Traditional time series imputation methods typically assume that time series data is derived from a linear dynamic system. However, the real-world time series show more non-linear characteristics. Therefore, a time series imputation model based on Long Short-Term Memory (LSTM) network with residual connection, called RSI-LSTM (ReSidual Imputation Long-Short Term Memory), was proposed to capture the non-linear dynamic characteristics of time series effectively and mine the potential relation between missing data and recent non-missing data. Specifically, the LSTM network was used to model the underlying non-linear dynamic characteristics of time series, meanwhile, the residual connection was introduced to mine the connection between the historical values and the missing value to improve the imputation capability of the model. Firstly, RSI-LSTM was applied to impute the missing data of the univariate daily power supply dataset, and then on the power load dataset of the 9th Electrical Engineering Mathematical Modeling Competition problem A, the meteorological factors were introduced as the multivariate input of RSI-LSTM to improve the imputation performance of the model on missing value in the time series. Furthermore, two general multivariate time series datasets were used to verify the missing value imputation ability of the model. Experimental results show that compared with LSTM, RSI-LSTM can obtain better imputation performance, and has the Mean Square Error (MSE) 10% lower than LSTM generally on both univariate and multivariate datasets.

In recent years, the short-term power load prediction model built with Recurrent Neural Network (RNN) as main part has achieved excellent performance in short-term power load forecasting. However, RNN cannot effectively capture the multi-scale temporal features in short-term power load data, making it difficult to further improve the load forecasting accuracy. To capture the multi-scale temporal features in short-term power load data, a short-term power load prediction model based on Multi-scale Skip Deep Long Short-Term Memory (MSD-LSTM) was proposed. Specifically, a forecasting model was built with LSTM (Long Short-Term Memory) as main part, which was able to better capture long short-term temporal dependencies, thereby alleviating the problem that important information is easily lost when encountering the long time series. Furthermore, a multi-layer LSTM architecture was adopted and different skip connection numbers were set for the layers, enabling different layers of MSD-LSTM can capture the features with different time scales. Finally, a fully connected layer was introduced to fuse the multi-scale temporal features extracted by different layers, and the obtained fusion feature was used to perform the short-term power load prediction. Experimental results show that compared with LSTM, MSD-LSTM achieves lower Mean Square Error (MSE) with the reduction of 10% in general. It can be seen that MSD-LSTM can better capture multi-scale temporal features in short-term power load data, thereby improving the accuracy of short-term power load forecasting.

When the traditional genetic algorithm is used to solve the computation-intensive task, the execution time of the fitness function increases rapidly, and the convergence rate of the algorithm is very low when the population size or generation increases. A "coarse-grained combined with master-slave" HyBrid Parallel Genetic Algorithm (HBPGA) was designed and implemented on Sunway "TaihuLight" supercomputer which is ranked first in the latest TOP500 list. Two-level parallel architecture was used and two different programming models, MPI and Athread were combined. Compared with the traditional genetic algorithm implemented on single-core or multi-core cluster with single-level parallel architecture, the algorithm using two-level parallel architecture was implemented on the Sunway many-core processors, better performance and higher speedup ratio were achieved. In the experiment, when using 16×64 CPEs (Computing Processing Elements), the maximum speedup can reach 544, and the CPE speedup ratio is more than 31.

AdaBoost is one of the most popular boosting algorithms in the area of data mining. By analyzing the disadvantages of the traditional AdaBoost using linear combination of the basic classifiers, a new algorithm was proposed, which changed the traditional linear addition into a nonlinear combination, and replaced the constant weights acquired in the training stage by a series of dynamic parameters based on the statistics of the K-nearest neighbors and decided by the instances in the predicting stage. In this way, the weight of each basic classifier was closer to reality. The experimental results show that, compared to the traditional AdaBoost, the new algorithm can increase the prediction accuracy nearly seven percentage points at most. The new algorithm is more accurate and it can achieve higher classification accuracy for most data sets.

Prevailing cloud storage systems normally use master/slave structure, which may cause performance bottlenecks and scalability problems in some extreme cases. So, fully distributed cloud storage system based on Distributed Hash Table (DHT) technology is becoming a new choice. How to solve load balancing problem for nodes, is the key for this technology to be applicable. The Kademlia algorithm was used to locate storage target in cloud storage system and its load balancing performance was investigated. Considering the load balancing performance of the algorithm significantly decreased in heterogeneous environment, an improved algorithm was proposed, which considered heterogeneous nodes and their storage capacities and distributed loads according to the storage capacity of each node. The simulation results show that the proposed algorithm can effectively improve load balance performance of the system. Compared with the original algorithm, after running a long period (more than 1500 hours in simulation), the number of overloaded nodes in system dropped at an average percentage 7.0%(light load) to 33.7%(heavy load), file saving success rate increased at an average percentage 27.2%(light load) to 35.1%(heavy load), and also its communication overhead is acceptable.

In a Base Station (BS) cooperation system, the capacity gains of two BSs by cooperating with each other are different owing to the channel asymmetry. Thus, in the process of selecting cooperative BS, when one BS wants to coordinate but another does not want to cooperate with it and prefers others, it will be difficult to judge whether to group the two into a cluster or not and the whole system is capacity-limited. To demonstrate this, the authors proposed an overlapped clustering scheme and then designed a joint precoding algorithm called Zero Force-Tomlinson-Harashima Precoding (ZF-THP) scheme. In this scheme, several BSs were adjusted to be overlapped and THP technique was used to eliminate the interference caused by overlapped BSs. The simulation results show that the proposed scheme solves the clustering contradiction well, efficiently increases the system capacity and enhances the system fairness.

In view of the problem of attribute mapping techniques in materialized data integration, the authors proposed a search algorithm of attribute mapping based on combined similarity calculation (SACS). The proposed algorithm was established through intuitive calculation factors and combined formula to traverses attribute mapping in data sources. The algorithm avoids the sample selection problem of machine learning in traditional attribute mapping techniques, and improves the precision rate and recall rate for attribute mapping.