Adaptive cost-sensitive assignment method for power system transient stability assessment

Traditional machine learning models are subject to changes in the data distribution. Once the topology of power grid changes, the distribution of power grid data will change, resulting in the consequence that the former transient stability assessment model based on machine learning is no longer applicable and needs to be retrained. However, the power network after the topology changes cannot provide enough data for the training of the new model. To solve the problem above, this paper proposes a method of power system transient stability assessment based on transfer learning, which fully taps the potential of past data and assists a small amount of newly acquired data in training the model. The simulation results on the CEPRI36 system show that the proposed method can effectively improve the accuracy of original classification models, especially for weak classifiers with insufficient generalization ability. The disadvantage of transfer learning lies in the long training time, which still needs to be improved.

Taking full advantage of synchrophasors provided by GPS-based wide-area measurement system (WAMS), a novel VBpMKL-based transient stability assessment (TSA) method through multifeature fusion is proposed in this paper. First, a group of classification features reflecting the transient stability characteristics of power systems are extracted from synchrophasors, and according to the different stages of the disturbance process they are broken into three nonoverlapped subsets; then a VBpMKL-based TSA model is built using multifeature fusion through combining feature spaces corresponding to each feature subset; and finally application of the proposed model to the IEEE 39-bus system and a real-world power system is demonstrated. The novelty of the proposed approach is that it improves the classification accuracy and reliability of TSA using multifeature fusion with synchrophasors. The application results on the test systems verify the effectiveness of the proposal.

Transient stability assessment with time-adaptive method based on spatial distribution

In this paper, an online power system transient stability assessment (TSA) problem is mapped as a two-class classification problem and a novel data mining algorithm the core vector machine (CVM) is proposed to solve the problem based on phasor measurement units (PMUs) big data. First of all, an offline training, online application framework is proposed, which contained four sub-steps, namely features selection, offline training, online application, and assessment evaluation. First, 24 features are selected to present the system status. Then in the offline training procedure, the PMU big data is generated by time domain simulation, and a CVM model is trained and tested. In the online application procedure, an interface between PMU data center and feature calculation program is set up to collect real time specific PMU big data and the CVM trained is applied to the TSA problem. Last but not least, the evaluation indices are calculated. Compared with other support vector machines, the proposed CVM based assessment algorithm has the higher precision, meanwhile, it has the least time consumption and space complexity. As long as online PMU big data are received, TSA can be done simultaneously. Case studies on the IEEE New England 39-bus system, and real systems in China and the U.S., exhibit the speed and effectiveness of the proposed algorithm.

An optimized algorithm for optimal power flow based on deep learning

Recently, data-driven methods have been widely assessed by researchers in the field of power system transient stability assessment (TSA). The differences in prediction difficulty among the samples are ignored by most previous studies. To address this problem, anchor loss (AL) is introduced, which can dynamically reshape loss values based on the prediction difficulty of samples. Thereby, easy samples are suppressed by reducing their loss values to avoid being paid too much attention when they are misclassified. Meanwhile, hard samples are emphasized by increasing their loss values, in order to be predicted correctly as much as possible. On basis of the AL, historical information in the model training process is considered. A novel loss function named historical information anchor loss (HIAL) is designed. The loss values can be adaptively rescaled according to the previous prediction results as well as the prediction difficulty of samples. Finally, the HIAL is combined with the deep brief network (DBN) and applied in the IEEE 39-bus system, and a realistic system is produced to verify its effectiveness. By incorporating prediction difficulty and historical training information, the accuracy (with a reduction in misjudgment rate exceeding 30%) and convergence speed of the TSA model can be significantly improved.

Convolutional neural network-based power system transient stability assessment and instability mode prediction

Transient voltage stability assessment of renewable energy grid based on residual SDE-Net

Deep Neural Network (DNN) has been introduced to power system transient stability assessment (TSA) and achieved quite remarkable success. However, since the DNNs are ‘blackbox’ models, it is very hard to reveal misbehaviors of the DNNs and repair them to get better performance. To address these challenges, this paper proposes to use Arachne, a fresh program repair technique for DNNs, to repair TSA DNNs. Arachne can fix DNN's specific misclassifications at little cost to overall accuracy. In TSA, missed detections is more severe to the system than false alarms due to the conservation requirements. Case studies have demonstrated the usefulness of the proposed method.

In recent years, the power system transient stability assessment (TSA) based on a data-driven method has been widely studied. However, the topology and modes of operation of power systems may change frequently due to the complex time-varying characteristics of power systems. which makes it difficult for prediction models trained on stationary distributed data to meet the requirements of online applications. When a new working situation scenario causes the prediction model accuracy not to meet the requirements, the model needs to be updated in real-time. With limited storage space, model capacity, and infinite new scenarios to be updated for learning, the model updates must be sustainable and scalable. Therefore, to address this problem, this paper introduces the continual learning Sliced Cramér Preservation (SCP) algorithm to perform update operations on the model. A deep residual shrinkage network (DRSN) is selected as a classifier to construct the TSA model of SCP-DRSN at the same time. With the SCP, the model can be extended and updated just by using the new scenarios data. The updated prediction model not only complements the prediction capability for new scenarios but also retains the prediction ability under old scenarios, which can avoid frequent updates of the model. The test results on a modified New England 10-machine 39-bus system and an IEEE 118-bus system show that the proposed method in this paper can effectively update and extend the prediction model under the condition of using only new scenarios data. The coverage of the updated model for new scenarios is improving.

In the transient stability assessment, machine learning based model has the problem that sample imbalance makes the model have a certain degree of evaluation tendency. From the loss function of the machine learning based model, it is found that the fitting degree to various samples after training can be mirrored by the loss function value of the samples. Therefore, a cost-sensitive strategy based on the imbalance degree of the loss function is proposed. Firstly, the machine learning based model is trained to obtain the mean value of loss functions of various samples. Then, the sample imbalance degree is calculated by the mean ratio of the loss function of unstable samples to that of stable samples. After that, the loss function is modified by the sample imbalance degree combined with the cost-sensitive strategy. Finally, to remedy the assessment inclination, the model is trained once again. When compared to conventional approaches, the suggested method comprehensively considers the impact of the imbalance in quantity and spatial distribution of samples. This model achieves higher global accuracy and greater corrective effect. Simulation findings on IEEE 39-bus and IEEE 145-bus systems are used to confirm the usefulness of the suggested strategy.

Machine learning approaches have exhibited high potential in power system transient stability assessment (TSA), yet their initial preparation stages of stability knowledge base generation (SKBG) based on time-domain simulations often undergo high computational costs. In fact, how to ease the heavy computational burden of SKBG without sacrificing the reliability is still a significant challenge. To address this problem, this paper develops a semi-supervised ensemble learning (SSEL) framework for reliable SKBG acceleration, without additional need for computing hardware upgrading. In particular, it performs detailed simulations for a minority of cases and fast simulations for the majority ones to reduce the total computation time. Considering the absence of stability status information of those fast simulated cases, an SSEL scheme is systematically designed to reliably label their stability status. Before implementing SSEL, two concise feature descriptors are first introduced to efficiently extract transient features from multiplex system trajectories. Then, all the cases are characterized in a unified feature space, whereby a series of semi-supervised support vector machines are trained in randomly formed subspaces. Afterward, these single machines are systematically combined to derive an enhanced SSEL model, which is able to make reliable and robust labeling decisions. Further, a backtrace strategy is carefully devised for SSEL, so as to maintain the high reliability of SKBG. Test results on the IEEE 39-bus system and the realistic GD Power Grid in South China illustrate the excellent performances of the proposed framework on SKBG acceleration.

This article presents the Lyapunov functions based on the artificial intelligence (AI) method for transient stability assessment and the determination of stability region (SR). First, Lyapunov stability theory and the definition of SR are introduced. Then, the characteristics of neural networks as a general function approximator are employed as the Lyapunov function learner, and the Lyapunov function is constructed combined with stochastic gradient descent (SGD). Then, the falsifier's task is to find the state vectors that violate Lyapunov stability conditions, and the counterexamples would be added to the training set for the function learner to accelerate convergence. After obtaining the Lyapunov function of power system, the estimation of SR boundary can be represented by the maximum level set of Lyapunov function. Finally, the IEEE 9-bus 3-machine system is used as test system to demonstrate the validity and effectiveness of the proposed construction method of Lyapunov functions for power system transient stability analysis.

Transient stability assessment (TSA) based on machine learning techniques is a classical problem of classification. In this paper, a new transient stability assessment model using the weighted random forest (WRF) is proposed. Random forest, which runs fast and can scale to very high dimensional datasets, is one of the most popular methods applied to TSA after faults occurred. However, in the actual operation of modern power system, unstable samples constitute only a very small minority of the cases. Besides, the traditional random forest algorithm does not work very well on imbalanced data. To tackle the problem, weighted random forest is adopted to lean the interest towards to the correct classification of rare class, and improve the prediction performance. Case study on the New England 39-bus test power system exhibits that WRF has better performance than the original random forest.

Machine learning is playing an increasingly important role in transient stability assessment. One of the key challenges of machine learning-based transient stability assessment in the case of dealing with large-scale and high-dimensional data samples of AC-DC power systems is to eliminate redundant variables to improve the training speed and quickly determine the optimal parameters of the adopted algorithms. Aiming at this problem, this paper proposes a Bayesian optimization-based XGBoost (B-XGBoost) method for transient stability assessment of power systems. The method uses serial integration of multiple regression trees to calculate the importance ranking of transient stability features. Combined with correlation analysis, the relationship between input data and transient stability can be explored more intuitively, and redundant variables can be eliminated; Bayesian optimization is used in training to quickly determine the best parameters of the model. Simulations are performed in a 39-node AC-DC system, and the results show that the B-XGBoost transient stability evaluation method has higher accuracy and a significant advantage in speed compared with other machine learning-based methods such as GBDT, DBN, RF, SVM.