Industrial carbon emission data missing filling method and system based on process procedure and time sequence similarity knowledge migration

Abstract

The invention discloses an industrial carbon emission data missing filling method based on process procedure and time sequence similarity knowledge migration, which comprises the following steps of: aiming at input data, adding descriptive texts related to process procedures, extracting text features of industries, products and process routes by utilizing a large language model, constructing vector representation, and in a pre-established vector database, establishing a data library; matching the first n most relevant data sets based on the similarity of industries, products and process routes, and further calculating the similarity between the carbon emission relevant data and the input carbon emission relevant data; according to the method, the similarity of industries, products and process routes and the similarity of carbon emission related data are comprehensively considered, a deep learning model is selected, the input carbon emission related data are used for training on the basis, and finally the model after knowledge migration is used for completing missing value filling. The invention further provides an industrial carbon emission data missing filling system. The missing value filling accuracy is improved under the condition that data samples are insufficient.

Description

Technical Field

[0001] The present invention relates to the fields of data science and data processing, and in particular to a method and system for filling missing data on industrial carbon emissions based on knowledge migration of process steps and time series similarity. Background Art

[0002] Carbon emission data, as a typical time series, is crucial for monitoring carbon emission levels, assessing emission reduction effectiveness, and formulating sound policies. However, due to factors such as human error, equipment failure, and signal transmission anomalies, missing data are inevitable, which can seriously impact subsequent data analysis, trend forecasting, and policy decisions. Therefore, developing accurate and efficient data infill methods is crucial to ensuring the integrity and reliability of carbon emission data.

[0003] Traditional missing value imputation methods primarily include statistical methods such as mean imputation, mode imputation, and median imputation. With the advancement of machine learning, methods such as support vector machines (SVMs), random forests, Adaboost, and K-means clustering have been gradually applied to missing value imputation. These methods have further evolved into deep learning methods such as long short-term memory (LSTM) networks and diffusion models, significantly improving imputation accuracy. However, deep learning models rely on large-scale data to learn complex time series features. When data is insufficient, they struggle to extract effective information, impacting imputation effectiveness. Therefore, knowledge transfer effectively improves learning efficiency and performance under small sample conditions by applying existing models or experience to related tasks.

[0004] In knowledge transfer, the choice of deep learning model has a significant impact on the transfer effect. Traditional methods typically use algorithms such as dynamic time warping (DTW) to evaluate the similarity between the sequence to be filled and the existing sequence, and select a deep learning model based on this. However, when there is a shortage of samples, the difference between the sequence to be filled and the true sequence may lead to inaccurate similarity measurements, thus affecting the filling effect. Therefore, a deep learning model selection method that does not rely entirely on the existing sequence may further improve filling performance. Summary of the Invention

[0005] In order to overcome the shortcomings of the existing technology, the present invention proposes a method for filling missing values in industrial carbon emission data based on knowledge transfer of process procedure and time series similarity. First, for the input data, descriptive text related to the process procedure is added, such as the factory's output, production process, process, production products and other information; then, a large language model is used to extract industry, product, and process route text features and construct a vector representation; in a pre-established vector database, the top n most relevant data groups are matched based on the similarity of industry, product, and process route, and the similarity between their carbon emission-related data and the input carbon emission-related data is further calculated; taking into account the similarity of industry, product, process route and carbon emission-related data, a deep learning model is selected, and on this basis, the input carbon emission-related data is used for training, and finally the model after knowledge transfer is used to complete the missing value filling.

[0006] The technical solution adopted by the present invention to solve its technical problem is:

[0007] A method for filling missing data of industrial carbon emissions based on knowledge transfer of process and time series similarity includes the following steps:

[0008] Step 1: Establish a vector database: For the input text, use the large language model specified by the prompt word project to extract process feature information from the description text, generate text feature vectors, and build a vector database based on this. For each carbon emission-related time series dataset, train a corresponding deep learning model and associate the deep learning model with the text feature vectors and time series data for storage.

[0009] Step 2: Input data includes descriptive text closely related to carbon emissions and carbon emission related time series data X = {x1, x2, ..., x L}, where x t Represents the carbon emission-related data at time t, where t ranges from 1 to L and L represents the length of the time series. The prompt word is used to guide the large language model to extract the feature vector of the process step in the input descriptive text to generate the feature vector INPUT_E;

[0010] Step 3: Deep learning model selection: Using the feature vector INPUT_E as the query vector, retrieve the top n data groups ranked by text similarity from the pre-built vector database. For each of these data groups, further calculate the similarity between their time series data and the input time series data. Using a similarity metric algorithm suitable for time series data, comprehensively consider both text similarity and time series data similarity to select a deep learning model to achieve knowledge transfer.

[0011] Step 4: Model training and missing value filling: On the selected deep learning model, use the input carbon emission-related time series data for training to complete the knowledge transfer process, and use the trained model to fill in the missing values in the time series data to improve data integrity and the accuracy of subsequent analysis.

[0012] Furthermore, the model, input, output and training methods in step 1 are as follows:

[0013] 1.1 Define model input and output: Input time series data Where B is the batch size, L is the sequence length, F is the number of features, X={x1,x2,…,x L} Each time step data x t It consists of two parts: carbon emission-related data and time characteristics. Missing carbon emission data are filled with 0. After the time series data X is processed by the deep learning model, the output is the filling result of the missing carbon emission data.

[0014] 1.2 Training loss function: Construct a joint loss function based on time domain and frequency domain information, time domain loss Loss time Use SmoothL1 function to measure the predicted value and the true value y t The differences:

[0015]

[0016] Frequency domain loss FFT By performing Fourier transform on the predicted and true sequences, we get the frequency domain representation and calculate the mean absolute difference between the two:

[0017]

[0018] Among them, FFT(·) represents the fast Fourier transform. In summary, the total loss function Loss total for

[0019] Loss total =Loss time +λ FFT *Loss FFT

[0020] Among them, λ FFT is the scaling factor of frequency domain loss;

[0021] 1.3 Model training: In each training cycle, the middle part of the input data X is first masked, and the mask window size is set to the maximum window W of the model prediction missing value max , the masked data is fed into the model defined in step 1.1, the masked area is reconstructed and the prediction result is generated Calculate the loss using the loss function defined in step 1.2, backpropagate the calculated loss and use the optimizer to update the model parameters, repeating all steps in the training cycle until the pre-set training cycle is reached.

[0022] Furthermore, in steps 1 and 2, the method of extracting the text feature vector INPUT_E using a large-scale pre-trained model is as follows:

[0023] 2.1 Text preprocessing and word segmentation: Given an input text, first use the word segmenter to split the text into n tokens.

[0024] Text→(token1,token2,…,token n )

[0025] Among them, each token i Represents a token in the text and corresponds to a unique token ID;

[0026] 2.2Token Embedding: Each token ID is mapped to an embedding vector of fixed dimension. Let the embedding matrix be Where VL is the vocabulary size and d is the embedding dimension, then for token i , whose embedding vector is

[0027]

[0028] 2.3 Add position code: Set the position code to p i , then the final input embedding is

[0029] z i =e i +p i

[0030] The entire sequence input is recorded as

[0031] Z=(z1,z2,…,z n )

[0032] 2.4 Transformer encoder processing: The embedded sequence Z is input into the Transformer encoder of the large language model. Each layer of the encoder uses the self-attention mechanism to process the input. The core calculation is

[0033]

[0034] Among them, the query Q, key K and value V are input Z, and the trainable weight matrix W Q 、W K 、W VThe linear transformation is performed to obtain:

[0035] Q=ZW Q ,K=ZW K ,V=ZW V

[0036] After passing through multiple layers of encoders, the hidden representation of each token is obtained

[0037]

[0038] Among them, h i Refers to the hidden representation of the i-th token.

[0039] 2.5 Text feature vector extraction: Use pooling strategy to obtain a fixed-length representation of the entire text. Take max pooling as an example, and take the maximum value of all token hidden states for each dimension.

[0040]

[0041] in, This is the final extracted text feature vector.

[0042] Furthermore, in step 3, the process of deep learning model selection is as follows:

[0043] 3.1 Text similarity calculation: Assume that the input text is extracted through a large language model to obtain a feature vector INPUT_E, and the vector database contains the text feature vector E corresponding to each data group i (i=1,2,…,n), calculate the text similarity between the input text and each data group:

[0044] S text (i) = Similarity(INPUT_E,E i )

[0045] Select the top n data groups with similarity values;

[0046] 3.2 Time series data similarity calculation: For each data group ranked in the top n, take out its corresponding time series X i , with the input time series X in Compare and use the DTW algorithm to calculate the distance D between the two sequences DTW (X in ,X i ), convert the distance into a similarity score S time

[0047]

[0048] 3.3 Comprehensive Similarity Calculation and Deep Learning Model Selection: Using the Average Method to Calculate the Comprehensive Similarity Score

[0049]

[0050] Among them, S(i) represents the comprehensive similarity score of the i-th candidate data group, and the corresponding deep learning model with the highest comprehensive score is finally selected for subsequent knowledge transfer training.

[0051] In step 4, the model structure and training process are consistent with the deep learning model constructed in step 1, but the pre-training weights of the selected deep learning model must be loaded before training to achieve effective knowledge transfer. After the training is completed, the model outputs W each time max The filling data are used as candidate filling data, and then the corresponding prediction results are selected for filling according to the location of the missing value.

[0052] A system for filling missing data of industrial carbon emissions based on knowledge transfer of process and time series similarity, comprising a deep learning model training module, a vector database module, a text feature vector extraction module and a deep learning model selection module.

[0053] The deep learning model training module uses carbon emission-related time series data to train the deep learning model. The module masks the input data and updates the model parameters based on the joint loss function, so that the model has the ability to reconstruct missing data. The module can choose whether to load weights to adapt to the different requirements of steps one and four.

[0054] The vector database module builds and maintains a vector database to store feature vectors extracted from the descriptive text of each dataset and the corresponding pre-trained deep learning model. This module obtains data from open source and private data sources to achieve efficient text feature storage and model association management, providing support for rapid retrieval of subsequent deep learning models.

[0055] The text feature vector extraction module first embeds the input text according to the predefined prompt word template, and uses a large-scale pre-trained language model to preprocess, segment, tokenize, positionally encode, and Transformer encode the text to obtain a fixed-dimensional text feature vector. This module is used to extract descriptive text feature vectors from each data set when building a vector database, and is also used to generate a feature vector INPUT_E for the descriptive text of the input data, thereby providing an accurate basis for deep learning model selection;

[0056] The deep learning model selection module uses the input text feature vector INPUT_E as the query vector, retrieves the top n data groups with the highest text similarity in the vector database by cosine similarity, and further uses methods such as DTW to calculate the similarity of time series data. It comprehensively evaluates each candidate data group based on text and time series similarity, and selects the deep learning model with the highest comprehensive score for knowledge transfer.

[0057] The technical concept of this invention is as follows: It uses a large-scale pre-trained model to extract feature vectors of industry and process information from descriptive text, constructs a text vector database, and associates carbon emission-related time series data with the corresponding descriptive text. For the data to be filled, the cosine similarity and dynamic time warping algorithms are used to comprehensively evaluate the similarity between the text and the time series data. A corresponding deep learning model is selected for knowledge transfer training, and the trained model is used to accurately fill in the missing values in the carbon emission-related data.

[0058] The present invention has the following beneficial effects: It fully integrates process descriptions with time series data, enabling precise selection of deep learning models and knowledge transfer training. It also improves the accuracy of missing value filling in the absence of sufficient data samples, significantly enhancing the integrity of carbon emission-related data and the reliability of subsequent analysis, while reducing the difficulty of model training. Using a joint loss function, which comprehensively considers both time and frequency domain losses, the training process better captures subtle changes in carbon emission-related data. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] Figure 1 A block diagram of the method of the present invention.

[0060] Figure 2 These are the input and output requirements for the deep learning model of the present invention.

[0061] Figure 3 It is the prompt word of the large language model in the present invention. DETAILED DESCRIPTION

[0062] The present invention will be further described below with reference to the accompanying drawings.

[0063] Reference Figure 1-Figure 3 A method for filling missing data of industrial carbon emissions based on knowledge transfer of process and time series similarity includes the following steps:

[0064] Step 1: Create a vector database: Use the prompt word project for the input text, and the prompt words used are as follows: Figure 3 As shown in the prompt, a large language model is specified to extract process feature information from the description text, generate text feature vectors, and construct a vector database based on them. For each carbon emission-related time series dataset, a corresponding deep learning model is trained and the model is associated with the text feature vectors and time series-related data for storage.

[0065] The model, input, output and training methods in step 1 are as follows:

[0066] 1.1 Define model input and output: Input time series data Where B is the batch size, L is the sequence length, F is the number of features, X={x1,x2,…,x L Each time step data x in} t It consists of two parts: carbon emission-related data and time characteristics. The missing parts of carbon emission-related data are filled with 0. After the time series data X is processed by the deep learning model, the output is the filling result of the missing parts of carbon emission-related data.

[0067] 1.2 Training loss function: Construct a joint loss function based on time domain and frequency domain information, time domain loss Loss time Use Smooth L1 function to measure the predicted value and the true value y t The differences:

[0068]

[0069] Frequency domain loss FFT By performing Fourier transform on the predicted and true sequences, we get the frequency domain representation and calculate the mean absolute difference between the two:

[0070]

[0071] Among them, FFT(·) represents the fast Fourier transform. In summary, the total loss function Loss total for

[0072] Loss total =Loss time +λ FFT *Loss FFT

[0073] Among them, λ FFT is the scaling factor of frequency domain loss;

[0074] 1.3 Model training: In each training cycle, first, the middle part of the input data X is masked, and the mask window size is set to the maximum window W of the model prediction missing value max , the masked data is fed into the model defined in step 1.1, the masked area is reconstructed and the prediction result is generated Calculate the loss using the loss function defined in step 1.2, backpropagate the calculated loss and use the optimizer to update the model parameters, repeating all steps in the training cycle until the pre-set training cycle is reached;

[0075] Step 2: Input data includes descriptive text closely related to carbon emissions, such as factory output, process flow, and process information, and carbon emission-related time series data X = {x1, x2, ..., x L}, where x t Represents the carbon emission-related data at time t, where t ranges from 1 to L and L represents the length of the time series. The prompt word is used to guide the large language model to extract the feature vector of the process step in the input descriptive text to generate the feature vector INPUT_E;

[0076] The method for extracting the text feature vector INPUT_E using a large-scale pre-trained model in steps 1 and 2 is as follows:

[0077] 2.1 Text preprocessing and word segmentation: Given an input text, first use the word segmenter to split the text into n tokens.

[0078] Text→(token1,token2,…,token n )

[0079] Among them, each token i Represents a token in the text and corresponds to a unique token ID;

[0080] 2.2Token Embedding: Each token ID is mapped to an embedding vector of fixed dimension. Let the embedding matrix be Where VL is the vocabulary size and d is the embedding dimension, then for token i , whose embedding vector is

[0081]

[0082] 2.3 Add position code: Set the position code to p i , then the final input embedding is

[0083] z i =e i +p i

[0084] The entire sequence input is recorded as

[0085] Z=(z1,z2,…,z n )

[0086] 2.4 Transformer encoder processing: The embedded sequence Z is input into the Transformer encoder of the large language model. Each layer of the encoder uses the self-attention mechanism to process the input. The core calculation is

[0087]

[0088] Among them, the query Q, key K and value V are input Z, and the trainable weight matrix W Q 、W K 、W V The linear transformation is performed to obtain:

[0089] Q=ZW Q ,K=ZW K ,V=ZW V

[0090] After passing through multiple layers of encoders, the hidden representation of each token is obtained

[0091]

[0092] Among them, h i Refers to the hidden representation of the i-th token;

[0093] 2.5 Text feature vector extraction: Use pooling strategy to obtain a fixed-length representation of the entire text. Take max pooling as an example, and take the maximum value of all token hidden states for each dimension.

[0094]

[0095] in, This is the final extracted text feature vector;

[0096] Step 3: Deep learning model selection: In the pre-built vector database, the feature vector INPUT_E is used as the query vector. The cosine similarity method is used to retrieve the top n data groups ranked by text similarity. For each of these data groups, the similarity between their time series data and the input time series data is further calculated. A similarity measurement algorithm suitable for time series data is used, comprehensively considering both text similarity and time series data similarity, to select a deep learning model to achieve knowledge transfer.

[0097] In step 3, the process of deep learning model selection is as follows:

[0098] 3.1 Text similarity calculation: Assume that the input text is extracted through a large language model to obtain a feature vector INPUT_E, and the vector database contains the text feature vector E corresponding to each data group i , i=1,2,…,n, and the cosine similarity is used to calculate the text similarity between the input text and each data group. The formula is

[0099]

[0100] Select the top n data groups with similarity values;

[0101] 3.2 Time series data similarity calculation: For each data group ranked in the top n, take out its corresponding time series X i , with the input time series X in Compare and use the DTW algorithm to calculate the distance D between the two sequences DTW (X in ,X i ), convert the distance into a similarity score S time

[0102]

[0103] 3.3 Comprehensive Similarity Calculation and Deep Learning Model Selection: Using the Average Method to Calculate the Comprehensive Similarity Score

[0104]

[0105] Among them, S(i) represents the comprehensive similarity score of the i-th candidate data group. Finally, the deep learning model corresponding to the group with the highest comprehensive score is selected for subsequent knowledge transfer training;

[0106] Step 4: Model training and missing value filling: The selected deep learning model is trained using the input carbon emission-related time series data to complete the knowledge transfer process. The trained model is then used to fill in missing values in the time series data, improving data integrity and the accuracy of subsequent analysis.

[0107] In step 4, the model structure and training process are consistent with the deep learning model constructed in step 1, but the pre-training weights of the selected deep learning model must be loaded before training to achieve effective knowledge transfer. After the training is completed, the model outputs W each time max The filling data are used as candidate filling data, and then the corresponding prediction results are selected for filling according to the location of the missing value.

[0108] This embodiment also provides an industrial carbon emission data missing filling system based on process and time series similarity knowledge migration, including a deep learning model training module, a vector database module, a text feature vector extraction module and a deep learning model selection module.

[0109] The deep learning model training module uses carbon emission-related time series data to train the deep learning model. The module masks the input data and updates the model parameters based on the joint loss function, so that the model has the ability to reconstruct missing data. The module can choose whether to load weights to adapt to the different requirements of steps one and four.

[0110] The vector database module builds and maintains a vector database to store feature vectors extracted from the descriptive text of each dataset and the corresponding pre-trained deep learning model. This module obtains data from open source and private data sources to achieve efficient text feature storage and model association management, providing support for rapid retrieval of subsequent deep learning models.

[0111] The text feature vector extraction module first embeds the input text according to a predefined prompt word template and uses a large-scale pre-trained language model such as Qwen 2.5 to preprocess the text, perform word segmentation, token embedding, position encoding, and Transformer encoding to obtain a fixed-dimensional text feature vector. This module is used to extract descriptive text feature vectors from each dataset when building a vector database, and is also used to generate a feature vector INPUT_E for the descriptive text of the input data, thereby providing an accurate basis for deep learning model selection.

[0112] The deep learning model selection module uses the input text feature vector INPUT_E as the query vector, retrieves the top n data groups ranked by text similarity in the vector database through cosine similarity, and further uses methods such as DTW to calculate the similarity of time series data. It comprehensively evaluates each candidate data group based on text and time series similarity, and selects the deep learning model with the highest comprehensive score for knowledge transfer.

[0113] As described above, the specific implementation steps of this patent make the present invention clearer. Any modifications and changes made to the present invention within the spirit of the present invention and the scope of protection of the claims fall within the scope of protection of the present invention.

Claims

1. A method for filling missing data of industrial carbon emissions based on knowledge transfer of process and time series similarity, characterized by: The method comprises the following steps: Step 1: Establish a vector database: For the input text, use the large language model specified by the prompt word project to extract process feature information from the description text, generate text feature vectors, and build a vector database based on this. For each carbon emission-related time series dataset, train a corresponding deep learning model and associate the deep learning model with the text feature vectors and time series data for storage. Step 2: Input data includes descriptive text closely related to carbon emissions and carbon emission related time series data X = {x1, x2, ..., x L }, where x t Represents the carbon emission-related data at time t, where t ranges from 1 to L and L represents the length of the time series. The prompt word is used to guide the large language model to extract the feature vector of the process step in the input descriptive text to generate the feature vector INPUT_E; Step 3: Deep learning model selection: In the pre-built vector database, the feature vector INPUT_E is used as the query vector. A similarity method is used to retrieve the top n data groups ranked by text similarity. For each of these data groups, the similarity between their time series data and the input time series data is further calculated. A similarity measurement algorithm suitable for time series data is used, comprehensively considering both text similarity and time series data similarity, to select a deep learning model to achieve knowledge transfer. Step 4: Model training and missing value filling: On the selected deep learning model, use the input carbon emission-related time series data for training to complete the knowledge transfer process, and use the trained model to fill in the missing values in the time series data to improve data integrity and the accuracy of subsequent analysis.

2. The method for filling missing data of industrial carbon emissions based on process and time series similarity knowledge transfer according to claim 1 is characterized in that: The model, input, output and training methods in step 1 are as follows: 1.1 Define model input and output: Input time series data Where B is the batch size, L is the sequence length, F is the number of features, X={x1,x2,…,x L } Each time step data x t It consists of two parts: carbon emission-related data and time characteristics. Missing carbon emission data are filled with 0. After the time series data X is processed by the deep learning model, the output is the filling result of the missing carbon emission data. 1.2 Training loss function: Construct a joint loss function based on time domain and frequency domain information, time domain loss Loss time Use SmoothL1 function to measure the predicted value and the true value y t The differences: Frequency domain loss FFT By performing Fourier transform on the predicted and true sequences, we get the frequency domain representation and calculate the mean absolute difference between the two: Among them, FFT(·) represents the fast Fourier transform. In summary, the total loss function Loss total for Loss total =Loss time +λ FFT *Loss FFT Among them, λ FFT is the scaling factor of frequency domain loss; 1.3 Model training: In each training cycle, the middle part of the input data X is first masked, and the mask window size is set to the maximum window W of the model prediction missing value max , the masked data is fed into the model defined in 1.1, the masked area is reconstructed and the prediction result is generated The loss function defined in 1.2 is used to calculate the loss, the calculated loss is propagated forward and the model parameters are updated using the optimizer, and all steps in the training cycle are repeated until the pre-set training cycle is reached.

3. The method for filling missing data of industrial carbon emissions based on process and time series similarity knowledge transfer according to claim 1 or 2, characterized in that: The method for extracting the text feature vector INPUT_E using a large-scale pre-trained model in steps 1 and 2 is as follows: 2.1 Text preprocessing and word segmentation: Given an input text, first use the word segmenter to split the text into n tokens. Text→(token1,token2,…,token n ) Among them, each token i Represents a token in the text and corresponds to a unique token ID; 2.2 Token Embedding: Each token ID is mapped to an embedding vector of fixed dimension. Let the embedding matrix be Where VL is the vocabulary size and d is the embedding dimension, then for token i , whose embedding vector is e i =E[token i ], 2.3 Add position code: Set the position code to p i , then the final input embedding is with i =e i +p i The entire sequence input is recorded as Z=(z1,z2,…,z n ) 2.4 Transformer encoder processing: The embedded sequence Z is input into the Transformer encoder of the large language model. Each layer of the encoder uses the self-attention mechanism to process the input. The core calculation is Among them, the query Q, key K and value V are input Z, and the trainable weight matrix W Q 、W K 、W V The linear transformation is performed to obtain: Q=ZW Q ,K=ZW K ,V=ZW V After passing through multiple layers of encoders, the hidden representation of each token is obtained H=(h1,h2,…,h n ), Among them, h i Refers to the hidden representation of the i-th token; 2.5 Text feature vector extraction: Use pooling strategy to obtain a fixed-length representation of the entire text. Take max pooling as an example, and take the maximum value of all token hidden states for each dimension. in, This is the final extracted text feature vector.

4. The method for filling missing data of industrial carbon emissions based on process and time series similarity knowledge transfer according to claim 1 or 2, characterized in that: In step 3, the process of deep learning model selection is as follows: 3.1 Text similarity calculation: Assume that the input text is extracted through a large language model to obtain a feature vector INPUT_E, and the vector database contains the text feature vector E corresponding to each data group i , i = 1, 2, ..., n, calculate the text similarity between the input text and each data group: S text (i)=Similarity(INPUT_E,E i ) Select the top n data groups with similarity values; 3.2 Time series data similarity calculation: For each data group ranked in the top n, take out its corresponding time series X i , with the input time series X in Compare and use the DTW algorithm to calculate the distance D between the two sequences DTW (X in ,X i ), convert the distance into a similarity score S time 3.3 Comprehensive Similarity Calculation and Deep Learning Model Selection: Using the Average Method to Calculate the Comprehensive Similarity Score Among them, S(i) represents the comprehensive similarity score of the i-th candidate data group, and the corresponding deep learning model with the highest comprehensive score is finally selected for subsequent knowledge transfer training.

5. The method for filling missing data of industrial carbon emissions based on process and time series similarity knowledge transfer according to claim 1 or 2, characterized in that: In step 4, the model structure and training process are consistent with the deep learning model constructed in step 1, but the pre-training weights of the selected deep learning model must be loaded before training to achieve effective knowledge transfer. After the training is completed, the model outputs W each time max The filling data are used as candidate filling data, and then the corresponding prediction results are selected for filling according to the location of the missing value.

6. A system implementing the method for filling missing data of industrial carbon emissions based on process and time series similarity knowledge transfer as claimed in claim 1, characterized in that: The system includes a deep learning model training module, a vector database module, a text feature vector extraction module and a deep learning model selection module. The deep learning model training module uses carbon emission-related time series data to train the deep learning model. The module masks the input data and updates the model parameters based on the joint loss function, so that the model has the ability to reconstruct missing data. The module can choose whether to load weights to adapt to the different requirements of steps one and four. The vector database module builds and maintains a vector database to store feature vectors extracted from the descriptive text of each dataset and the corresponding pre-trained deep learning model. This module obtains data from open source and private data sources to achieve efficient text feature storage and model association management, providing support for rapid retrieval of subsequent deep learning models. The text feature vector extraction module first embeds the input text according to the predefined prompt word template, and uses a large-scale pre-trained language model to preprocess, segment, tokenize, positionally encode, and Transformer encode the text to obtain a fixed-dimensional text feature vector. This module is used to extract descriptive text feature vectors from each data set when building a vector database, and is also used to generate a feature vector INPUT_E for the descriptive text of the input data, thereby providing an accurate basis for deep learning model selection; The deep learning model selection module uses the input text feature vector INPUT_E as the query vector, retrieves the top n data groups ranked by text similarity in the vector database through cosine similarity, and further uses methods such as DTW to calculate the similarity of time series data. It comprehensively evaluates each candidate data group based on text and time series similarity, and selects the deep learning model with the highest comprehensive score for knowledge transfer.

Images