 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- YHXISWVBGDMDLQ-UHFFFAOYSA-N moclobemide Chemical compound C1=CC(Cl)=CC=C1C(=O)NCCN1CCOCC1 YHXISWVBGDMDLQ-UHFFFAOYSA-N 0.000 description 1
- 238000003062 neural network model Methods 0.000 description 1
- 238000007781 pre-processing Methods 0.000 description 1
- 238000010223 real-time analysis Methods 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 230000003068 static effect Effects 0.000 description 1
- 238000013179 statistical model Methods 0.000 description 1
- 230000008685 targeting Effects 0.000 description 1
Images
Landscapes
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention provides an online learning method and system of IoT observation data, which relate to the technical field of data processing and are characterized in that an online deep learning model is initialized according to acquired initial time sequence observation data; acquiring time sequence observation data generated by a sensor in real time, and forming an input data stream according to the time sequence observation data; processing an input data stream by an online deep learning model to generate a final prediction result; in the process of processing the input data stream, the input data stream is immediately learned, and an online deep learning model is dynamically updated in real time; the instant learning is to learn data distribution based on the mean value and variance of data flow, construct quasi-normal distribution, reconstruct a new sample, realize a variation attention network, and dynamically adjust a model based on distribution difference, reconstruction difference and reasoning difference; the method learns the hidden information among different hidden layers, improves the accuracy of model reasoning, and dynamically adjusts the parameters among the different hidden layers through online learning.
Description
Technical Field
The invention belongs to the technical field of data processing, and particularly relates to an IoT observation data online learning method and system.
Background
The statements in this section merely provide background information related to the present disclosure and may not necessarily constitute prior art.
In recent years, with the high-speed development of the Internet of Things (IoT), various types of sensors of the Internet of Things are becoming more and more popular; correspondingly, time series observation data generated by the sensor of the internet of things are exponentially increased, and real-time intelligent analysis of the data is particularly urgent and necessary in the application fields of intelligent environment monitoring, industrial monitoring, data center monitoring and the like.
With the rapid development of artificial intelligence, in the application field, knowledge needs to be accurately mined from data; deep Learning (DL) has a strong nonlinear modeling capability and the ability to extract abstract features layer by layer, with great success in modeling various data modalities, such as images, natural languages, speech, biological signals, etc.; therefore, a large number of deep learning models are continuously proposed and verified in the data modeling of the internet of things.
It should be noted that the deep learning model is mainly a static and basic parameter set, and after the deep learning model is trained offline, the parameters of the deep learning model are not changed; however, the traditional offline batch training and online reasoning solution mainly aims at internet of things data with small distribution variation, such as license plate recognition tasks; the scheme cannot meet the real-time analysis requirement of the data stream of the Internet of things with high sampling rate and unknown change; most of observation data of the internet of things are non-stationary data, the probability distribution of the observation data changes along with the time, and therefore the problem of concept drift is caused, and the concept drift refers to the fact that the data distribution changes along with the time, and therefore an original predictor or model cannot continuously and accurately predict the trend or attribute of new data. In the field of data stream prediction, concept drift is an important challenge because it affects the accuracy and effectiveness of the model.
In the case of concept drift, it is very important how to obtain relatively stable reasoning performance, for which some studies related to the concept drift problem are continuously discussed and explored; although various artificial intelligence based time series prediction models have been widely studied and applied, these models are unreliable for prediction of internet of things data streams with concept drift and can lead to overall performance degradation; how to effectively learn and infer from non-stationary data streams is still a great challenge, and particularly, how to adjust a model to adapt to concept drift is a problem which is concerned and tried to be solved by many researchers, in recent years, a method combining deep learning has a good effect on solving the problem, but the researches have certain limitations, and particularly, the researches only make limited assumptions on solving some types of concept drift modes; therefore, there is a lack of suitable methods and theories to effectively deal with the concept drift problem without affecting the accuracy and reliability of the final result.
As described above, the deep learning model trained offline cannot adapt well to the data flow with evolution distribution, and even if the model is updated periodically and new knowledge is "injected" into the model, the data between two adjacent updates must still face the problem of performance degradation; the higher the sampling rate of the data of the internet of things is, the more serious the problem of performance reduction when the model updating strategy is adopted is; furthermore, the expressive power of these models is limited by their predefined fixed structure; thus, the model will not be able to learn the extremely complex patterns that future data may have; the above problems in the current method will lead to cumulative bias decisions, and since the flow data analysis follows a one-pass processing scheme, each data point is processed only once without further inspection, such errors may be amplified continuously and cannot be repaired, which will cause chain reaction, and potentially significant adverse effects are generated on the monitoring systems of the internet of things such as smart oceans, smart cities, smart data centers and the like, and even the operation of the whole system is endangered.
For time series prediction, the main methods are roughly divided into two categories, namely statistical-based models and deep learning-based models. The statistical model is interpretable and the prediction of smoothed data is very good, provided that the time series data trends are consistent and do not change; such as ARIMA models and statistical machine learning methods GBDT, GBRT, etc., which require extensive data preprocessing and labor-intensive feature engineering and therefore cannot capture complex patterns in time series.
The emphasis of deep learning-based models in the field of time series data prediction is of increasing interest, for example, RNN-based models, seq2seq and LSNET have been shown to be effective in capturing short-term or long-term patterns; compared with the model, the recently proposed Transformer-based model shows more excellent performance in a long sequence, however, the Transformer-based model still needs to rely on abundant data to optimize specific parameters thereof, and cannot meet the accurate inference and prediction of stream data with concept drift; the method cannot effectively solve the concept drift problem, and cannot solve the cold start problem of the model.
Disclosure of Invention
In order to overcome the defects of the prior art, the invention provides the method and the system for online learning of the IoT observation data, so that the hidden information between different hidden layers is learned, the inference accuracy of the model is improved, and meanwhile, the parameters between different hidden layers are dynamically adjusted through online learning, the problems of concept drift and cold start are solved, so that the model can continuously and accurately learn and update the data along with the time, thereby better adapting to the continuously changing data characteristics and modes and ensuring that the performance and the accuracy of the model are always kept in the optimal state.
In order to achieve the above object, one or more embodiments of the present invention provide the following technical solutions:
the invention provides an online learning method of IoT observation data in a first aspect;
an online learning method of IoT observation data, comprising:
initializing an online deep learning model according to the obtained initial time sequence observation data;
acquiring time sequence observation data generated by a sensor in real time, and forming an input data stream according to the time sequence observation data;
the online deep learning model processes an input data stream to generate and output a final prediction result;
in the process of processing the input data stream, the input data stream is immediately learned, and an online deep learning model is dynamically updated in real time;
the instant learning is based on the mean value and the variance of the data flow, learning data distribution, constructing quasi-normal distribution, reconstructing a new sample, realizing a variation attention network, and dynamically adjusting the model based on distribution difference, reconstruction difference and reasoning difference.
Further, the online deep learning model includes an encoder, a decoder, and a result fusion module, and takes the data stream as input, and outputs the prediction of the data value at the next time.
Further, the encoder performs two operations:
(1) Extracting hidden information between hidden layers, inputting the hidden information into a result fusion module, and calculating a final predicted value;
(2) Targeting a standard normal distributionCoding hidden information, and outputting mean value satisfying quasi-normal distribution through reparameterization setting
And standard deviation->
。
Further, the decoder randomly selects samples from the quasi-normal distribution generated by the encoder, and outputs data consistent with the original characteristic structure of the data stream.
Further, the result fusion module performs weighted fusion on the plurality of features extracted by the encoder to obtain a current predicted value;
the weight is used for dynamically adjusting the contribution degree of each hidden layer, and the weight of each hidden layer is calculated in real time after each input sample is processed through the variation attention network.
Further, the variation attention network calculates the weight of each hidden layer according to the relationship between each hidden layer in the encoder and the necessary degree of each substructure.
Further, the loss function of the online deep learning model comprises:
reconstructing loss, measuring the difference between the true value and the reconstructed value;
KL loss, which measures the difference between the quasi-normal distribution and the standard normal distribution;
predicting loss, and measuring the difference between the true value and the current predicted value;
and measuring the difference between the final predicted value and the true value through the weighted fusion of the three losses.
A second aspect of the present invention provides an online learning system of IoT observation data.
An online learning system of IoT observation data comprises an initialization module, an acquisition module, a prediction module and a learning module:
an initialization module configured to: initializing an online deep learning model according to the acquired initial time sequence observation data;
an acquisition module configured to: acquiring time sequence observation data generated by a sensor in real time, and forming an input data stream according to the time sequence observation data;
a prediction module configured to: the online deep learning model processes an input data stream to generate and output a final prediction result;
a learning module configured to: in the process of processing the input data stream, the input data stream is immediately learned, and an online deep learning model is dynamically updated in real time;
the instant learning is based on the mean value and the variance of the data flow, learning data distribution, constructing quasi-normal distribution, reconstructing a new sample, realizing a variation attention network, and dynamically adjusting the model based on distribution difference, reconstruction difference and reasoning difference.
A third aspect of the present invention provides a computer-readable storage medium, on which a program is stored, which when executed by a processor, implements the steps in an online learning method of IoT observation data according to the first aspect of the present invention.
A fourth aspect of the present invention provides an electronic device, including a memory, a processor, and a program stored on the memory and executable on the processor, where the processor implements the steps in the method for online learning of IoT observation data according to the first aspect of the present invention when executing the program.
The above one or more technical solutions have the following beneficial effects:
the invention designs a self-adaptive online deep learning model which is used for learning a time series data stream of an Internet of things sensor in real time; the model can start from a shallow network, has higher convergence rate, has higher generalization capability on the Internet of things observation data with concept drift, can automatically evolve along with the change of observation time sequence data, can flexibly adjust the complexity and the representation capability of the model, effectively solves the problem of concept drift in the Internet of things observation data stream, and can be directly applied to process the data stream in a cold start mode; relevant verification is carried out on the data collected by the reference data set and the monitoring system, and the result shows that the real-time accuracy of the reasoning result can be obviously improved compared with the traditional method through the online deep learning model.
Advantages of additional aspects of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention.
Drawings
The accompanying drawings, which are incorporated in and constitute a part of this specification, are included to provide a further understanding of the invention, and are included to illustrate an exemplary embodiment of the invention and not to limit the invention.
FIG. 1 is a flow chart of the method of the first embodiment.
Fig. 2 is a structural diagram of an online deep learning model of the first embodiment.
Fig. 3 is a schematic diagram of an online learning operation mechanism according to a first embodiment.
Fig. 4 is a schematic diagram illustrating the evolution of weights of the layers of the inference network according to the first embodiment.
Fig. 5 is a system configuration diagram of the second embodiment.
Detailed Description
The invention is further described with reference to the following figures and examples.
In order to solve the concept drift problem without influencing the accuracy and reliability of a final result, a method for quickly and real-timely learning a data stream is needed, the method is generally called as 'online learning' or 'lifelong learning', the online learning is a learning mode of a time sequence data stream, most of the existing schemes have defects in training speed and reasoning speed, real-time performance cannot be really realized, and the effect of online learning is influenced. For deep learning models, its parameters can be simply updated periodically, however, this approach has the following disadvantages:
1. only the adjustment of different parameters among the online multiple tasks is focused on, the importance of the parameters of different hidden layers is ignored, and the model cannot be adjusted flexibly.
2. The complexity of the model cannot be adjusted, namely, the model selection cannot be carried out, and in an online learning environment, a model is needed to learn the data flow from shallow to deep, so that the problems of overfitting when the data volume is small (cold start stage) and underfitting when the data volume is large are solved.
Therefore, the invention designs an online learning method called as 'instant learning', which can learn hidden information between different hidden layers, meet the final reasoning accuracy of the model, simultaneously meet the dynamic adjustment of parameters between different hidden layers, and realize the real-time dynamic adjustment of the model.
The design concept of the invention follows the following existing online learning paradigm:
with respect to the time-series observation data,
generated by a sensor in real time, wherein T represents the number of data,x t the data representing the time at the t-th instant,Xrepresenting stream data with unbounded T. In an online learning environment, this time series is not available all at once, but is processed one after the other in the form of a stream, for each time t only historical data
Is accessible via a sensor, but can be determined by historical data +>
To predict a value of ^ greater or greater at time t>
And then observe and reveal the true valuesx t 。
Further, calculating the instantaneous loss of the current time t through the residual error between the real value and the predicted value; of course, for a particular application scenario, the time series
The value at the current time t can also be predicted in the form of a window; according to the predicted value>
Calculating the current instantaneous loss with the corresponding real value: />
The algorithm can meet and realize the online training of mass data by training each input data by using the loss and gradient iterative update generated by each input training sample.
The existing online learning framework can only realize data improvement, but the framework provided by the invention can also realize data generation and learn the distribution of potential variables; and learning the mean value and the variance of the quasi-normal distribution by constructing the quasi-normal distribution, further generating a new sample, and carrying out the following inference prediction based on a fusion module.
Example one
In one or more embodiments, a method for online learning of IoT observation data is disclosed, and fig. 1 is a flow chart of the method, as shown in fig. 1, comprising the following steps:
step S1: and initializing an online deep learning model according to the acquired initial time sequence observation data.
Step S2: and acquiring time sequence observation data generated by the sensor in real time, and forming an input data stream according to the time sequence observation data.
And step S3: the online deep learning model processes an input data stream to generate and output a final prediction result.
And step S4: in the process of processing the input data stream, the input data stream is subjected to instant learning, and an online deep learning model is dynamically updated in real time, wherein the instant learning is to learn data distribution based on the mean value and the variance of the data stream, construct quasi-normal distribution, reconstruct a new sample, realize a variation attention network, and dynamically adjust the model based on distribution difference, reconstruction difference and reasoning difference.
The following describes in detail an implementation procedure of the online learning method of IoT observation data according to this embodiment.
FIG. 2 is a block diagram of an online deep learning model, as shown in FIG. 2, which includes an encoder, a decoder, and a result fusion module, with a data stream as an input, outputting a prediction of a data value at a next time.
The encoder is connected with the result fusion module and used for predicting the data value, and the decoder is introduced and matched with the encoder and the result fusion module and used for dynamically updating the online deep learning model.
Encoder for encoding a video signal
Based on deep neural network construction, the method is composed of a plurality of hidden layers, and the maximum network capacity can be k +1 hidden layers in consideration of available computing power
The method carries out multi-scale learning in a data stream mode from sample to sample, and a learned hidden layer has two functions:
(1) Computing prediction vectors and attention weights for hidden layers
In order to fully utilize information in different hidden layers of an encoder, hidden information in each hidden layer is extracted to obtain k +1 prediction vectorsp i Inputting the prediction vector into a result fusion module, and obtaining a final prediction result through weighted fusion, wherein the prediction vectorp i The calculation formula of (2) is as follows:
wherein ,pred()is a feed-forward network that translates the hidden layer representation into local prediction,h i representing the ith hidden layer.
In order to realize dynamic adjustment of the contribution weight of each hidden layer in the final prediction, a variation attention network capable of meeting the above requirements is designed to learn the attention weight of the hidden layer, that is:
wherein ,attention()is a feed-forward network with a Softmax function, which is primarily responsible for computing the weights of each hidden layer, normalized to
The objective is to determine the relationship between each hidden layer in the overall model and how much the substructures are necessary in the final prediction.
In order to realize the iterative update of the model, an important technology of the heavy weight operation is realized based on the above, that is, the real-time dynamic adjustment of the weight is realized, and through the heavy weight operation, the weights of all hidden layers are recalculated and normalized after each input sample processing, which is completed after each iteration.
(2) Constructing a quasi-normal distribution
The encoder attempts to transcode the distribution of all the characteristic information carried by the sample into a quasi-normal distribution, which is referred to herein as a potential distribution, although it is usually not coded into a standard normal distribution (target distribution), although it is coded with the standard normal distribution as a target; in this process, the encoder directly outputs the mean value satisfying the quasi-normal distribution
And standard deviation->
As a result of the encoding, i.e. the mean value generated in the encoding process +>
And standard deviation->
Not calculated by mean or standard deviation definition, but output directly from the encoder.
To ensure mean value of encoder output
And standard deviation>
And (3) satisfying the standard normal distribution, setting a penalty term in the loss function, and once the fed-back quasi normal distribution (actual distribution) based on the mean and the standard deviation has a difference with the standard normal distribution (target distribution), further adjusting the model of the embodiment under the influence of a penalty mechanism.
Therefore, in the actual operation flow of the algorithm, the encoder is responsible for outputting the mean value and the standard deviation, the loss function ensures that the mean value and the standard deviation conform to a certain quasi-normal distribution, which is equivalent to encoding the original data to the quasi-normal distribution direction, outputting the mean value and the standard deviation of the quasi-normal distribution, and randomly sampling to generate the data
Subsequent decoder operations are performed.
At the same time, to guarantee sampling data
Is matched to the input sample, it is first assumed that there is a->
The posterior probability P (Z | X) for X, and further assume that this probability distribution is a normal distribution: />
Then the sample data is pick>
The probability distribution P (Z) of (A) is:
thus, P (Z) is a priori distributed and
the posterior distribution conforms to the standard normal distribution.
In addition, from a standard normal distribution
In a randomly selected sample->
And inputs the samples to a decoder. Since random sampling exists in the process, the gradient information flow in the framework is broken, and back propagation cannot be performed through the traditional chain rule, so that the back propagation of the model of the embodiment is designed to be completed by using a unique re-parameterization skill so as to meet the characteristics of a typical neural network. In reparameterization, this embodiment uses random sampling to generate a set of variables that satisfy a normal distribution, slave->
Middle sample z, equivalent to slave->
The middle of the samples is the epsilon, then make->
Therefore, a gradient descent method may be applied; finally, the decoder decodes z and outputs reconstructed data for sample x.
Thus, in the encoding stage of the encoder, each input sample is calculated
In (d) is based on the mean value>
And variance>
For constructing a quasi-normal distribution.
Decoder
Randomly selecting samples from the quasi-normal distribution generated by the encoder, and outputting data consistent with the original characteristic structure of the data stream, i.e. restoring the original input structure by the decoder to generate new data samples
The data sampled from the potential variable space is expected to >>
Follow the probability distribution of the original data X such that the sampled data->
New data generated
The probability distribution of the original data is also followed, and the larger the latent variable space of the embodiment is, the more information about the probability distribution can be kept relatively.
Result fusion module
The final prediction of the model is based on the prediction vector of each hidden layer
Performing weighted fusion to obtain a final prediction result, namely a prediction value at the next moment, and performing weighted fusion based on a variational attention mechanism, wherein the weighted fusion specifically comprises the following steps:
wherein ,
is predicted from the representation of the i-th hidden layer>
Is the weight of the i-th hidden layer,
is/>
The parameter (c) of (c).
The above is the formation of the online deep learning model, and after the online deep learning model is constructed, the online deep learning model is divided into two flows: real-time prediction and online learning.
Fig. 3 is a schematic diagram of online learning, as shown in fig. 3, in the process of processing samples of a data stream input into a linear deep learning model one by one, characteristics of the samples are continuously learned, and parameters of the model are updated through changes of the parameters, so as to realize dynamic real-time adjustment of the model, as shown in fig. 4, three hidden layers are set as verification findings through the embodiment, weights of the hidden layer-1, the hidden layer-2 and the hidden layer-3 are continuously changed along with the increase of the number of the samples of the data stream, and weights among the layers are also changed at different stages, which can better explain that the embodiment can dynamically and accurately capture changes when dealing with a concept drift problem; in the learning process, the loss and gradient generated by each input sample are updated iteratively, and each input sample is trained to meet and realize the online training of mass data, wherein the loss is quantified by a loss function, and the final total loss is calculated by using three losses: loss of reconstruction, loss of KL, and loss of prediction.
(1) Predicting loss
Unlike traditional neural network models, the prediction of the model is not the final representation from the last layer, and the final prediction vector is obtained by weighting the prediction vector obtained from each hidden layer to ensure absolute high accuracy.
Thus, for the observation of streaming IoT data, the prediction penalty Lt for the current prediction phase time t is:
wherein, y is the true value,
is a predicted value, is greater than or equal to>
Representing the predicted instantaneous loss function at time t.
The prediction loss function here may be any convex function, and the present embodiment uses a mean square loss function (MSE), where in the original backward propagation, the error derivative is backward propagated from the last output layer, unlike the original backward propagation, the error derivative is backward propagated from each predictor pj, such as:
wherein ,
parameters of the j-th layer network at time t and at time t + 1, respectively>
Indicates a renewal rate, is greater than or equal to>
Represents a gradient, <' > based on>
Indicates the predictor value of each layer, is>
Indicating a loss at the current time.
Calculating the gradient of each layer of parameters according to the equation, updating the parameters of the jth layer of the embodiment model in time t, and calculating the final prediction according to the gradient of each layer of parameters; it is worth to be noted that when the data sample is small, the model can realize fast convergence by fast learning the shallow network of the hidden information; using a variable attention mechanism may allow dynamic adjustment of different depth networks as data increases
Thus conforming to the scalability of the model capacity.
(2) Loss of KL
KL Loss (Kullback-Leibler Loss) is a measure of the difference between a quasi-normal distribution and a normal distribution, i.e. the mean
And variance->
The difference between them.
Given having a known mean value
And variance->
K, KL loss is defined as:
wherein ,
representing the mean and variance generated for each dimension.
(3) Loss of reconstruction
The reconstruction loss here uses the standard L2 loss, i.e., MSE, assuming that when there are n samples, there are the original value x and the reconstructed value
The reconstruction loss is defined as:
finally, to enhance the expressive power and prediction accuracy of the model, the total loss of this embodiment consists of the above three loss functions, with coefficients of
The method specifically comprises the following steps:
thus, at the end of each round, the algorithm will self-adjust according to the instantaneous loss of the next few rounds, which also solves the problem of model flexibility in the concept drift scenario.
And real-time prediction, namely inputting a data stream into an online deep learning model, obtaining a prediction vector corresponding to each hidden layer through the coding of the hidden layer of a coder, and performing weighted fusion on the prediction vectors to obtain a final prediction result, namely a prediction value at the next moment.
Example two
The embodiment discloses an online learning system of IoT observation data;
as shown in fig. 5, an online learning system of IoT observation data includes an initialization module, an acquisition module, a prediction module, and a learning module:
an initialization module configured to: initializing an online deep learning model according to the obtained initial time sequence observation data;
an acquisition module configured to: acquiring time sequence observation data generated by a sensor in real time, and forming an input data stream according to the time sequence observation data;
a prediction module configured to: the online deep learning model processes an input data stream to generate and output a final prediction result;
a learning module configured to: in the process of processing the input data stream, the input data stream is immediately learned, and an online deep learning model is dynamically updated in real time;
the instantaneous learning is based on the mean value and the variance of data flow, learning data distribution, constructing quasi-normal distribution, reconstructing a new sample, realizing a variation attention network, and dynamically adjusting a model based on distribution difference, reconstruction difference and reasoning difference.
EXAMPLE III
An object of the present embodiment is to provide a computer-readable storage medium.
A computer-readable storage medium, on which a computer program is stored, which when executed by a processor, implements the steps in an online learning method of IoT observation data according to one embodiment of the present disclosure.
Example four
An object of the present embodiment is to provide an electronic device.
An electronic device comprising a memory, a processor and a program stored on the memory and executable on the processor, wherein the processor when executing the program implements the steps of the method for online learning of IoT observation data according to the first embodiment of the present disclosure.
The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention, and various modifications and changes may be made by those skilled in the art. Any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.
Claims (10)
1. A method for online learning of IoT observation data, comprising:
initializing an online deep learning model according to the obtained initial time sequence observation data;
acquiring time sequence observation data generated by a sensor in real time, and forming an input data stream according to the time sequence observation data;
the online deep learning model processes an input data stream to generate and output a final prediction result;
in the process of processing the input data stream, the input data stream is immediately learned, and an online deep learning model is dynamically updated in real time;
the instantaneous learning is based on the mean value and the variance of data flow, learning data distribution, constructing quasi-normal distribution, reconstructing a new sample, realizing a variation attention network, and dynamically adjusting a model based on distribution difference, reconstruction difference and reasoning difference.
2. The method of claim 1, wherein the online deep learning model comprises an encoder, a decoder, and a result fusion module, and the data stream is used as input to output a prediction of a next-time data value.
3. The method of online learning of IoT observation data in accordance with claim 2, wherein the encoder performs two part operations:
(1) Extracting hidden information between hidden layers, inputting the hidden information into a result fusion module, and calculating a final predicted value;
(2) Coding hidden information by taking standard normal distribution as a target, and outputting a mean value meeting the quasi-normal distribution through reparameterization setting
And standard deviation->
。
4. The method of claim 2, wherein the decoder randomly selects samples from the quasi-normal distribution generated by the encoder and outputs data that is consistent with the original feature structure of the data stream.
5. The method of claim 2, wherein the result fusion module performs weighted fusion on the plurality of features extracted by the encoder to obtain the current predicted value;
the weight is used for dynamically adjusting the contribution degree of each hidden layer, and the weight of each hidden layer is calculated in real time after each input sample is processed through the variation attention network.
6. The method of claim 5, wherein the variational attention network calculates the weight of each hidden layer according to the relationship between each hidden layer in the encoder and the necessity degree of each substructure.
7. The method of online learning of IoT observation data recited in claim 1, wherein the loss function of the online deep learning model comprises:
reconstructing loss, measuring the difference between the true value and the reconstructed value;
KL loss, which measures the difference between the quasi-normal distribution and the standard normal distribution;
predicting loss, and measuring the difference between the true value and the current predicted value;
and measuring the difference between the final predicted value and the true value through the weighted fusion of the three losses.
8. An online learning system of IoT observation data is characterized by comprising an initialization module, an acquisition module, a prediction module and a learning module:
an initialization module configured to: initializing an online deep learning model according to the obtained initial time sequence observation data;
an acquisition module configured to: acquiring time sequence observation data generated by a sensor in real time, and forming an input data stream according to the time sequence observation data;
a prediction module configured to: the online deep learning model processes an input data stream to generate and output a final prediction result;
a learning module configured to: in the process of processing the input data stream, the input data stream is immediately learned, and an online deep learning model is dynamically updated in real time;
the instant learning is based on the mean value and the variance of the data flow, learning data distribution, constructing quasi-normal distribution, reconstructing a new sample, realizing a variation attention network, and dynamically adjusting the model based on distribution difference, reconstruction difference and reasoning difference.
9. An electronic device, comprising:
a memory for non-transitory storage of computer readable instructions; and
a processor for executing the computer readable instructions,
wherein the computer readable instructions, when executed by the processor, perform the method of any of claims 1-7.
10. A storage medium storing non-transitory computer-readable instructions, wherein the non-transitory computer-readable instructions, when executed by a computer, perform the instructions of the method of any one of claims 1-7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310214205.9A CN115879569B (en) | 2023-03-08 | 2023-03-08 | Online learning method and system for IoT observation data |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310214205.9A CN115879569B (en) | 2023-03-08 | 2023-03-08 | Online learning method and system for IoT observation data |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN115879569A true CN115879569A (en) | 2023-03-31 |
| CN115879569B CN115879569B (en) | 2023-05-23 |
Family
ID=85762015
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310214205.9A Active CN115879569B (en) | 2023-03-08 | 2023-03-08 | Online learning method and system for IoT observation data |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115879569B (en) |
Citations (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109102126A (en) * | 2018-08-30 | 2018-12-28 | 燕山大学 | One kind being based on depth migration learning theory line loss per unit prediction model |
| US20200160207A1 (en) * | 2018-11-15 | 2020-05-21 | General Electric Company | Automated model update based on model deterioration |
| CA3097655A1 (en) * | 2019-10-30 | 2021-04-30 | Royal Bank Of Canada | System and method for machine learning architecture with differential privacy |
| CN112836808A (en) * | 2021-03-22 | 2021-05-25 | 哈尔滨理工大学 | Time-space coupling learning method for deep learning-based distributed parameter system modeling |
| WO2021121354A1 (en) * | 2019-12-19 | 2021-06-24 | Beijing Didi Infinity Technology And Development Co., Ltd. | Model-based deep reinforcement learning for dynamic pricing in online ride-hailing platform |
| US20220036201A1 (en) * | 2020-08-02 | 2022-02-03 | Actimize Ltd. | Real drift detector on partial labeled data in data streams |
| CN114186668A (en) * | 2021-12-09 | 2022-03-15 | 山东省计算中心（国家超级计算济南中心） | Online deep learning method, system and equipment for Internet of things data flow prediction |
| CN114676622A (en) * | 2022-02-17 | 2022-06-28 | 国网冀北电力有限公司承德供电公司 | Short-term photovoltaic power prediction method based on self-encoder deep learning model |
| CN114881331A (en) * | 2022-05-09 | 2022-08-09 | 西安交通大学 | Learner abnormal learning state prediction method facing online education |
| CN114911831A (en) * | 2022-05-12 | 2022-08-16 | 大连大学 | Agricultural real-time early warning method and system based on Flink streaming calculation and deep learning |
| CN114970674A (en) * | 2022-04-15 | 2022-08-30 | 北京大学 | Time sequence data concept drift adaptation method based on relevance alignment |
| CN115293662A (en) * | 2022-10-10 | 2022-11-04 | 山东省计算中心（国家超级计算济南中心） | Parallel and distributed ocean observation data fusion intelligent calculation method and system |
| CN115731396A (en) * | 2022-11-28 | 2023-03-03 | 辽宁工程技术大学 | Continuous learning method based on Bayesian variation inference |
-
2023
- 2023-03-08 CN CN202310214205.9A patent/CN115879569B/en active Active
Patent Citations (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109102126A (en) * | 2018-08-30 | 2018-12-28 | 燕山大学 | One kind being based on depth migration learning theory line loss per unit prediction model |
| US20200160207A1 (en) * | 2018-11-15 | 2020-05-21 | General Electric Company | Automated model update based on model deterioration |
| CA3097655A1 (en) * | 2019-10-30 | 2021-04-30 | Royal Bank Of Canada | System and method for machine learning architecture with differential privacy |
| WO2021121354A1 (en) * | 2019-12-19 | 2021-06-24 | Beijing Didi Infinity Technology And Development Co., Ltd. | Model-based deep reinforcement learning for dynamic pricing in online ride-hailing platform |
| US20220036201A1 (en) * | 2020-08-02 | 2022-02-03 | Actimize Ltd. | Real drift detector on partial labeled data in data streams |
| CN112836808A (en) * | 2021-03-22 | 2021-05-25 | 哈尔滨理工大学 | Time-space coupling learning method for deep learning-based distributed parameter system modeling |
| CN114186668A (en) * | 2021-12-09 | 2022-03-15 | 山东省计算中心（国家超级计算济南中心） | Online deep learning method, system and equipment for Internet of things data flow prediction |
| CN114676622A (en) * | 2022-02-17 | 2022-06-28 | 国网冀北电力有限公司承德供电公司 | Short-term photovoltaic power prediction method based on self-encoder deep learning model |
| CN114970674A (en) * | 2022-04-15 | 2022-08-30 | 北京大学 | Time sequence data concept drift adaptation method based on relevance alignment |
| CN114881331A (en) * | 2022-05-09 | 2022-08-09 | 西安交通大学 | Learner abnormal learning state prediction method facing online education |
| CN114911831A (en) * | 2022-05-12 | 2022-08-16 | 大连大学 | Agricultural real-time early warning method and system based on Flink streaming calculation and deep learning |
| CN115293662A (en) * | 2022-10-10 | 2022-11-04 | 山东省计算中心（国家超级计算济南中心） | Parallel and distributed ocean observation data fusion intelligent calculation method and system |
| CN115731396A (en) * | 2022-11-28 | 2023-03-03 | 辽宁工程技术大学 | Continuous learning method based on Bayesian variation inference |
Non-Patent Citations (1)
| Title |
|---|
| 潘志松;唐斯琪;邱俊洋;胡谷雨;: "在线学习算法综述", 数据采集与处理 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN115879569B (en) | 2023-05-23 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Yu et al. | Probabilistic prediction of regional wind power based on spatiotemporal quantile regression | |
| CN109165664B (en) | Attribute-missing data set completion and prediction method based on generation of countermeasure network | |
| Sehovac et al. | Forecasting building energy consumption with deep learning: A sequence to sequence approach | |
| Ma et al. | A hybrid attention-based deep learning approach for wind power prediction | |
| CN108900346B (en) | Wireless network flow prediction method based on LSTM network | |
| CN112364975B (en) | Terminal running state prediction method and system based on graph neural network | |
| CN113177633B (en) | Depth decoupling time sequence prediction method | |
| CN114529051A (en) | Long-term power load prediction method based on hierarchical residual self-attention neural network | |
| CN114493014A (en) | Multivariate time series prediction method, multivariate time series prediction system, computer product and storage medium | |
| CN114021836B (en) | Multi-variable reservoir water inflow prediction system based on different angle fusion, training method and application | |
| CN114817773A (en) | Time sequence prediction system and method based on multi-stage decomposition and fusion | |
| CN116432697A (en) | Time sequence prediction method integrating long-term memory network and attention mechanism | |
| Basterrech et al. | Evolutionary Echo State Network: A neuroevolutionary framework for time series prediction | |
| WO2023179609A1 (en) | Data processing method and apparatus | |
| CN117424208A (en) | Probability photovoltaic power generation prediction method for comprehensive missing data | |
| CN115713044B (en) | Method and device for analyzing residual life of electromechanical equipment under multi-condition switching | |
| CN115879569A (en) | IoT observation data online learning method and system | |
| CN116485025A (en) | Dual-migration prediction model generation method and waste household appliance recovery amount prediction method | |
| CN112667394B (en) | Computer resource utilization rate optimization method | |
| CN114298411A (en) | GDP prediction method based on N-BEATS | |
| CN114239934A (en) | Power grid load sequence prediction method and system based on generative countermeasure network | |
| Li et al. | Mutual information variational autoencoders and its application to feature extraction of multivariate time series | |
| Zhan et al. | Learning Latent ODEs with Graph RNN for Multi-Channel Time Series Forecasting | |
| CN110648183A (en) | Grey correlation and QGNN-based resident consumption price index prediction method | |
| Ma et al. | A novel spatial–temporal generative autoencoder for wind speed uncertainty forecasting |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |