With the increasing integration of measurement devices, the control and observation of power distribution systems have become significantly more dependent on cyberspace, making them more vulnerable to false data injection attacks (FDIAs). Contrary to the detection of FDIAs in power transmission systems, less attention has been paid to power distribution systems due to poor data quality, large volume sizes, and unbalanced data. This article proposes a dual-stage localization detection strategy for FDIAs in power distribution systems to detect and localize stealth FDIAs. Considering the time scale, arithmetic, and calculation overhead, this strategy can be transformed into two stages: presence detection and localization detection. Specifically, a cost-sensitive loss convolutional neural network based on Gaussian mixture autoencoder architecture is leveraged to capture data features from unbalanced data in presence detection. In localization detection, a Markov chain based on a cumulative state transfer probability (CSTP) is leveraged to locate the FDIAs exactly after presence detection. In this subject, presence detection can support the operator in rapidly filtering out compromised data, and localization detection can drive the control center to deploy countermeasures accurately. Based on the adjusted IEEE 14 and 118-bus test systems, numerical results indicate the effectiveness of the proposed strategy.

In real-world vehicle operation, lithium-ion battery capacity cannot be directly measured, and on-board charging data are often noisy and heterogeneous. To address these issues, this paper proposes a unified framework for monthly capacity estimation and multi-step forecasting based on real-world charging data. Monthly capacity labels are first constructed, and monthly feature maps are developed to characterize charging behavior. A convolutional neural network (CNN)-based model is then used for monthly capacity estimation, while a CNN-Transformer+Multi-Head network is designed for multi-step forecasting. Experimental results show that the proposed method achieves a mean absolute error (MAE) of 0.9843 Ah, a root mean square error (RMSE) of 1.2087 Ah, and a mean absolute percentage error (MAPE) of 0.8034% for monthly capacity estimation on the test set. For multi-step forecasting, the proposed model attains the best overall performance across all three forecasting horizons, with the MAPE at t+3 reaching 1.0957%. Further analysis shows that a 3-month input window provides a good balance between forecasting accuracy and computational cost. These results confirm the effectiveness of the proposed framework under real-world vehicle operating conditions.

The increasing use of energy conversion facilities has highlighted the interdependence between energy systems. Current literature has paid enormous attention to integrated heat-power networks and particularly focused on a holistic optimization framework. However, distribution networks and district heating networks belong to different entities and hold varied interests. In this paper, we propose a market model of integrated heat-power systems and a distributed algorithm for solving the equilibrium point. In the market model, energy trading between the heat market and the power market are taken into account, as well as the response of heterogeneous energy consumers to price signals from the markets. Besides, electricity and heat consumers are separately modeled based on corresponding features to make the modelling more accurate. In the solution method, a distributed algorithm based on fixed-point iteration is implemented after the model reformulation, to seek the market equilibrium with sensitive information of both networks well preserved. Numerical experiments are conducted on a testing system to validate the proposed model and the solution method.

This paper responds to official policies by exploring the positive role of electric power data in areas such as financial risk control. It aims to provide decision support for State Grid Corporation in data monetization and innovative business models, while addressing pain points for governments, financial institutions, and other businesses. By integrating electric power data, the project establishes an enterprise credit evaluation model to overcome the limitations of traditional credit profiling. Leveraging data mining methods and statistical modeling techniques, a credit scorecard model is developed based on electric power data, enabling the calculation of credit scores and corresponding risk levels. This model enhances risk prevention and decision-making effectiveness for relevant institutions. The project contributes to the systematic, normalized, and sustainable development of electric power big data credit reporting, serving other credit markets and promoting the improvement of various credit platforms. The findings provide a foundation for market participants to assess overall risk situations and improve credit risk management. This paper demonstrates the value of integrating electric power data in credit assessment and highlights its potential for enhancing risk control and decision-making processes.

When the system fails, a large amount of alarm information floods into the dispatching center in a short period of time, making it difficult to quickly diagnose and process alarm information. Dispatchers rely on a wealth of expertise, historical and real-time grid data to make optimal decisions based on changing conditions. Aiming at the complex power grid alarm information and high requirements for real-time judgment, this paper proposes a method for grid alarm information analysis and decision-making based on knowledge graph, which condenses the knowledge of power grid topology, operating procedures, disposal plans and other knowledge represented by a large number of texts into a knowledge graph that can be represented, operable and reasonable.

As a new cyber-attack method in power cyber physical systems, false-data-injection attacks (FDIAs) mainly disturb the operating state of power systems by tampering with the measurement data of sensors, thereby avoiding bad-data detection by the power grid and threatening the security of power systems. However, existing FDIA detection methods usually only focus on the detection feature extraction between false data and normal data, ignoring the feature correlation that easily produces diverse data redundancy, resulting in the significant difficulty of detecting false-data-injection attacks. To address the above problem, we propose a multi-source self-attention data fusion model for designing an efficient FDIA detection method. The proposed data fusing model firstly employs a temporal alignment technique to integrate the collected multi-source sensing data to the identical time dimension. Subsequently, a symmetric hybrid deep network model is built by symmetrically combining long short-term memory (LSTM) and a convolution neural network (CNN), which can effectively extract hybrid features for different multi-source sensing data. Furthermore, we design a self-attention module to further eliminate hybrid feature redundancy and aggregate the differences between attack-data features and normal-data features. Finally, the extracted features and their weights are integrated to implement false-data-injection attack detection using a single convolution operation. Extensive simulations are performed over IEEE14 node test systems and IEEE118 node test systems; the experimental results demonstrate that our model can achieve better data fusion effects and presents a superior detection performance compared with the state-of-the-art.

The analysis and utilization of massive recording data of distribution grid fault indicator is beneficial to improve the effect of distribution grid fault diagnosis. In this paper, the distribution grid monitoring are realized by the random matrix theory (RMT) of high dimensional statistical analysis. The fault diagnosis method based on RMT has the advantages of no need for detailed distribution grid topology, comprehensive utilization of wide-area spatiotemporal data, and observation from a muti-dimensional view. By explaining the application principle of limit spectrum distribution function, the linear eigenvalue statistics (LES) is proposed as the state monitoring index. Distribution grid fault diagnosis based on fault indicator and RMT provides a new data-driven method for distribution grid fault diagnosis while efficiently utilizing massive fault indicator data.

In recent years, with the clean energy consumption, decentralized supply and market-oriented trading, the composition of power system has become increasingly complex. As one of the most important consumer groups, the mining of residents' power consumption behavior is of great value to strengthen demand side management, improve energy efficiency and promote the development of smart grid. Therefore, this paper studies the pattern recognition and associated factors of power consumption behavior based on unsupervised clustering and Apriori. We use the hourly load curve of Shanghai residents from 2016 to 2018 to carry out the experiment. According to the results of the survey, we distinguish the single household characteristics and combined household characteristics, analyze their relationship with these typical power consumption modes, and eliminate the impact of unbalanced distribution of categories. The experimental results show that socio-economic factors, environmental cognitive factors and housing factors will affect Chinese residents' power consumption behavior to varying degrees. The association rules of combined household characteristics formed in different seasons are also quite different.

This paper explores the fusion and application of big data mining and artificial intelligence technology to delve into the value of power data assets. It advances the research on the business model of power data in enterprise credit and explores the feasibility of implementing power data assets. Based on the exploration of business models and existing project databases, the paper, considering the characteristics of the power industry and data foundation, uses data mining methods, label system construction methods, and credit evaluation system methods to construct an attentional convolution neural network-based credit rating (ACNNCR) model for power big data. Utilizing clustering algorithms, expert rules, statistical modelling, mining algorithms, etc., the paper develops a set of power credit label models, including factual labels, rule-based labels, and predictive labels, within dimensions such as user attributes, electricity usage characteristics, and credit features, based on non-residential user data. A total of 185 features are successfully established for attentional convolution neural network model, forming a new type of power data asset. Case verification on real-life datasets is used to validate the effectiveness of the constructed power credit labels.

In the daily work of electric power marketing, for all kinds of risks in the electric power marketing business to carry out systematic sorting and in-depth analysis, can effectively reduce the production and operation risks of electric power enterprises.At present, power companies still use more traditional power marketing model, the lack of lean analysis of business data and business risk control, enterprise production and operation there are more problems.For example, in terms of management, the current management system and management level is relatively backward.Electricity marketing work directly affects the operating efficiency of electric power enterprises, and the abovementioned issues have increased the operating risks of electric power enterprises to a certain extent, bringing more uncontrollable factors to the production and operation of electric power enterprises.For the current situation of power marketing risk control work, this paper uses AHP hierarchical analysis and power data to analyze business users.By constructing the judgment matrix and calculating the affiliation degree, the credit scoring rating system of enterprise users is built and a new credit management model for enterprise users is proposed.The experimental results show that our method can effectively evaluate the credit of corporate users.

To effectively analyze multi-type loads characteristics in extreme weather scenarios, a multi-type loads characteristics analysis method based on principal component analysis (PCA) and K-means clustering algorithm (PCA-K-means) model is proposed in this paper. Meteorological indicators were downscaled by PCA. The principal component with the most significant impact on the load is identified through correlation analysis and used to construct the principal component index. Integrating the principal component index and K-means clustering algorithm, the load data in extreme weather scenarios are clustered to form figures of multi-type loads associated with features, and the load characteristics are analyzed. The validity of the model is verified by arithmetic examples.

A knowledge graph is an intelligent database that integrates artificial intelligence technology and traditional database. It is a knowledge base that represents concepts, entities and their relationships in the objective world in the form of a graph. It can well reflect the relationship between entities. In the current era of big data, although the accuracy of abnormal value detection of power data is much higher than that of traditional methods, big data is lack of explicability, so it is difficult to trace the origin of abnormal values. Abnormal values caused by errors must also be checked manually on site, which wastes human and material resources. As a knowledge embodiment of "big data + artificial intelligence", it has natural advantages in the integrity and interpretability of behavior description. Therefore, this paper studies and discusses the relevant theories and technologies of knowledge graph construction and anomaly detection algorithm. Aiming at the power database table processing process that will cause errors, this paper constructs the knowledge graph and uses the Neo4j diagram database for storage, so as to visualize the data processing process, clarify the data processing relationship, and facilitate the tracking and traceability of the data. In addition, it also carries out experiments and analysis of different anomaly detection algorithms on the sample power consumption data.

False data injection attack (FDIA) is a deliberate modification of measurement data collected by the power grid using vulnerabilities in power grid state estimation, resulting in erroneous judgments made by the power grid control center. As a symmetrical defense scheme, FDIA detection usually uses machine learning methods to detect attack samples. However, existing detection models for FDIA typically require large-scale training samples, which are difficult to obtain in practical scenarios, making it difficult for detection models to achieve effective detection performance. In light of this, this paper proposes a novel FDIA sample generation method to construct large-scale attack samples by introducing a hybrid Laplacian model capable of accurately fitting the distribution of data changes. First, we analyze the large-scale power system sensing measurement data and establish the data distribution model of symmetric Laplace distribution. Furthermore, a hybrid Laplace-domain symmetric distribution model with multi-dimensional component parameters is constructed, which can induce a deliberate deviation in the state estimation from its safe value by injecting into the power system measurement. Due to the influence of the multivariate parameters of the hybrid Laplace-domain distribution model, the sample deviation generated by this model can not only obtain an efficient attack effect, but also effectively avoid the recognition of the FDIA detection model. Extensive experiments are carried out over IEEE 14-bus and IEEE 118-bus test systems. The corresponding results unequivocally demonstrate that our proposed attack method can quickly construct large-scale FDIA attack samples and exhibit significantly higher resistance to detection by state-of-the-art detection models, while also offering superior concealment capabilities compared to traditional FDIA approaches.

With the rapid advancement of national strategies such as "Internet+" and "Made in China 2025", industrial control systems have been widely used in various industries such as energy, municipalities, transportation, water conservancy and aerospace, etc. The security of industrial control networks is always affecting the lifeline of the national economy. Therefore, the security of industrial control networks has not been given enough attention, resulting in frequent industrial control network security incidents, which makes us realize the importance of creating an industrial control network situational awareness system that integrates assessment and prediction. In this paper, we study the situational awareness technology of industrial control network based on big data, and integrate situational extraction as the premise, situational assessment as the core, and situational prediction as the goal to comprehensively sense the situational awareness of industrial control network system. Firstly, we adopt two ways of industrial control network data collection in the situational extraction, one is the use of traffic mirroring bypass access to sensor-aware terminals, without affecting the original production services on the premise of network traffic data collection. The second is the use of WireShark tools to achieve industrial control network traffic packet collection and statistics of traffic packets per second, analysis of the current network state, build Hadoop big data platform to achieve offline data pre-processing and feature extraction, and build Flink and TensorFlow model for graph neural network model training and complete prediction. Secondly, we use a combination of hierarchical analysis and correlation analysis to evaluate the situation, and use the evaluation graph to show the current network security state. Thirdly, the situation prediction is done by training the graph neural network model offline. We use the Flink real-time computation engine to read the data of the industrial control network into the graph neural network model in real time, which is used to enhance the feature representation of each node and to predict the probability of anomaly occurrence of the industrial control network in the future period. Finally, the graph neural network model of this paper is compared with traditional neural networks and machine learning models, and the accuracy and false alarm rate indexes are used to demonstrate the high accuracy and robustness of this model.

Power consumption forecasting is an important part of the macro planning of the industry and energy sector, and accurate forecasting of power load is very important for power grid management and power dispatching. At present, most of the power load forecasting takes the region as the object, but residents and small and medium-sized enterprise users are the basic units of electricity consumption, and their power load forecasting is as important as regional power load forecasting. compared with the regional power load, the electricity load of residents and small and medium-sized enterprises is more uncertain and more difficult to forecast. Therefore, this study combines the adaptive spectral clustering (ASC) method with the support vector quantile regression model (SVQR) to analyze the electricity consumption behavior of smart grid users and predict the residential power load. In this paper, the grid search is used to optimize the parameters of the Gaussian kernel SVQR model (GSVQR) to predict the power load, and compare it with other algorithms. From the two error evaluation index values of MAPE and pinball loss, the prediction effect of the GSVQR model is the best. In order to effectively provide uncertain information of power load, the GSVQR algorithm is used to predict the load of ultra-high energy consumption users and medium energy consumption users at any time in the future. Extensive experimental results show that: compared with other models, the prediction accuracy of the GSVQR model is higher; and the prediction results of the GSVQR model still have high reliability. Therefore, the method used in this paper can solve the problem of uncertainty of load forecasting.

Through research and literature review, we analyzed the model architecture of credit rating in credit industry and its business model, including data collection, model design, service targets, fees and profit model. We found that corporate credit, as the foundation of social credit system, is the cornerstone of national life and business economy development. And the basis of credit evaluation formed by power credit is to build a power credit label with power characteristics and unique application scenarios and values based on power data. Therefore, in this paper, using technologies such as clustering algorithms, expert rules, statistical modeling, mining algorithms and other mining algorithms and electricity big data, where the electricity big data is the data on transaction tariff, electricity sales, electricity consumption customers, etc. We build a set of electricity credit tags including fact tags, rule tags and model tags for the existing Shanghai non-residential customer data database. From the experimental results, we can see that the model can explore the power consumption characteristics of enterprise users, tap the behavioral characteristics of users’ production and life, quickly understand the abnormalities of users’ power consumption behavior and avoid the occurrence of undesirable events.

False data injection attacks (FDIA) destroy the integrity of information transmission by evading the bad data detection mechanism, and thus affects the stability of power cyber-physical systems (PCPS). Existing studies simply introduce complex neural network models for FDIA detection, ignoring spatial-temporal correlation and interpretability of neural networks. As a result, the accuracy and reliability of false detection may be negatively affected. To address the challenges above, this paper proposes an interpretable deep learning framework based on the spatial-temporal attention mechanism. Firstly, based on the gated recurrent unit (GRU), a dual attention mechanism is designed by combining spatial and temporal features of deep neural network to dynamically mine the potential correlations between the FDIA detection and the input features. Besides, the quantification of attention weights is introduced to interpret the spatial-temporal correlations between normal and attack data, which can effectively enhance the interpretability and reliability of detection results. Finally, based on the IEEE 14-bus test system and real operation data, simulations are conducted and the results show that the proposed STAGN model can detect FDIA effectively, has higher accuracy and stability than the latest detection models, and also has reasonable interpretability.