A cems carbon emission data failure processing method and related equipment

Abstract

The application discloses a CEMS carbon emission data failure processing method and related equipment, and belongs to the field of carbon emission online monitoring and data quality guarantee. The method comprises the following steps: constructing a working condition knowledge base containing historical "backtracking window-future sequence" key-value pairs; monitoring and identifying CEMS carbon emission data failure in real time; taking the backtracking window before failure as a query key, retrieving the most similar K historical fragments from the knowledge base as evidence; inputting the backtracking window and the retrieved evidence into a time series prediction model for fusion, generating a predicted value of the failure period as substitute data; and synchronously outputting an interpretability report containing evidence sources, similarity and contribution degree. The system is an intelligent agent for executing the method. The application explicitly introduces historical similar working condition evidence to constrain and enhance prediction, solves the problems of low precision and poor interpretability caused by the existing fixed rule substitution method, and significantly improves the accuracy, auditability and adaptability to complex working conditions of data repair.

Description

Technical Field

[0001] This application relates to the field of carbon emission monitoring and data quality assurance technology, and in particular to a method and related equipment for handling CEMS carbon emission data failure. Background Technology

[0002] With the increasing urgency of global action to address climate change, accurate monitoring and accounting of major carbon emission sources such as thermal power generation have become crucial. CEMS (Continuous Emission Monitoring System) serves as a key technology for real-time, continuous monitoring of flue gas emissions. However, during long-term operation, CEMS equipment often experiences data loss, anomalies, or sudden changes due to sensor malfunctions, calibration errors, and communication interruptions.

[0003] To address the issue of data failure, current domestic and international regulations have proposed corresponding handling standards. For example, the US 40 CFR Part 75 uses average or extreme values ​​for replacement based on monitoring availability and failure duration. While this method ensures compliance, it is often overly conservative, leading to a systematic overestimation of carbon emissions. Domestically, the highest historical emission value is typically used for replacement, a more coarse approach. These conservative replacement schemes based on fixed rules fail to fully consider the time-varying characteristics of actual power plant operating conditions, easily causing significant deviations between replacement data and actual emission trajectories, affecting the fairness and accuracy of accounting.

[0004] In addition, there are some patents in the existing technology related to CEMS monitoring or time-series data processing. For example, some solutions improve monitoring accuracy by data modeling or adding measuring points, but do not specifically propose repair methods for data failure; other solutions propose time-series data retrieval or prediction methods, but their application scenarios are mostly general power load or traffic flow prediction, without taking into account the characteristics of CEMS carbon emission data, nor considering the introduction of historical experience knowledge in data failure repair to enhance the accuracy and interpretability of the results.

[0005] Therefore, how to effectively integrate historical operational knowledge to achieve accurate, reliable, and interpretable repair processing of CEMS carbon emission failure data has become a technical problem that urgently needs to be solved in this field. Summary of the Invention

[0006] The main objective of this application is to propose a method, electronic device, storage medium, and program product for handling CEMS carbon emission data failure based on a knowledge retrieval-enhanced time-series large model. By explicitly introducing historical evidence of similar operating conditions and integrating it with the predictive capabilities of the time-series large model, it achieves refined and highly accurate replacement of failed data and simultaneously generates auditable explanatory reports, thereby improving the integrity, reliability, and transparency of carbon emission monitoring data.

[0007] To achieve the above objectives, one aspect of this application proposes a method for handling CEMS carbon emission data failure, the method comprising: A historical operating condition knowledge base is constructed, which stores multiple historical operating condition segments. Each segment includes a backtracking window sequence and its corresponding subsequent real carbon emission sequence. Real-time monitoring of CEMS carbon emission data streams and identification of periods of data failure; A backtracking window sequence of a preset length before the data failure period is extracted as the query key; Based on the query key, K similar historical operating condition segments are retrieved from the historical operating condition knowledge base, where K≥1; The backtracking window sequence and the K similar historical operating condition segments are input into the time series prediction model to generate a carbon emission prediction sequence for the data failure period, which serves as a substitute value. The alternative value is output, and an interpretability evidence report is generated simultaneously. The report includes at least the identifiers of the similar historical working condition segments used, their similarity information, and their contribution to the prediction results.

[0008] In some embodiments, the construction of the historical operating condition knowledge base includes: Multiple time segments are extracted from historical CEMS runtime data using a sliding window approach; For each time segment, the data from the previous L times are defined as keys, and the carbon emission data from the next H times are defined as values, forming key-value pairs; After the keys are standardized to eliminate the influence of offset, they are stored in the historical operating condition knowledge base.

[0009] In some embodiments, retrieving K similar historical operating condition segments from the historical operating condition knowledge base based on the query key includes: Calculate the similarity between the query key and each fragment key in the knowledge base; Select one or more segments whose similarity exceeds a preset threshold; When there are segments exceeding the preset threshold, the Top-K segments are selected as search results based on similarity from high to low.

[0010] In some embodiments, the similarity is calculated using Pearson correlation coefficient, cosine similarity, or a distance-based metric; the preset threshold is 0.85.

[0011] In some embodiments, the time-series prediction model is a Time-MoE model; the specific formula for generating the prediction sequence is:

[0012] in, For the predicted sequence, To integrate weights, For time series prediction models, the backtracking window The original prediction output, This is the evidence vector obtained by weighted aggregation of subsequent sequences of the K similar historical work condition segments retrieved.

[0013] In some embodiments, the weights of the weighted aggregation are calculated using a temperature-based Softmax function, as follows:

[0014] in, For the first Similarity score of each segment For temperature coefficient, This is the set of indexes for the retrieved fragments.

[0015] In some embodiments, retrieving K similar historical operating condition segments from the historical operating condition knowledge base based on the query key includes: If the similarity of all retrieved historical working condition segments is lower than the preset safety threshold, the search enhancement prediction will be abandoned, and the method will automatically switch to a conservative data substitution method based on preset rules.

[0016] In some embodiments, the keys in the backtracking window sequence and / or the historical operating condition knowledge base, in addition to the carbon emission sequence, may be extended to one or more covariate features; the covariate features include at least one of power generation load, fuel parameters, unit start-up and shutdown status, and ambient temperature.

[0017] In some embodiments, the interpretability evidence report is structured data that can be directly read and verified by the audit system.

[0018] To achieve the above objectives, another aspect of this application provides an electronic device, which includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the method described above.

[0019] To achieve the above objectives, another aspect of the embodiments of this application proposes a computer-readable storage medium storing a computer program that, when executed by a processor, implements the method described above.

[0020] To achieve the above objectives, another aspect of the embodiments of this application proposes a computer program product, including a computer program that, when executed by a processor, implements the method described above.

[0021] Compared with the prior art, this application has the following beneficial effects: 1) High accuracy: By retrieving historical evidence most similar to the current working conditions and using a time series large model for constrained prediction, it can more accurately characterize local dynamic trends and significantly reduce the error between substitute data and true values.

[0022] 2) Strong interpretability and auditability: The accompanying evidentiary reports clearly reveal the basis and process of data substitution, providing transparent and traceable technical support for carbon emission accounting, trading and compliance auditing, and enhancing the credibility of the results.

[0023] 3) Good adaptability and robustness: Through multi-scale retrieval and adaptive fusion mechanisms, it can effectively handle data failures of varying durations, from short-term missing data to long-term interruptions. When similar evidence is insufficient, the system can smoothly revert to preset conservative rules to ensure processing reliability.

[0024] 4) Easy to implement and expand: This method is implemented in software, requiring no modification to existing CEMS hardware, making deployment simple. The system architecture naturally supports the integration of multiple covariates such as load, fuel, and start / stop status, and facilitates the subsequent integration of new business rules or algorithm models, demonstrating strong scalability.

[0025] 5) High processing efficiency: The intelligent agent can be deployed on edge devices to realize real-time, online processing of data failures, which greatly improves the operating efficiency of the carbon emission monitoring system compared with manual repair or offline replenishment.

[0026] 6) Dynamic updates: The historical database can be dynamically updated according to the actual operation, and at the same time, it continuously records various abnormal fluctuations in the operation status. Attached Figure Description

[0027] Figure 1 This is a flowchart illustrating the CEMS carbon emission data failure handling method provided in this application.

[0028] Figure 2 The structural block diagram of the CEMS carbon emission data failure processing intelligent agent system based on knowledge retrieval enhanced temporal series large model provided in this application.

[0029] Figure 3 This is a flowchart illustrating the knowledge retrieval enhancement database construction and interpretable audit output in this embodiment of the application.

[0030] Figure 4 This is a diagram of the fusion architecture of temporal large model and knowledge retrieval enhancement in the embodiments of this application.

[0031] Figure 5 This is an overview of the carbon emission rate data of the U.S. gas-fired power plant used in the experiments of this application.

[0032] Figure 6This is a comparison chart showing the effects of data imputation using the method of this application and the baseline method.

[0033] Figure 7 This is a schematic diagram of the hardware structure of the electronic device provided in the embodiments of this application. Detailed Implementation

[0034] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of this application and are not intended to limit it. In the following description, when referring to the accompanying drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with those of this application; they are merely examples of apparatuses and methods consistent with some aspects of the embodiments of this application as detailed in the appended claims.

[0035] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing embodiments of this application only and is not intended to limit this application.

[0036] Before providing a detailed description of the embodiments of this application, some of the nouns and terms involved in the embodiments of this application will be explained first. The nouns and terms involved in the embodiments of this application are subject to the following interpretations.

[0037] 1) CEMS is an abbreviation for Continuous Emission Monitoring System. It refers to a device that continuously monitors the concentration and total emission of gaseous pollutants and particulate matter emitted from air pollution sources and transmits the information to the competent authorities in real time. It is also known as "automatic flue gas monitoring system", "continuous flue gas emission monitoring system" or "online flue gas monitoring system".

[0038] With the increasing severity of global climate change, reducing greenhouse gas emissions has become an international consensus. Carbon emission data measured by CEMS can be directly used for total emission accounting, emissions trading, and environmental compliance audits. As a real-time, continuous monitoring technology for flue gas emissions, it plays an increasingly important role in carbon emission monitoring at thermal power plants. Monitoring CO2 emissions through CEMS can compensate for the shortcomings of traditional accounting methods, providing more accurate and real-time data support for carbon emission management. However, during long-term operation, CEMS equipment often experiences data failures due to problems such as sensor drift, sampling probe blockage, calibration anomalies, or communication interruptions, including data loss, abnormal offsets, and abrupt changes. If failed to properly repair the failed data, it will directly affect the total emission accounting results, and may even lead to overestimation or underestimation of carbon emissions, resulting in economic losses and compliance risks. Therefore, how to effectively handle failed CEMS data has become an important issue in ensuring the integrity of the carbon emission monitoring system.

[0039] Both domestic and international regulations have proposed methods to address the issue of expired CEMS carbon emission data, but these methods still have significant shortcomings. First, while international regulations have introduced mechanisms for replacing missing data, the algorithms are overly conservative. For example, the US 40 CFR Part 75 classifies expired data based on monitoring availability and duration of expiration, gradually transitioning replacement values ​​from averages to extreme values ​​to avoid underestimating emissions. While this principle ensures compliance, it often leads to a systematic overestimation of emissions, increasing the carbon cost pressure on companies. Second, current domestic regulations do not categorize expired data according to duration of expiration; they only select the highest CO2 emission value from the past period for replacement. This method is coarser than the 40 CFR Part 75 standard.

[0040] To address the shortcomings in regulations, several technologies and methods have been proposed. For example, the first existing technical solution combines conventional fuel analysis and CEMS system data, using data modeling and calibration factors to correct CEMS data and improve monitoring accuracy; however, it does not address data loss issues caused by equipment failure, calibration, or maintenance. The second existing technical solution monitors carbon emissions using gas detection and data analysis devices, which does not fall under the category of data loss processing. The third existing technical solution addresses the uneven distribution of flue gas concentration and velocity in the cross-section of the flue gas duct by proposing a multi-channel continuous flue gas emission monitoring method and system, increasing the number of channels and measurement points to improve the accuracy and reliability of flue gas concentration and velocity; however, it does not consider data loss issues caused by equipment failure or maintenance.

[0041] It is worth noting that some related patents have also emerged in the fields of time series data processing and retrieval enhancement. For example, the fourth prior art solution achieves efficient retrieval of time series data by constructing fast indexes and full-text indexes, but it mainly addresses the retrieval performance problem under massive data conditions and does not involve predictive replacement processing in the case of data failure, nor does it consider using the retrieved historical information to enhance the performance of the prediction model. The fifth prior art solution extracts high-level features through self-supervised pre-training, retrieves relevant features from the feature pool using a retrieval module, and finally selects the optimal expert network for prediction through an expert selection module. Although this patent introduces a retrieval enhancement mechanism in time series prediction, its application scenario focuses on general prediction tasks such as power load and traffic flow, and does not address the specific scenario of CEMS carbon emission data failure. Furthermore, its retrieval object is high-dimensional features rather than original historical operating condition data, lacking a specific design for the data characteristics and compliance requirements in the field of carbon emission monitoring.

[0042] It is evident that current technologies for handling expired carbon emission data from thermal power plant CEMS (Consumer Energy Management System) do not adequately consider utilizing historical experience to accurately process such data. In particular, they lack solutions that combine retrieval enhancement mechanisms with large-scale time-series models to refine the processing of carbon emission data based on its characteristics. Furthermore, large-scale time-series models that solely rely on general knowledge and strong generalization capabilities lack interpretability. Historical experience, on the other hand, can not only improve the accuracy of processing expired data but also provide interpretable processing guidelines, which is precisely what the current technological system lacks.

[0043] In view of this, this application provides a method, electronic device, storage medium, and program product for handling CEMS carbon emission data failure based on a knowledge retrieval-enhanced time-series large-scale model. The approach is as follows: A historical operational data-driven operating condition knowledge base is constructed. When CEMS carbon emission data fails, historical fragments similar to the current operating condition are retrieved in real time. The retrieved "similar evidence" is used as external knowledge context and fused with the time-series large-scale model to generate predictive replacement values ​​for the replacement interval. Simultaneously, interpretable evidence such as the source fragment, similarity weight, and contribution decomposition is output. This method, by explicitly introducing historical experience, can significantly improve the accuracy and interpretability of data replacement without modifying existing CEMS hardware and data reporting processes. Furthermore, it can flexibly integrate multiple covariates such as load, fuel, and start-stop status to further enhance performance.

[0044] The CEMS carbon emission data failure handling method provided in this application relates to the field of carbon emission monitoring and data quality assurance technology. The CEMS carbon emission data failure handling method provided in this application can be applied to a terminal, a server, or software running on a terminal or server. In some embodiments, the terminal can be a smartphone, tablet, laptop, desktop computer, smart speaker, smartwatch, or vehicle terminal, but is not limited to these. The server can be configured as an independent physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms. The server can also be a node server in a blockchain network. The software can be an application implementing the CEMS carbon emission data failure handling method, but is not limited to the above forms.

[0045] This application can be used in a wide variety of general-purpose or special-purpose computer system environments or configurations. Examples include: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronics, network PCs, minicomputers, mainframe computers, and distributed computing environments including any of the above systems or devices. This application can be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc., that perform specific tasks or implement specific abstract data types. This application can also be practiced in distributed computing environments where tasks are performed by remote processing devices connected via a communication network. In distributed computing environments, program modules can reside in local and remote computer storage media, including storage devices.

[0046] like Figure 1 As shown, this embodiment provides a method and intelligent agent system for handling CEMS carbon emission data failure based on a knowledge retrieval-enhanced time-series large model. First, a backtracking window of data is constructed, and historical similarity searches are performed. The retrieved data is then aggregated as evidence, and subsequently, gaps are filled through fusion using a time-series prediction model, resulting in the final audit output. This embodiment also supports the retrieval of covariates and rule constraints related to carbon emissions. The technical solution, technical effects, and implementation methods of this application are described in detail below with reference to the accompanying drawings.

[0047] The overall modeling process in this embodiment is as follows: The CEMS continuous carbon emission monitoring sequence can be defined as... , For a moment tThe CO2 emission rate. Extensible covariates such as power generation load, fuel consumption, and start / stop flags are defined as... Let the observation vector be... . The sequence can be divided into historical carbon emission window segments and retrospective window segments. L and data failure segment H .for When a sequence data failure detection tool detects data failure, it first determines whether the data is invalid. Normal data is directly stored in the data storage device, while the result of the invalid data determination is returned to the Language Large Model (LLM). The LLM constructs a backtracking window:

[0048] The ultimate goal is to produce alternative sequences:

[0049] In the formula H To fill the prediction step size, D This is a work condition knowledge base dataset, specifically consisting of historical key-value pairs. Retrieval Provide with X t Similar evidence of future trends allows this paradigm to explicitly inject historical experience into the prediction process, thereby improving the accuracy and interpretability of failure replacements.

[0050] The carbon emission history window retrieval process is actually a retrieval of similar key-value segments, dividing the historical sequence into sliding window steps. s Extract key-value fragment pairs, where the key is the backtrack window fragment and the value is its subsequent value. H The future trajectory at this point in time.

[0051]

[0052] To mitigate the effects of scale and offset, the backtracking window and key fragment are offset and aligned:

[0053] The similarity between the backtracking window and each key segment can be calculated using Pearson coefficient, cosine similarity, or distance-based metrics. This invention's implementation uses Pearson coefficient by default. ρ i The closer the value is to 1, the more similar the change patterns of the two are. The calculation formula is:

[0054] The Pearson coefficient is used for similarity assessment. When the similarity reaches 0.85 or higher (this is an initial empirical value), the top-m sequences are selected from largest to smallest as candidate sequences, and the index set is... JThe weights are obtained using temperature-based Softmax:

[0055] In the formula Temp The aggregation temperature is used; the smaller the value, the more attention is paid to search sequences with high similarity. Finally, the aggregation yields future evidence for the search results.

[0056] The offset is recovered from the result, and the predicted value is obtained by the RAG (Retrieval Augmented Generation) method:

[0057] Unlike existing technologies, the retrieval module in this embodiment not only considers the similarity of historical time series, but also introduces covariates as retrieval features, thereby improving the context matching accuracy of the retrieval.

[0058] The prediction model in this embodiment Using linear fusion, the final prediction calculation formula is as follows:

[0059] In the formula, α is the fusion weight. f The system uses the Time-MoE prediction function for large time-series models. It supports the introduction of external covariates, which can be incorporated into the key value during the retrieval stage to enhance the context-awareness of the retrieval.

[0060] Intelligent agent system diagram as follows Figure 2 As shown, this intelligent agent consists of a large language model (LLM), a data failure detection tool, a carbon emission history window retrieval tool, a time-series model prediction tool, and data storage. The LLM acts as the agent's brain, utilizing general-purpose large models with reasoning and knowledge integration capabilities, such as Deepseek, Claude, ChatGPT, and Qwen, for task decomposition, semantic understanding, and knowledge injection scheduling. The data failure detection tool identifies missing data in the CEMS real-time monitoring data and returns the results to the LLM. The carbon emission history window retrieval tool, based on a similar operating condition matching mechanism, retrieves emission data and knowledge fragments from the historical database that are closest to the current operating state, providing reference information for the model. The time-series model prediction tool integrates internal and external memory to complete the missing data. The workflow of this intelligent agent is as follows: S201: CEMS continuously senses and outputs carbon emission rate data streams, which can be divided into historical carbon emission window segments, retrospective window segments, and data failure segments. The historical carbon emission window segment is used to build an operating condition knowledge base, while the retrospective window segment is used for query input.

[0061] S202: LLM extracts the carbon emission rate data stream and calls the data failure detection tool to monitor the real-time stream. Once a data failure is detected, the agent is triggered to enter the completion process and notify the operation and maintenance personnel. If no failure is detected, the data is written to the data storage device as usual and the process ends.

[0062] S203: After data failure is triggered, LLM calls the historical retrieval tool, takes the backtracking window series as the query input, starts to search for historical fragments in the historical window library, performs similarity calculation and sorting on the backtracking window data and historical fragments, and selects the top-m most similar results.

[0063] S204: If the aggregation similarity is greater than 0.85, it is considered that the historical memory is sufficient to support the retrieval-based imputation; otherwise, it is judged as insufficient similarity, and it is transferred to the alternative processing path of regulations or rules, and the reason for the rollback is marked in the audit record.

[0064] S205: Once the similarity is determined by the threshold, the LLM calls the time series large model prediction tool to make a prediction. This tool inputs the backtracking window data into the model and merges it with the historical knowledge retrieved from the outside (the future values ​​corresponding to the Top-m segments) for prediction. The results are then stored in the data storage device and properly labeled.

[0065] S206: Input the similar fragment sources, similarity scores, contribution weights and final decision reasons generated in the above process into the LLM model. After processing, the LLM model outputs an evidentiary, auditable and interpretable report.

[0066] See Figure 3 , Figure 3 This document outlines a flowchart for enhancing the database construction and interpretable audit output for knowledge retrieval. The process includes building a work condition knowledge base, calculating and ranking similarities, aggregating search results, and providing interpretable output. The specific steps are as follows: S301: Construction of the Operating Condition Knowledge Base: Key information is extracted from the raw data to construct key-value pairs, ensuring a clear data structure and ease of retrieval. Subsequently, offset elimination techniques are used to standardize the data, eliminating systematic biases caused by differences in data sources or acquisition conditions. Finally, the processed data is stored in the knowledge base, forming a structured operating condition knowledge base, providing a foundation for subsequent retrieval and analysis.

[0067] S302: Similarity Calculation and Ranking: For the input carbon emission failure data, the system constructs a corresponding query key and performs a similarity search in the knowledge base. By calculating the similarity between the query key and each segment in the knowledge base, the system selects the top-m results with the highest similarity as the basis for subsequent aggregation analysis.

[0068] S303: After obtaining the Top-m similar segments, the system further calculates the attention weight of each segment to measure its contribution to the final output. Subsequently, the search results are weighted and aggregated based on the weights, and offset recovery technology is used to restore the original semantic information of the data. This step not only enhances the accuracy of the results but also provides a basis for interpretable output.

[0069] S304: The system generates interpretable output that includes the source of similarity, similarity score, contribution weight, and the rationale for the final decision. Carbon emission auditors can use this output to clearly understand the basis for their decisions and the impact of each data segment on the conclusions, thereby enhancing the transparency and credibility of the audit process.

[0070] See Figure 4 , Figure 4 The diagram shows the fusion architecture of the Time-MoE large-scale model and knowledge retrieval enhancement. This framework includes the Time-MoE large-scale model and the RAG module. The data processing flow is as follows: S401: The design data input dimensions are batch size, input length, and feature variable. In this implementation case, the batch size is designed to be 32, the input length is 512, and the feature variable is 1.

[0071] S402: Time-MoE Data Processing Flow for Large Time-Series Models. In this implementation case, the pre-trained weight parameters activation size is 50M. As one implementation method, step S402 specifically includes the following steps S21-S26: S21: After the input layer receives the data, it first undergoes normalization processing through the root mean square layer. S22: Then, the causal attention mechanism is used to capture the temporal dependencies in the sequence; S23: Enhance information flow and training stability through residual connections and root mean square layer normalization; S24: Input Hybrid Expert (MoE) module, which consists of multiple expert networks working together to process different temporal patterns and improve the model's expressive power; S25: Perform residual connections again to preserve the original information and alleviate gradient vanishing; S26: Finally, preliminary time series prediction results are generated through the prediction head of the base model.

[0072] S403: RAG data processing flow. As one implementation method, step S403 specifically includes the following steps S31-S33: S31: The retrieval module, based on the built-in pre-processed database, performs multi-cycle decomposition on the input time series and extracts key pattern features; S32: Perform a Top-m search to retrieve the m most relevant segments to the current sequence from the knowledge base; S33: Weighted aggregation of search results, combining multi-source information to form an enhanced representation.

[0073] S404: The output of the Time-MoE path is fused with the enhanced representation of the RAG path for prediction. Combining time series modeling and external knowledge, the final output dimensions are batch size, prediction length, and prediction results of the target variable, thus completing the entire time series prediction process.

[0074] The solutions of the embodiments of this application will be described in detail and explained below with reference to specific application examples.

[0075] This embodiment uses real production data from a US gas-fired power plant to verify the practicality, inventiveness, and novelty of the invention. The carbon emission data comes from the US Environmental Protection Agency (EPA), and the overall data is as follows: Figure 5 As shown. Specifically, this experiment selected carbon emission rate data (in short tons) from the Barry gas-fired power plant for the entire year of 2024. Data was collected hourly, spanning from 00:00 on January 1, 2024 to 23:00 on December 31, 2024, totaling 8784 continuous observations. This dataset has the following significant characteristics: the carbon emission rate exhibits a clear multi-periodic fluctuation pattern, including daily, weekly, and seasonal cycles, as well as irregular sudden fluctuations, which fully reflects the complex impact of power load changes on the carbon emissions of the gas-fired power plant. The numerical dynamic range is wide, with the carbon emission rate fluctuating dramatically throughout the year, ranging from a minimum of 239.575 short tons / hour to a maximum of 1718.1 short tons / hour, a dynamic range exceeding 7 times, reflecting the diversity of power plant operating conditions and the existence of extreme conditions. The data contains multiple typical operating modes, including the stable base load period, the high load peak period, the low load trough period, and the transitional phase of rapid load ramp-up and decline. The dataset contains several significant anomalous fluctuations. For example, the carbon emission rate plummeted to 845.55 short tons per hour at 9:00 AM on July 20, 2024, before rapidly rebounding. On October 28, 2024, an anomalous jump occurred, from 915.3 short tons per hour to 468.1 short tons per hour. These anomalies likely correspond to special operating conditions such as equipment maintenance, emergency shutdowns, or load adjustments, providing challenging samples for robustness testing of the model. The dataset has no missing values, continuous and uniformly distributed timestamps, and high-quality data collection, accurately reflecting the carbon emission characteristics of gas-fired power plants under different operating conditions. This provides a reliable experimental basis for the training, validation, and testing of time series prediction models.

[0076] In this embodiment, the ratio of training, validation, and test sets follows the official settings of the Time-MoE large-scale time-series model, with a ratio of 7:1:2. The training set is used to build the working condition knowledge base, and the test set is used to test the performance. The backtracking window step size is 512, and the prediction step size is 96 steps. Table 1 shows the performance comparison of the knowledge retrieval-enhanced time-series large-scale model under different fusion parameter configurations. As shown in Table 1, the baseline model is the Time-MoE zero-sample prediction result, with MSE and MAE of 0.4702 and 0.4866, respectively, representing the performance level without external knowledge fusion. The pure retrieval method, due to its complete reliance on historical similar fragments for substitution, fails to consider the dynamic evolution of the temporal context, resulting in a significant increase in error. The MSE rises to 1.2682, an increase of approximately 170% compared to the baseline, verifying that relying solely on retrieval generation cannot effectively characterize emission changes under complex working conditions. The fusion strategy experiment shows that reasonable fusion weights are necessary. α Setting the retrieval parameter k can significantly improve model performance. When α =0.95、 k =3、 Temp When the coefficient of performance (COP) is 0.01, the model achieves optimal performance, with the mean squared error (MSE) decreasing by 1.25% compared to the baseline, while the mean squared error (MAE) remains almost unchanged. This indicates that under this configuration, the time-series prediction results and the retrieved evidence achieve good complementarity. The above experiments verify that the retrieval enhancement mechanism of this invention can introduce historical evidence similar to the current operating conditions while maintaining the stability of the large-scale time-series model prediction. This achieves lightweight accuracy improvement without retraining the model, verifying the effectiveness of the knowledge retrieval enhancement paradigm in the CEMS carbon emission data gap completion task.

[0077] Table 1. Performance Comparison of Knowledge Retrieval Enhancement Temporal Series Large Scale Models under Different Fusion Parameter Configurations

[0078] The results in Table 1 show that the knowledge retrieval augmented time-series large-scale model in this embodiment outperforms the baseline model in terms of overall error metrics. In particular, with appropriate parameter settings, it can effectively improve the accuracy of data completion for missing data. To further verify the model's application in real-world missing data scenarios, this embodiment designed an experiment with artificially constructed missing data. By setting missing windows in normal CEMS data and using the proposed intelligent agent method for prediction and completion, its completion capability and stability were evaluated. Figure 6 To fill in the data in the rendering, by Figure 6As can be seen, the data imputation effect obtained by the technique of this application is better than the zero-shot prediction effect of the Time-MoE model alone, reducing MAE by 3.1% and MSE by 1.2%. Furthermore, comparing the top 10 predicted points (as shown in Table 2), compared with the traditional method using historical high values, the technique of this application can generate replacement results that are closer to the actual working conditions, significantly improving the rationality and accuracy of data imputation. Moreover, the agent can output an auditable report including the source of similarity, similarity score, contribution weight, and the reasoning for the final decision, greatly enhancing the transparency and credibility of the data repair process.

[0079] Table 2 Comparison of Failure Point Filling Effects

[0080] In summary, compared with existing conservative substitution methods based on monitoring availability and failure duration using averages, extreme values, or fixed rules, this embodiment effectively avoids the systematic biases that may result from a one-size-fits-all substitution through a collaborative mechanism of "retrieval enhancement—predictive imputation—evidence-based interpretation." While ensuring process compliance and verifiable results, it significantly improves the fit and verifiability between substitution values ​​and actual emission trajectories. The overall technical effects include: (1) Improved accuracy: This invention can effectively integrate the generalization ability of time series large model with the advantages of knowledge retrieval enhancement generation, and use similar working condition evidence to constrain prediction, characterize local trends, seasonality and variable load patterns. Compared with simple statistical substitution, it can significantly reduce reconstruction error and bias.

[0081] (2) Enhanced interpretability: Generate audit-friendly results containing “evidence fragments - similarity - contribution - reasons for adoption”, supporting internal and external review and retrospection.

[0082] (3) Robust handling of failures of different durations: Based on the multi-scale retrieval and integration mechanism, it has good adaptability to both short-term missing data and long-term interruption. When there is insufficient evidence, it can smoothly fall back to the rule-based backup strategy to ensure the robustness of the system.

[0083] (4) Low modification and easy expansion: It can be implemented by using the existing data collection and reporting system. The architecture has good scalability to covariates and legal provisions, which is convenient for further iteration and upgrading in subsequent versions.

[0084] (5) Dynamic updates: The historical database can be dynamically updated according to the actual operation, and at the same time, it continuously records various abnormal fluctuations in the operation status.

[0085] (6) Plug and play: The model does not need to be retrained and fine-tuned. Domain knowledge can be injected into the prediction model by connecting to the database.

[0086] (7) High timeliness: This intelligent agent can be directly deployed on the field terminal equipment and achieve rapid response through edge computing or lightweight inference. It can handle the situation of missing monitoring data in real time. Compared with manual repair and offline replenishment methods, it significantly improves the online operation efficiency of carbon emission monitoring system.

[0087] This application also provides an electronic device, which includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the above-described method. This electronic device can be any smart terminal, including tablet computers, in-vehicle computers, etc.

[0088] It is understood that the content of the above method embodiments is applicable to this device embodiment. The specific functions implemented by this device embodiment are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0089] Please see Figure 7 , Figure 7 The hardware structure of an electronic device according to another embodiment is illustrated. The electronic device includes: The processor 701 can be implemented using a general-purpose CPU (Central Processing Unit), microprocessor, application-specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of this application. The memory 702 can be implemented as a read-only memory (ROM), a static storage device, a dynamic storage device, or a random access memory (RAM). The memory 702 can store the operating system and other application programs. When the technical solutions provided in the embodiments of this specification are implemented through software or firmware, the relevant program code is stored in the memory 702 and is called and executed by the processor 701 using the methods described in the embodiments of this application. The input / output interface 703 is used to implement information input and output; The communication interface 704 is used to enable communication and interaction between this device and other devices. Communication can be achieved through wired means (such as USB, Ethernet cable, etc.) or wireless means (such as mobile network, WIFI, Bluetooth, etc.). Bus 705 transmits information between various components of the device (e.g., processor 701, memory 702, input / output interface 703, and communication interface 704); The processor 701, memory 702, input / output interface 703, and communication interface 704 are connected to each other within the device via bus 705.

[0090] This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described method.

[0091] It is understood that the content of the above method embodiments is applicable to this storage medium embodiment. The specific functions implemented in this storage medium embodiment are the same as those in the above method embodiments, and the beneficial effects achieved are also the same as those achieved in the above method embodiments.

[0092] Memory, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs and non-transitory computer-executable programs. Furthermore, memory may include high-speed random access memory, and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some embodiments, memory may optionally include memory remotely located relative to the processor, and these remote memories can be connected to the processor via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0093] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the above-described method.

[0094] It is understood that the content of the above method embodiments is applicable to the embodiments of this program product. The specific functions implemented in the embodiments of this program product are the same as those in the above method embodiments, and the beneficial effects achieved are also the same as those achieved in the above method embodiments. The executable computer program code or "code" used to perform the various embodiments can be written in high-level programming languages ​​such as C, C++, Python, Smalltalk, Java, JavaScript, Visual Basic, Structured Query Language (e.g., Transact-SQL), Perl, or in various other programming languages.

[0095] The embodiments described in this application are for the purpose of more clearly illustrating the technical solutions of the embodiments of this application, and do not constitute a limitation on the technical solutions provided by the embodiments of this application. As those skilled in the art will know, with the evolution of technology and the emergence of new application scenarios, the technical solutions provided by the embodiments of this application are also applicable to similar technical problems.

[0096] Those skilled in the art will understand that the technical solutions shown in the figures do not constitute a limitation on the embodiments of this application, and may include more or fewer steps than shown, or combine certain steps, or different steps.

[0097] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0098] Those skilled in the art will understand that all or some of the steps in the methods disclosed above, as well as the functional modules / units in the systems and devices, can be implemented as software, firmware, hardware, or suitable combinations thereof.

[0099] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in the specification and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms “comprising” and “having,” and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0100] It should be understood that in this application, "at least one (item)" means one or more, and "more than" means two or more. "And / or" is used to describe the relationship between related objects, indicating that three relationships can exist. For example, "A and / or B" can represent three cases: only A exists, only B exists, and both A and B exist simultaneously, where A and B can be singular or plural. The character " / " generally indicates that the preceding and following related objects are in an "or" relationship. "At least one (item) of the following" or similar expressions refer to any combination of these items, including any combination of single or plural items. For example, at least one (item) of a, b, or c can represent: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, and c can be single or multiple.

[0101] In the several embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of the units described above is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0102] The units described above as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0103] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0104] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes multiple instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this application. The aforementioned storage medium includes various media capable of storing programs, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0105] The preferred embodiments of the present application have been described above with reference to the accompanying drawings, but this does not limit the scope of the claims of the present application. Any modifications, equivalent substitutions, and improvements made by those skilled in the art without departing from the scope and substance of the embodiments of the present application shall be within the scope of the claims of the present application.

Claims

1. A method for handling expired CEMS carbon emission data, characterized in that, The method includes the following steps: A historical operating condition knowledge base is constructed, which stores multiple historical operating condition segments. Each segment includes a backtracking window sequence and its corresponding subsequent real carbon emission sequence. Real-time monitoring of CEMS carbon emission data streams and identification of periods of data failure; A backtracking window sequence of a preset length before the data failure period is extracted as the query key; Based on the query key, K similar historical operating condition segments are retrieved from the historical operating condition knowledge base, where K≥1; The backtracking window sequence and the K similar historical operating condition segments are input into the time series prediction model to generate the carbon emission prediction sequence for the data failure period, which serves as a substitute value. Output the replacement value and simultaneously generate an interpretability evidence report, which includes at least the identifier of the similar historical working condition segment used, similarity information, and contribution information of each similar historical working condition segment to the prediction result; The time-series prediction model is the Time-MoE model; the specific formula for generating the prediction sequence is: in, For the predicted sequence, To integrate weights, For time series prediction models, the backtracking window The original prediction output, This is the evidence vector obtained by weighted aggregation of subsequent sequences of the K similar historical work condition segments retrieved; The weights of the weighted aggregation are calculated using a temperature-based Softmax function, as shown in the formula: in, For the first Similarity score of each segment For temperature coefficient, This is the set of indexes for the retrieved fragments.

2. The method according to claim 1, characterized in that, The construction of the historical operating condition knowledge base includes: Multiple time segments are extracted from historical CEMS operation data using a sliding window approach; For each time segment, the data from the previous L times are defined as keys, and the carbon emission data from the next H times are defined as values, forming key-value pairs; After standardizing the keys in the key-value pairs to eliminate the influence of offsets, they are stored in the historical operating condition knowledge base.

3. The method according to claim 1, characterized in that, The step of retrieving K similar historical operating condition segments from the historical operating condition knowledge base based on the query key includes: Calculate the similarity between the query key and each fragment key in the knowledge base; Select one or more segments whose similarity exceeds a preset threshold; When there are segments exceeding the preset threshold, the top K segments are selected as search results based on similarity from high to low, where K≥1.

4. The method according to claim 1, characterized in that, The step of retrieving K similar historical operating condition segments from the historical operating condition knowledge base based on the query key includes: If the similarity of all retrieved historical working condition segments is lower than the preset safety threshold, the search enhancement prediction will be abandoned, and the method will automatically switch to a conservative data substitution method based on preset rules.

5. The method according to claim 1, characterized in that, The keys stored in the backtracking window sequence and / or historical operating condition knowledge base, in addition to the carbon emission sequence, are extended to one or more covariate features; the covariate features include at least one of power generation load, fuel parameters, unit start-up and shutdown status, and ambient temperature.

6. The method according to claim 1, characterized in that, The interpretability evidence report is structured data that can be directly read and verified by the audit system.

7. An electronic device, characterized in that, The electronic device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the method according to any one of claims 1 to 6.

8. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the method of any one of claims 1 to 6.

Images