Vol. 05, No. 06 [June 2019]

Based on the deep learning method, a feature reduction method of state monitoring index based on autoencoder network is first proposed, and then a convolution neural network state monitoring classification model is built. The feature vectors of dimensionality reduction generated by autoencoders are used as input of convolution neural network, and the multi classifier of transformer state is realized through convolution neural network classifier. The test of the monitoring data of the transformer of the proposed method using test results prove the correctness and effectiveness of the feature reduction method based on auto encoder, also proved the effectiveness of the convolutional neural network state monitoring classification model based on classification performance is better than that of SVM and MLP.
Keywords:state monitoring; autoencoder; convolution neural network; transformer

Marine Propellers of different types have been used in vessel propulsion systems. Surface Piercing Propeller or Semi submerged propellers (SPP) are today used for high-speed crafts. These propellers operate at the surface of the water for a superior performance, and thus they are subjected to severe hydrodynamic pressure. SPPs are commonly made of stainless steel materials having the required stiffness and strength to withstand the applied loads. Designing a novel composite SPP is the basic purpose of our investigation for its lower friction, no corrosion, and low detestability. The mechanical loads considered in our work are the hydrodynamic pressure, the centrifugal force, and the gravity. Through a boundary element analysis of the unsteady flow characteristics, we obtained the non-uniform distribution of the hydrodynamic pressure on the surface of the blades. The fluid dynamic analysis has been performed in an apart-related investigation, which only was presented and used its result as the input to the structural analysis of our work. In order to have a basis for the comparison, we analyzed a stainless steel propeller in addition to the composite one.
Keywords: PSemi-Submersible Propeller, Composite, MARC, AUTOCAD

With the continuous expansion of the scale of new energy power system, voltage collapse will seriously affect the safe and stable operation of power system. To avoid voltage collapse, it is necessary to predict the distance between the system operation state and the critical point of voltage stability, that is, voltage stability margin. Through the analysis of voltage stability margin, power grid operators can judge the current operation state of power grid, and then know whether the current system is static voltage stability. Traditional static voltage stability assessment methods are mostly based on complex power flow equations. Because of manual operation, the speed of solution cannot meet the demand of online real-time prediction. In order to satisfy the rapid automatic evaluation of system voltage stability, artificial intelligence technology is introduced in the current research, but most of the research focuses on improving the speed of off-line training. When noise data is mixed into the collected samples, noise cannot be filtered well, which leads to the poor fault tolerance rate of the training model and poor power supply. Pressure stability margin is predicted. Combining the advantages of Stacked Denoising Auto Encoder (SDAE) and Support Vector Regression (SVR) classifiers, this paper proposes a hybrid prediction model based on SDA-SVR, which can reduce input dimension and filter noise to reconstruct input. Over-stacked multi-de-noising self-encoding filters the mixed noise input data hierarchically, reconstructs the input, and directly inputs the reconstructed data to the support vector regression classifier, and then obtains the static voltage stability margin. In this paper, the model is validated by the simulation data of 9-node power network of WSCC 3 generators. Experiments show that the proposed model can reduce dimension and fault-tolerant processing of large-scale power network data in high latitudes. The model trained can effectively reduce the prediction error on the test sample set and achieve accurate prediction of static voltage stability margin. It has good adaptability, generalization ability and real-time.
Keywords: Power System Static Voltage Stability; Support Vector Regression; Stacked Denoising Self-Encoder; Deep Feature Learning; Deep Learning Prediction

The methodology was used to obtain new functions of distribution of normal probability and extreme value by Bayesian inference and stochastic mixing of Gaussians. The proposed methodology is oriented to data with Gaussian behavior and consists of adjusting the normal distribution function to the data of time series of maximum temperature data, to give a behavior of the temperatures maximum, later we use the Bayesian inference for normal data, in this case we are looking for behavior and trends with new Gaussian functions and extreme value. To validate the model we use the following statistical estimators: measurement of the root of the quadratic error, quadratic error, coefficient of determination and prediction approximation.
Using the new means and variances of the new functions of extreme distribution we generate two new functions of normal distribution a minimum and a maximum. Thus already having the three normal probability distribution functions, the adjusted first and the two new normal distribution functions we introduce the Gaussian stochastic mixing method to give a new function of general normal probability distribution for the trend of the maximum temperatures for Mexico City. In addition to making a temperature forecast with the WRF for comparison. The database that is used is from the page of the City of Mexico http://www.aire.cdmx.gob.mx/
Keywords: Temperature, Random and Extreme Variable Distribution Functions, Bayesian Inference, WRF.

In Cloud computing day by day number of users are increasing with large amount of data processing requirements. The biggest challenge for cloud data centers is how to service millions of requests arriving very frequently from end users efficiently and correctly. Therefore, Load balancing is the optimal approach solving the issue of growing demands for the resources of data centers. The need to efficient and powerful load balancing algorithms is one of the most important issues in cloud computing to improve the performance. This paper proposed hybrid load balancing algorithm to improve the performance and efficiency in cloud computing environment. The algorithm considers the advantages of random, greedy, and Throttled algorithms. The hybrid algorithm has been evaluated and compared with other algorithms using cloud Analyst simulator in hardware complexity environment. The experiment results show that the proposed algorithm improves the average response time and average processing time compared with other algorithms. This approach is used to minimize the response time, avoid the bottleneck problem, maximize the services and reduce the machine cost.
Keywords: Cloud computing, Load Balancing, Data Center, Virtual Machines, Cloud Analyst