CN114781744A - Deep learning multi-step long radiance prediction method based on codec - Google Patents

Deep learning multi-step long radiance prediction method based on codec Download PDF

Info

Publication number
CN114781744A
CN114781744A CN202210492332.0A CN202210492332A CN114781744A CN 114781744 A CN114781744 A CN 114781744A CN 202210492332 A CN202210492332 A CN 202210492332A CN 114781744 A CN114781744 A CN 114781744A
Authority
CN
China
Prior art keywords
data
lstm
output
tcn
input
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.)
Pending
Application number
CN202210492332.0A
Other languages
Chinese (zh)
Inventor
谢利萍
童俊龙
张晗津
张侃健
魏海坤
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Southeast University
Original Assignee
Southeast University
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Southeast University filed Critical Southeast University
Priority to CN202210492332.0A priority Critical patent/CN114781744A/en
Publication of CN114781744A publication Critical patent/CN114781744A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/044—Recurrent networks, e.g. Hopfield networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Computational Linguistics (AREA)
  • Health & Medical Sciences (AREA)
  • Biophysics (AREA)
  • Artificial Intelligence (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Biomedical Technology (AREA)
  • Business, Economics & Management (AREA)
  • Economics (AREA)
  • Human Resources & Organizations (AREA)
  • Strategic Management (AREA)
  • Development Economics (AREA)
  • Game Theory and Decision Science (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Marketing (AREA)
  • Operations Research (AREA)
  • Quality & Reliability (AREA)
  • Tourism & Hospitality (AREA)
  • General Business, Economics & Management (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The invention relates to a deep learning multi-step long-radiance prediction method based on a codec, and belongs to the technical field of photovoltaic power generation. The prediction method comprises the following steps: s1, acquiring training data, acquiring historical irradiance data of the target area and meteorological data corresponding to the historical irradiance data, and making a supervision data set; s2, preprocessing data, including meteorological information characteristic coding and data normalization; s3, training a coder-decoder model, wherein the coder model is composed of a TCN and LSTM cascade structure, and the decoder is composed of an LSTM and MLP cascade structure; by reading the current time period t0～tNIrradiance as supervisory information, and t0Historical irradiance and meteorological information before the moment are used as input data to train a coder-decoder model; s4, predicting, inputting the historical data into the coding solution obtained by training in the step S3And the encoder model predicts the solar irradiance of multiple steps in the future. The method can fully utilize the historical information of the irradiance sequence, and experiments show that the method can effectively improve the accuracy of multi-step long-irradiance prediction.

Description

Deep learning multi-step long radiance prediction method based on codec
Technical Field
The invention relates to a deep learning multistep long radiance prediction method based on a codec, and belongs to the technical field of photovoltaic power generation.
Background
Solar energy resources are the most promising renewable energy source, and a survey by the international renewable energy agency shows that as of 2020, 29% of the global electricity production comes from renewable energy sources, with the renewable energy power generation part solar energy accounting for 26.77% and rising year by year. However, due to uncertainty and intermittency of irradiance, photovoltaic power generation presents considerable instability, which increases grid connection and scheduling difficulty of photovoltaic power generation, and restricts wide application of solar energy resources.
At present, there are various irradiance prediction methods based on deep learning, such as irradiance prediction using models such as LSTM and CNN. Some scholars are dedicated to research on mixed prediction methods of RNN models and CNN models, such as LSTM-CNN combined prediction models. However, the CNN model is better at extracting spatial features, and has limited ability to extract time-dependent features; RNN models can maintain timing dependence but are difficult to handle long input sequences. The characteristics show that the deep learning model based on the RNN and the CNN is difficult to take long-sequence input and long-term dependence into consideration, and the model needs to be improved.
Disclosure of Invention
The technical problem is as follows:
the technical problem to be solved by the invention is as follows: the existing deep learning model based on RNN and CNN is difficult to give consideration to long-sequence input and long-term dependence, so that the irradiance prediction is difficult to carry out by fully utilizing historical information in a prediction model, and the model precision is difficult to guarantee after the prediction step length is increased.
In order to solve the technical problem, the invention provides a deep learning multi-step long irradiance prediction method based on a codec, so that irradiance prediction is performed by fully utilizing historical information, and the prediction effect is improved.
The technical scheme is as follows:
the invention provides a deep learning multi-step long radiance prediction method based on a codec, which comprises the following steps:
1. a deep learning multi-step long radiance prediction method based on a codec is characterized by comprising the following steps:
s1, acquiring training data, acquiring historical irradiance data of a target area and meteorological data corresponding to the historical irradiance data, and manufacturing a supervision data set according to a prediction task;
s2, preprocessing data, including weather information feature coding and data normalization;
s3, training the code decoder model, using the reading current time interval t0～tNIrradiance as supervisory information, and t0Historical irradiance and meteorological information before the moment are used as input data, and a codec model is trained;
and S4, forecasting, inputting the historical data into the codec model obtained by training S3, and forecasting future multi-step solar irradiance.
Further, the step S1 includes the following steps:
(1) obtaining historical irradiance data of a target area and corresponding meteorological data (including but not limited to temperature, humidity, air pressure, wind speed and the like);
(2) if a certain segment of the historical data is missing or illegal, the average value of the adjacent data before and after the historical data is used for substitution, so that the continuity and the authenticity of the data are ensured, and the quality of the training data is ensured;
(3) matching the supervision information to make a supervision data set by reading the current time period t0～tNThe irradiance is used as supervision information, the historical irradiance and the corresponding meteorological information are matched to be used as input information, and a supervision data set is manufactured.
Further, the step S2 specifically includes the following steps:
(1) encoding meteorological information corresponding to irradiance, encoding weather types by using a one-hot encoding mode, and taking numerical values as encoding values for numerical information;
(2) in order to ensure that the model gradient changes reasonably in the training process, normalization operation needs to be performed on input data, and a normalization formula is as follows:
Figure BDA0003632142090000021
wherein a represents a feature in the data set, a' represents a normalized feature value, amaxAnd aminRespectively representing the maximum and minimum values of the feature in the historical data.
Further, the codec structure in step S3 is as follows:
(1) the encoder is composed of a series structure of a time convolution network TCN and a long-short term memory network LSTM, the TCN is responsible for acquiring long sequence input and keeping time sequence dependency relationship, and the compressed short sequence keeps time sequence dependency through the LSTM; specifically, the encoder firstly receives a long sequence input through the TCN, the network layer number depends on the length of the input sequence, secondly, the feature sequence extracted by the TCN is compressed and output to the LSTM, and finally, the output of the LSTM is used as the encoding output of the encoder;
(2) the decoder consists of a series structure of a long-short term memory network LSTM and a multilayer perceptron MLP, the LSTM is used for obtaining multi-step length prediction in a time circulation mode, and the performance of multi-step output is balanced through a loss function; the decoder first receives the output of the encoder, is responsible for decoding by the LSTM, and outputs to the MLP after decoding, where the MLP is used to match the output dimensions.
Further, the codec is characterized as follows:
(1) the coder is formed by cascading TCN and LSTM; the TCN receives long sequence input, obtains an output sequence after multi-layer characteristic extraction, and intercepts the rear section of the sequence as the input of the LSTM;
(2) the decoder is formed by LSTM and MLP cascade; LSTM receives the encoder state, MLP is used for dimension matching of LSTM output;
(3) each layer of the TCN is composed of a TCN residual block, and the TCN residual block comprises a series module and a residual connection; the series module consists of two groups of identical void cause and effect convolution layers, a weight normalization layer, a ReLu activation unit and a discarding layer;
(4) balancing the prediction performance of the multi-step output by designing a loss function, wherein the loss function is designed as follows:
Figure BDA0003632142090000031
where K represents the number of predicted steps, lossiRepresents the output loss of the ith step, αiExpress lossiW represents the parameters of the model and β represents the regularization coefficients.
Further, the TCN and LSTM operation process of the codec includes:
(1) the hole causal convolution operation is described by:
Figure BDA0003632142090000032
wherein x represents a convolution operation, d represents a hole coefficient, x represents an input sequence, s represents an element of the sequence, f represents a convolution kernel, k represents the size of the convolution kernel, s-d · i represents an element selected by the hole convolution, and f(s) represents an output of the hole causal convolution;
(2) TCN residual block:
O1＝dropout(ReLU(Norm(F(s))))，
O2＝dropout(ReLU(Norm(O1)))，
Otcn＝s+O2，
where Norm denotes weight normalization, ReLU denotes activation function, and dropout denotes a discard layer; o istcnRepresents the output of the residual block of TCN, i.e., the output of each layer of TCN;
(3) the LSTM includes:
forgetting door ft：
ft＝sigmoid(Wifxt+bif+Whfht-1+bhf)，
Input door it：
it＝sigmoid(Wiixt+bii+Whiht-1+bhi)，
Activation function gt：
gt＝tanh(Wiixt+bii+Whiht-1+bhi)，
Output gate ot：
ot＝sigmoid(Wioxt+bio+Whoht-1+bho)，
The current time corresponds to the state c of the memory cellt：
Figure BDA0003632142090000033
LSTM output State ht：
Figure BDA0003632142090000041
Wherein WifAnd bifWeight matrix and bias matrix respectively representing the external input of a forgetting gate, WhfAnd bhfWeight matrix and bias matrix, W, representing respectively the forgetting gate hidden state inputiiAnd biiWeight matrix and bias matrix, W, respectively representing external inputs to the input gatehiAnd bhiWeight matrix and bias matrix, W, representing the hidden state input of the input gate, respectivelyioAnd bioWeight matrix and bias matrix, W, representing external inputs to the output gates, respectivelyhoAnd bhoWeight matrix and bias matrix respectively representing hidden state inputs of output gates, ft、itAnd otOutput of the forgetting gate, the input gate and the output gate at the time t, htHidden state at time t, xtIs an external input.
Has the advantages that:
the encoder part of the invention acquires long sequence input by utilizing the characteristic that the TCN has a long receptive field, intercepts short sequence input LSTM after TCN output, and acquires longer historical information under the condition of short sequence input; the encoder has the capacity of inputting long sequences and maintaining time characteristic time sequence dependency, and is more suitable for characteristic extraction of time sequences; the decoder structure can ensure that the model outputs a plurality of prediction steps in sequence, and the prediction performance of multi-step output is balanced through a loss function. On the premise of only depending on historical information, the structure of the coder-decoder can obtain the irradiance prediction effect with high precision.
Drawings
FIG. 1 is a flow chart of irradiance prediction of the present invention;
FIG. 2 is a block diagram of a codec of the present invention;
FIG. 3 is a schematic diagram of the TCN residual block inside the codec according to the present invention;
FIG. 4 is a schematic diagram of the elements of an internal LSTM module of the codec of the present invention;
FIG. 5 is a graph showing the results of the present invention.
Detailed Description
In order to more clearly illustrate the technical solution of the present invention, the present invention is described below with reference to the accompanying drawings. The examples are given solely for the purpose of illustration and are not intended to limit the scope of the invention.
Referring to fig. 1, a codec-based deep learning multi-step long radiance prediction method includes the following steps:
s1, acquiring training data, acquiring historical irradiance data of a target area and meteorological data corresponding to the historical irradiance data, and making a supervision data set, wherein S1 comprises the following contents:
(1) obtaining historical irradiance data of a target area and corresponding meteorological data thereof, and selecting temperature T, humidity H, air pressure P, wind speed W and the like as the meteorological data in the embodiment;
(2) if a certain segment of the historical data is missing or illegal, the average value of the adjacent data before and after the historical data is used for substitution, so that the continuity and the authenticity of the data are ensured, and the quality of the training data is ensured;
(3) matching supervisory information to produce supervisory numbersBy reading the current time period t0～tNIrradiance x of0～xNAnd as the monitoring information of multi-step length prediction, matching the historical irradiance with the corresponding meteorological information as input information to manufacture a monitoring data set. For example: the first 24 moments of the current moment are selected as historical information input models, irradiance of 6 moments in the future is predicted, and the data set can be represented as ([ x ]-24，T-24，W-24，P-24，H-24，...，x-1，T-1，W-1，P-1，H-1]；[x0，...，x5]) Wherein [ x ]0，...，x5]Denotes supervision information, [ x ]-t，T-t，W-t，P-t，H-t]Irradiance and weather information for t moments before the current moment are represented.
S2, preprocessing data, including meteorological information feature coding and data normalization, specifically including the following contents:
(1) the weather information corresponding to the irradiance is coded, the weather type is coded by utilizing a one-hot coding mode, and the numerical value is used as a coding value for the numerical value information;
(2) in order to ensure that the model gradient changes reasonably in the training process, normalization operation needs to be performed on input data, and a normalization formula is as follows:
Figure BDA0003632142090000051
wherein a represents a feature in the data set, a' represents a normalized feature value, amaxAnd aminRespectively representing the maximum value and the minimum value of the characteristic in historical data; the characteristics that need to be normalized in this embodiment are historical irradiance information and historical meteorological information, where the historical meteorological information includes temperature, humidity, wind speed, and barometric pressure.
S3, training a code decoder model;
the RNN model is easy to have gradient disappearance under the condition of long sequence input and time sequence dependency relationship disappearance, the CNN has a large receptive field, but due to structural limitation, time sequence dependency characteristics are difficult to extract fully, namely, the deep learning prediction model based on the RNN and the CNN is difficult to take both long sequence input and long-term time sequence dependency relationship into account, so that the model is difficult to fully utilize historical information to carry out irradiance prediction, and the model precision is difficult to ensure after the prediction step length is increased;
in order to fully utilize the useful information of the historical sequence, the present embodiment considers both long sequence input and long-term timing dependency relationship according to the codec structure proposed by the present invention, and reads the current time period t0～tNIrradiance as supervisory information, and t0Historical irradiance and meteorological information before the moment are used as input data, and a codec model is trained; referring to fig. 2, the codec described in S3 includes the following:
the encoder is composed of a series structure of a time convolution network TCN and a long-short term memory network LSTM, the TCN is responsible for obtaining long sequence input and keeping a time sequence dependency relationship, and the compressed short sequence keeps time sequence dependency after passing through the LSTM; specifically, the encoder receives a long sequence input through the TCN, the network layer number depends on the length of the input sequence, then the feature sequence extracted by the TCN is compressed and output to the LSTM, and finally the output of the LSTM is used as the encoding output of the encoder;
the decoder consists of a series structure of a long-short term memory network LSTM and a multilayer perceptron MLP, the LSTM is used for obtaining multi-step length prediction in a time circulation mode, and the performance of multi-step output is balanced through a loss function; the decoder first receives the output of the encoder, is responsible for decoding by the LSTM, and outputs to the MLP after decoding, where the MLP is used to match the output dimensions.
Further, the codec in step S3 is characterized as follows:
(1) the encoder is formed by cascading TCN and LSTM; the TCN receives long sequence input, obtains an output sequence after multi-layer characteristic extraction, and intercepts the rear segment of the sequence as the input of the LSTM;
(2) the decoder is formed by cascading LSTM and MLP; LSTM receives the encoder state, MLP is used for dimension matching of LSTM output;
(3) each layer of the TCN is composed of a TCN residual block, and the TCN residual block comprises a serial module and residual connection; the series module consists of two groups of identical void cause and effect convolution layers, a weight normalization layer, a ReLu activation unit and a discarding layer;
(4) balancing the prediction performance of the multi-step output by designing a loss function, wherein the loss function is designed as follows:
Figure BDA0003632142090000061
where K represents the number of predicted steps, lossiRepresents the output loss of the ith step, αiExpress lossiW represents a parameter of the model, β represents a regularization coefficient; in the present embodiment, the outputs at different times are regarded as being equally important, and therefore the weight coefficient αiThe values are the same.
Further, the structure of the TCN residual block and LSTM module in step S3 is as shown in fig. 3 and fig. 4, which specifically includes:
(1) the hole causal convolution operation is described by:
Figure BDA0003632142090000062
wherein x represents convolution operation, d represents a hole coefficient, x represents an input sequence, s represents an element of the sequence, f represents a convolution kernel, k represents convolution kernel size, s-d · i represents an element of hole convolution selection, and f(s) represents hole causal convolution output;
(2) TCN residual block:
O1＝dropout(ReLU(Norm(F(s))))，
O2＝dropout(ReLU(Norm(O1)))，
Otcn＝s+O2，
where Norm denotes weight normalization, ReLU denotes activation function, and dropout denotes a discard layer; o istcnRepresents the output of the TCN residual block, i.e., the output of each layer of the TCN;
(3) the LSTM includes:
a forgetting gate for discarding unimportant information; forgetting the output f of the door during the forgetting stagetThe sigmoid is calculated by an activation function, whether the information of the previous moment is discarded or not is determined by the output value of the activation function, and the calculation formula of the forgetting gate is as follows:
ft＝sigmoid(Wifxt+bif+Whfht-1+bhf)，
the input gate and the activation function jointly complete memory selection, and the input gate and the activation function perform matrix multiplication to determine which values are stored to the current state;
the calculation formula of the input gate it is as follows:
it＝sigmoid(Wiixt+bii+Whiht-1+bhi)，
activation function gtThe calculation formula of (a) is as follows:
gt＝tanh(Wiixt+bii+Whiht-1+bhi)，
the current time corresponds to the state c of the memory celltThe input gate and the forgetting gate jointly determine, the matrix multiplication of the forgetting gate and the last time state represents discarding some unnecessary information, the matrix multiplication of the input gate and the activation function represents storing important information, and the calculation formula is as follows:
Figure BDA0003632142090000071
output gate ot：
ot＝sigmoid(Wioxt+bio+Whoht-1+bho)，
The output gate determines the output value of the memory cell state at the current time, thereby obtaining the output state h of LSTMtThe calculation formula is as follows:
Figure BDA0003632142090000072
wherein WifAnd bifWeight matrix and bias matrix respectively representing the external input of a forgetting gate, WhfAnd bhfWeight matrix and bias matrix, W, representing respectively the forgetting gate hidden state inputiiAnd biiWeight matrix and bias matrix, W, respectively representing external inputs to the input gatehiAnd bhiWeight matrix and bias matrix, W, representing hidden state inputs to the input gate, respectivelyioAnd bioWeight matrix and bias matrix, W, representing external inputs to the output gates, respectivelyhoAnd bhoWeight matrix and bias matrix respectively representing hidden state inputs of output gates, ft、itAnd otRespectively the output of the forgetting gate, the input gate and the output gate at the moment t, htFor the hidden state at time t, xtIndicating an external input.
And S4, forecasting, inputting the historical data into the codec model obtained by the training of S3, and forecasting the future multi-step solar irradiance.
The codec model created in this embodiment selects the following parameters:
selecting four layers of TCN and two layers of LSTM as encoders, and selecting the input length of the TCN as one time of the input length of the LSTM, namely, the second half of the output sequence of the TCN layer is used as the input of the LSTM layer;
a single layer LSTM and two layers MLP are chosen as decoders, where the number of LSTM-MLP concatenations is consistent with the prediction step size.
In order to verify the prediction performance of the method, irradiance data of American measurement and instrument laboratories are selected, and an oak ridge is specifically selected as the location of a data set. The training set length is 4 years, the test set length is 1 year, and compared with the traditional LSTM and TCN methods, the method takes one-step prediction in advance as an example, and the specific error prediction is shown in Table 1 and attached figure 5;
TABLE 1 comparison of conventional LSTM and TCN processes with the present process
RMSE MAE nRMSE
LSTM 55.47 25.78 36.23％
TCN 55.16 28.03 36.03％
Method for producing a composite material 54.32 23.68 35.46％
The method of the invention has smaller prediction error than the traditional LSTM and TCN in each evaluation; therefore, the method has higher prediction precision.

Claims (6)

1. A deep learning multi-step long radiance prediction method based on a codec is characterized by comprising the following steps:
s1, acquiring training data, acquiring historical irradiance data of a target area and meteorological data corresponding to the historical irradiance data, and manufacturing a supervision data set according to a prediction task;
s2, preprocessing data, including meteorological information characteristic coding and data normalization;
s3, training the code decoder model, using the reading current time interval t0～tNIrradiance as supervisory information, and t0Historical irradiance and meteorological information before the moment are used as input data, and a codec model is trained;
and S4, forecasting, inputting historical data into the codec model obtained by training in the step S3, and forecasting future multi-step solar irradiance.
2. The prediction method according to claim 1, wherein the step S1 includes the following steps:
(1) obtaining historical irradiance data of a target area and corresponding meteorological data including but not limited to temperature, humidity, air pressure and wind speed;
(2) if a certain segment of the historical data is missing or illegal, the average value of the adjacent data before and after the historical data is used for substitution, so that the continuity and the authenticity of the data are ensured, and the quality of the training data is ensured;
(3) matching the supervision information to make a supervision data set by reading the current time period t0～tNThe irradiance is used as supervision information, the historical irradiance and the corresponding meteorological information are matched to be used as input information, and a supervision data set is manufactured.
3. The prediction method according to claim 1, wherein the step S2 specifically includes the following contents:
(1) the weather information corresponding to the irradiance is coded, the weather type is coded by utilizing a one-hot coding mode, and the numerical value is used as a coding value for the numerical value information;
(2) in order to ensure that the model gradient changes reasonably in the training process, normalization operation needs to be carried out on input data, and a normalization formula is as follows:
Figure FDA0003632142080000011
wherein a represents the feature in the data set, a' represents the normalized feature value, amaxAnd aminRespectively representing the maximum and minimum values of the feature in the historical data.
4. The prediction method according to claim 1, wherein the codec of step S3 comprises an encoder and a decoder, and wherein:
the encoder is composed of a series structure of a time convolution network TCN and a long-short term memory network LSTM, the TCN is responsible for acquiring long sequence input and keeping time sequence dependency relationship, and the compressed short sequence keeps time sequence dependency through the LSTM; specifically, the encoder receives a long sequence input through the TCN, the network layer number depends on the length of the input sequence, then the feature sequence extracted by the TCN is compressed and output to the LSTM, and finally the output of the LSTM is used as the encoding output of the encoder;
the decoder consists of a series structure of a long-short term memory network LSTM and a multilayer perceptron MLP, and obtains multi-step length prediction by using the LSTM in a time cycle manner, and balances the performance of multi-step output through a loss function; the decoder first receives the output of the encoder, is responsible for decoding by the LSTM, and outputs to the MLP after decoding, where the MLP is used to match the output dimension.
5. The prediction method of claim 4, wherein the codec is characterized by:
(1) the encoder is formed by cascading TCN and LSTM; the TCN receives long sequence input, obtains an output sequence after multi-layer characteristic extraction, and intercepts the rear section of the sequence as the input of the LSTM;
(2) the decoder is formed by cascading LSTM and MLP; LSTM receives the encoder state, MLP is used for the dimension matching of LSTM output;
(3) each layer of the TCN is composed of a TCN residual block, and the TCN residual block comprises a serial module and residual connection; the series module consists of two groups of identical void cause-and-effect convolution layers, a weight normalization layer, a ReLu activation unit and a discarding layer;
(4) balancing the prediction performance of the multi-step output by designing a loss function, wherein the loss function is designed as follows:
Figure FDA0003632142080000021
where K represents the predicted number of steps, lossiRepresents the output loss of the ith step, αiExpress lossiW represents a parameter of the model and β represents a regularization coefficient.
6. The prediction method of claim 4, wherein the TCN and LSTM operations of the codec comprise:
(1) the hole causal convolution operation is described by:
Figure FDA0003632142080000022
wherein x represents a convolution operation, d represents a hole coefficient, χ represents an input sequence, s represents an element of the sequence, f represents a convolution kernel, k represents the size of the convolution kernel, s-d · i represents an element selected by the hole convolution, and f(s) represents an output of the hole causal convolution;
(2) TCN residual block:
O1＝dropout(ReLU(Norm(F(s))))，
O2＝dropout(ReLU(Norm(O1)))，
Otcn＝s+O2，
where Norm denotes weight normalization, ReLU denotes activation function, and dropout denotes a discard layer; o istcnRepresents the output of the residual block of TCN, i.e., the output of each layer of TCN;
(3) the LSTM includes:
forgetting door ft：
ft＝sigmoid(Wifxt+bif+Whfht-1+bhf)，
Input gate it：
it＝sigmoid(Wiixt+bii+Whiht-1+bhi)，
Activation function gt：
gt＝tanh(Wiixt+bii+Whiht-1+bhi)，
Output gate ot：
ot＝sigmoid(Wioxt+bio+Whoht-1+bho)，
The current time corresponds to the state c of the memory cellt：
Figure FDA0003632142080000031
LSTM output State ht：
Figure FDA0003632142080000032
Wherein WifAnd bifWeight matrix and bias matrix, W, respectively representing the external input of a forgetting gatehfAnd bhfWeight matrix and bias matrix, W, representing respectively the forgetting gate hidden state inputiiAnd biiWeight matrix and bias matrix, W, respectively representing external inputs to the input gatehiAnd bhiWeight matrix and bias matrix, W, representing the hidden state input of the input gate, respectivelyioAnd bioWeight matrix and bias matrix, W, representing external inputs to the output gates, respectivelyhoAnd bhoWeight matrix and bias matrix respectively representing hidden state inputs of output gates, ft、itAnd otOutput of the forgetting gate, the input gate and the output gate at the time t, htHidden state at time t, xtIs an external input.
CN202210492332.0A 2022-05-07 2022-05-07 Deep learning multi-step long radiance prediction method based on codec Pending CN114781744A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202210492332.0A CN114781744A (en) 2022-05-07 2022-05-07 Deep learning multi-step long radiance prediction method based on codec

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202210492332.0A CN114781744A (en) 2022-05-07 2022-05-07 Deep learning multi-step long radiance prediction method based on codec

Publications (1)

Publication Number Publication Date
CN114781744A true CN114781744A (en) 2022-07-22

Family

ID=82434237

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202210492332.0A Pending CN114781744A (en) 2022-05-07 2022-05-07 Deep learning multi-step long radiance prediction method based on codec

Country Status (1)

Country Link
CN (1) CN114781744A (en)

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115618714A (en) * 2022-09-09 2023-01-17 北京思源知行科技发展有限公司 Solar radiation probability forecasting method and related equipment
CN115951430A (en) * 2023-03-15 2023-04-11 数字太空(北京)智能技术研究院有限公司 Geomagnetic index forecasting method based on LSTM
CN116991431A (en) * 2023-08-04 2023-11-03 沐曦集成电路(杭州)有限公司 GPU-based coding and decoding model static deployment method, electronic equipment and medium
CN117272002A (en) * 2023-11-23 2023-12-22 中国电建集团西北勘测设计研究院有限公司 Solar radiation amount estimation method and device, electronic equipment and storage medium

Cited By (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115618714A (en) * 2022-09-09 2023-01-17 北京思源知行科技发展有限公司 Solar radiation probability forecasting method and related equipment
CN115951430A (en) * 2023-03-15 2023-04-11 数字太空(北京)智能技术研究院有限公司 Geomagnetic index forecasting method based on LSTM
CN116991431A (en) * 2023-08-04 2023-11-03 沐曦集成电路(杭州)有限公司 GPU-based coding and decoding model static deployment method, electronic equipment and medium
CN116991431B (en) * 2023-08-04 2024-03-01 沐曦集成电路(杭州)有限公司 GPU-based coding and decoding model static deployment method, electronic equipment and medium
CN117272002A (en) * 2023-11-23 2023-12-22 中国电建集团西北勘测设计研究院有限公司 Solar radiation amount estimation method and device, electronic equipment and storage medium
CN117272002B (en) * 2023-11-23 2024-02-20 中国电建集团西北勘测设计研究院有限公司 Solar radiation amount estimation method and device, electronic equipment and storage medium

Similar Documents

Publication Publication Date Title
CN114781744A (en) Deep learning multi-step long radiance prediction method based on codec
CN111899510A (en) Intelligent traffic system flow short-term prediction method and system based on divergent convolution and GAT
CN111860982A (en) Wind power plant short-term wind power prediction method based on VMD-FCM-GRU
CN111815033A (en) Offshore wind power prediction method based on RCNN and meteorological time sequence characteristics
CN110222901A (en) A kind of electric load prediction technique of the Bi-LSTM based on deep learning
CN112733444A (en) Multistep long time sequence prediction method based on CycleGAN neural network
CN111861013B (en) Power load prediction method and device
CN111027772A (en) Multi-factor short-term load prediction method based on PCA-DBILSTM
CN111191856A (en) Regional comprehensive energy system multi-energy load prediction method considering time sequence dynamic characteristics and coupling characteristics
CN115409258A (en) Hybrid deep learning short-term irradiance prediction method
CN114462718A (en) CNN-GRU wind power prediction method based on time sliding window
CN114611792A (en) Atmospheric ozone concentration prediction method based on mixed CNN-Transformer model
CN113128113A (en) Poor information building load prediction method based on deep learning and transfer learning
Li et al. Deep spatio-temporal wind power forecasting
CN114218872A (en) Method for predicting remaining service life based on DBN-LSTM semi-supervised joint model
CN113705915A (en) CNN-LSTM-ARIMA-based combined short-term power load prediction method
CN112508286A (en) Short-term load prediction method based on Kmeans-BilSTM-DMD model
CN110458341B (en) Ultra-short-term wind power prediction method and system considering meteorological characteristics
CN113449919B (en) Power consumption prediction method and system based on feature and trend perception
CN113642255A (en) Photovoltaic power generation power prediction method based on multi-scale convolution cyclic neural network
CN117498296A (en) Multi-wind power plant power prediction method based on attention space-time synchronization diagram convolution network
CN115481788B (en) Phase change energy storage system load prediction method and system
CN115759343A (en) E-LSTM-based user electric quantity prediction method and device
CN112598186B (en) Improved LSTM-MLP-based small generator fault prediction method
CN115375002A (en) Short-term power load prediction method, system, storage medium and computing device

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