Search

In this paper, we present a novel training method based on Baum-Welch algorithm for hidden Markov models (HMM), named as Comprehensive HMM (CompHMM), which changes the traditional approach of training HMM from positive examples only to be able to utilize both positive and negative examples in training HMMs. By comparison, our method outperformed the standard Baum-Welch method and another HMM discriminative training method significantly through both synthetic and real data in membership prediction task.

With the implementation of coal-to-electricity project, a large amount of electric heating loads is connected to the power grid, which causes impact on the power grid, but also brings new distribution network planning problem. In order to solve this problem, this paper establishes the model of coal-to-electricity project from two aspects of spatial layout and construction time sequence. In the solution of the model, the weighted Voronoi diagram method is firstly used to partition the area to form multiple partition schemes. Then, using the idea of dynamic programming, the problem is divided into several stages. In a single stage, the genetic algorithm is used to obtain the optimal substation planning scheme of the corresponding partition scheme, and the backtracking algorithm is used to generate the timing sequence decisions. Finally, the optimal planning scheme is obtained by calculating the sum of annual investment in each stage through reverse recursive method. The results show that the scheme can meet the requirements of distribution network in practical construction.

RNA-seq is one of the most widely used methods to probe gene expression and alternative splicing events (ASE) at the transcriptome scale. It often generates large amounts of complex sequencing data, which expands rapidly in large scale studies containing many samples. Quantifying short RNA reads distribution on reference genomes is a key step for most analyses of RNA-seq. This process requires a large amount of computational resource, including high-end computing hardware. With the accumulation of high throughput sequencing results from modern big-data biology, fast and efficient processing of RNA-seq data has become an urgent requirement. In this paper, we observe that transcriptome quantification is a highly repetitive task suitable for parallel computation. We report on a GPU based transcriptome quantification tool called Parallel Transcriptome Quantification Analyzer (Paean), which achieves over 20X acceleration in processing RNA-seq data compared to existing CPU-based methods when analyzing splicing changes, while producing comparable accuracy to conventional methods. To our knowledge, this is the first reported use of a hybrid CPU-GPU approach for transcriptome quantification analysis with RNA-seq data.

According to the existing problems in the supervision of on-site operation of electrical power system, an on-site operation monitoring and command system is designed. The on-site video is transmitted to command center via servicer by wireless network. But due to the instability of wireless network bandwidth, the real-time transmission of video can’ t be guaranteed. Even the transmission delay is long and the frames are lost seriously. In order to make the video transmission real-time and integrity, a video optimal transmission method based on the frame similarity is proposed. On the basis of the characteristics of H.264 video coding mode, the frame similarity parameter is marked to distinguish the frame importance. The network bandwidth changes is monitored in real time and the Prio-Drop scheduling algorithm is presented, so the important frames can be transmitted in advance. The test indicates that the optimal transmission method can improve the decoding quality of transmitted video, and achieve the requirements of real-time video transmission, which provided the technical guidance for enhancing the power supervision and strengthening emergency command.

Installing heat accumulators or electric boiler in the thermoelectric plant, and installing electric boiler with heat storage in the user side are three common modes to accommodate abandoned wind power. In order to select the best mode from three kind of abandoned wind heating mode, this paper combines regional characteristics and market mechanism, proposes the unit cost of accommodation index based on Life-Cycle Cost Theory and Benefit/Cost method, and the optimization model of abandoned wind heating mode considering equipment LCC is established. Finally, the ability of accommodating abandoned wind power of three modes are evaluated by numerical examples, and the adaptability to different heat load of three modes are compared and analyzed, the optimization model is verified effectively and correctly by numerical example.

A short-term load forecasting method of photovoltaic (PV) output based on weak correlation between meteorological date and PV output is proposed. The Pearson cor-relation coefficient method is used to make out the main affecting factor of PV output. Fuzzy clustering analysis method is used to attain similar days and then support vector regression (SVR) forecasting model with excellent learning ability for small sample is established. And then a two-stage parameter-determining method is carried out. Finally, the effectiveness of the proposed method is verified.