CN-122020470-A - Big data-based carbon emission anomaly detection method and system

CN122020470ACN 122020470 ACN122020470 ACN 122020470ACN-122020470-A

Abstract

The invention discloses a method and a system for detecting abnormal carbon emission based on big data, which comprise the steps of separating trend components and season components according to a standardized time sequence data sequence by adopting a time sequence decomposition method, determining the influence of production cycle factors contained in the trend components to obtain a decomposed trend sequence, extracting local dynamic characteristics from a marked potential abnormal sequence by adopting a sliding window technology, determining hidden mode changes in characteristic vectors to obtain a characteristic extraction sequence, isolating abnormal points through the characteristic extraction sequence by adopting an isolated forest algorithm, judging that abnormal fluctuation positions are confirmed if the isolation score is higher than a threshold value to obtain a confirmed abnormal position set, acquiring associated equipment state data aiming at the confirmed abnormal position set, performing cross verification, determining abnormal event types after verification, and obtaining a classified abnormal event list.

Inventors

ZHANG JUFENG
LI ZUOQUAN
YANG FENGFENG
WANG RUIYUN
MIAO ZAIQUAN
XIE YADONG
ZHENG CHAO
ZHANG JIANJIANG
Quan Jiye

Assignees

陇东学院

Dates

Publication Date: 20260512
Application Date: 20260129

Claims (8)

1. A method for detecting abnormal carbon emissions based on big data, comprising: obtaining a standardized time sequence data sequence according to an enterprise emission data source; according to the standardized time sequence data sequence, a decomposed trend sequence is obtained; Aiming at the decomposed trend sequence, acquiring external environment variable data and integrating the external environment variable data into the sequence, judging that if the fluctuation of the integrated sequence exceeds a preset threshold value, marking the integrated sequence as a potential abnormal segment, and obtaining a marked potential abnormal sequence; Extracting local dynamic characteristics from the marked potential abnormal sequences by adopting a sliding window technology, determining hidden mode changes in the characteristic vectors, and obtaining characteristic extraction sequences; obtaining a confirmed abnormal position set through the feature extraction sequence; and aiming at the confirmed abnormal position set, acquiring associated equipment state data, performing cross verification, determining the type of the verified abnormal event, and obtaining a classified abnormal event list.
2. The method for detecting abnormal carbon emission based on big data according to claim 1, wherein the standardized time series data sequence is obtained by acquiring the enterprise emission data source and performing the standardized processing on the sequence by adopting a preset sampling interval.
3. The method for detecting abnormal carbon emission based on big data according to claim 2, wherein the trend component and the season component are separated by a time series decomposition method based on a standardized time series data sequence, and the influence of the production cycle factor contained in the trend component is determined to obtain a decomposed trend sequence.
4. The method for detecting abnormal carbon emission based on big data according to claim 3, wherein the abnormal points are isolated by a feature extraction sequence by using an isolated forest algorithm, and if the isolation score is higher than a threshold value, the abnormal fluctuation position is confirmed to obtain a confirmed abnormal position set.
5. A carbon emission anomaly detection system based on big data, characterized by comprising: the first processing module is used for obtaining a standardized time sequence data sequence according to the enterprise emission data source; The second processing module is used for obtaining a decomposed trend sequence according to the standardized time sequence data sequence; The third processing module is used for acquiring external environment variable data aiming at the decomposed trend sequence and integrating the external environment variable data into the sequence, judging that if the fluctuation of the integrated sequence exceeds a preset threshold value, the integrated sequence is marked as a potential abnormal segment, and obtaining a marked potential abnormal sequence; The fourth processing module is used for extracting local dynamic characteristics from the marked potential abnormal sequences by adopting a sliding window technology, determining hidden mode changes in the characteristic vectors and obtaining a characteristic extraction sequence; the fifth processing module is used for obtaining a confirmed abnormal position set through the feature extraction sequence; And the sixth processing module is used for acquiring the state data of the associated equipment aiming at the confirmed abnormal position set, performing cross verification, determining the type of the verified abnormal event and obtaining a classified abnormal event list.
6. The big data based carbon emission anomaly detection system of claim 5, wherein the first processing module obtains the multi-source time series sequence by collecting enterprise emission data sources, and performs a normalization process on the sequence by using a preset sampling interval to obtain the normalized time series data sequence.
7. The big data based carbon emission anomaly detection system of claim 6, wherein the second processing module separates the trend component and the season component according to a standardized time series data sequence using a time series decomposition method, determines a production cycle factor influence contained in the trend component, and obtains the decomposed trend sequence.
8. The big data based carbon emission anomaly detection system of claim 7, wherein the fifth processing module isolates anomaly points using an isolated forest algorithm via a feature extraction sequence, determines that if the isolation score is above a threshold, determines an anomaly fluctuation location, and obtains a set of determined anomaly locations.

Description

Big data-based carbon emission anomaly detection method and system Technical Field The invention belongs to the technical field of information processing, and particularly relates to a carbon emission anomaly detection method and system based on big data. Background In the field of environmental protection and sustainable development, carbon emission monitoring and management is a global focus of attention, directly related to climate change control and ecological system stabilization. In particular, in the context of acceleration of industrial processes, abnormal fluctuations in the discharge amount of enterprises may have serious influence on the environment, so that it is important to monitor the discharge data in real time and recognize the abnormality. The research in this field is not only the basis for policy formulation, but also an important link for enterprises to fulfill social responsibilities. However, currently in emissions monitoring, a common problem is the lack of adaptability to complex dynamics. Many approaches often rely on fixed thresholds or simple rules that fail to cope with the varying patterns and abrupt changes hidden in the emission data. This limitation makes it difficult for some potential anomalies to be captured in time, especially in the face of large-scale, multi-source data, where the efficiency and accuracy of human intervention is extremely challenging. The technical difficulty of the deeper level is how to extract meaningful dynamic characteristics from mass data and accurately locate abnormal behaviors. Firstly, the time sequence characteristic of the emission data determines that the change trend has high complexity, is influenced by multiple factors such as production period, equipment state, external environment and the like, and is difficult to reveal the real fluctuation rule by simply relying on static analysis. Second, this complexity further results in concealment of abnormal behavior, such as a sudden increase in emissions due to equipment failure in a particular period of time by an enterprise, but due to the huge amount of data and subtle changes, such anomalies may be submerged in normal fluctuations and fail to trigger an alarm in time. Therefore, how to accurately identify hidden abnormal fluctuations in the dynamically changing emission data and distinguish them from normal fluctuations becomes a key problem to be solved currently. The problem is not only related to breakthrough in the technical level, but also directly influences whether an enterprise can take measures in time to reduce environmental damage, and whether the supervision department can effectively implement the emission reduction policy. Disclosure of Invention In order to solve the problems in the prior art, the invention provides a method and a system for detecting abnormal carbon emission based on big data. In order to achieve the above object, the present invention provides the following solutions: A carbon emission anomaly detection method based on big data comprises the following steps: obtaining a standardized time sequence data sequence according to an enterprise emission data source; according to the standardized time sequence data sequence, a decomposed trend sequence is obtained; Aiming at the decomposed trend sequence, acquiring external environment variable data and integrating the external environment variable data into the sequence, judging that if the fluctuation of the integrated sequence exceeds a preset threshold value, marking the integrated sequence as a potential abnormal segment, and obtaining a marked potential abnormal sequence; Extracting local dynamic characteristics from the marked potential abnormal sequences by adopting a sliding window technology, determining hidden mode changes in the characteristic vectors, and obtaining characteristic extraction sequences; obtaining a confirmed abnormal position set through the feature extraction sequence; and aiming at the confirmed abnormal position set, acquiring associated equipment state data, performing cross verification, determining the type of the verified abnormal event, and obtaining a classified abnormal event list. Preferably, the multi-source time sequence is acquired by collecting enterprise emission data sources, and the sequence is subjected to standardized processing by adopting a preset sampling interval to obtain a standardized time sequence data sequence. Preferably, the trend component and the season component are separated according to the standardized time sequence data sequence by adopting a time sequence decomposition method, and the influence of the production cycle factors contained in the trend component is determined, so as to obtain a decomposed trend sequence. Preferably, the feature extraction sequence is used for isolating abnormal points by adopting an isolated forest algorithm, and if the isolation score is higher than a threshold value, the abnormal fluctuation position is confirmed, s