The hybrid energy storage system (HESS) is a key component for smoothing fluctuation of power in micro-grids. An appropriate configuration of energy storage capacity for micro-grids can effectively improve the system economy. A new method for HESS capacity allocation in micro-grids based on the artificial bee colony (ABC) algorithm is proposed. The method proposed a power allocation strategy based on low pass filter (LPF) and fuzzy control. The strategy coordinates battery and supercapacitor operation and improves the battery operation environment. The fuzzy control takes the state of charge (SOC) of the battery and supercapacitors as the input and the correction coefficient of the time constant of the LPF filter as the output. The filter time constant of the LPF is timely adjusted, and the SOC of the battery and supercapacitor is stable within the limited range so that the overcharge and over-discharge of the battery can be avoided, and the lifetime of the battery is increased. This method also exploits sub-algorithms for supercapacitors and battery capacity optimization. Besides, the Monte Carlo simulation of the statistic model is implemented to eliminate the influence of uncertain factors such as wind speed, light intensity and temperature. The ABC algorithm is used to optimize the capacity allocation of hybrid energy storage, which avoids the problem of low accuracy and being easy to fall into the local optimal solution of the supercapacitors and battery capacity allocation sub-algorithms, and the optimal allocation of the capacity of the HESS is determined. By using this method, the number of supercapacitors required for the HESS is unchanged, and the number of battery is reduced from 75 to 65, which proves the rationality and economy of the proposed method.

Concerning the prediction problems’accuracy of the state-of-charge(SOC) of the battery，this paper proposes a prediction method based on an improved genetic algorithm-radial basis function neural network for power battery charged state. The prediction method, based on intensity of information interaction and neural activity, adjusts the size of the neural network online and solves the problem that radial basis function neural network structure adjustment influences the accuracy of charged state prediction. The simulation results show that，compared with the method of radial basis function neural network based on genetic algorithm , the accuracy of charged state prediction is more stable and more precise.

This paper presents a method based improved particle swarm optimization optimizing PI control parameters. When the load changes, parallel active power filter, in the traditional PI control strategies, can not well control DC voltage fluctuations leading the deterioration of its dynamic performance. This paper proposes the improved particle swarm algorithm by increasing symbolic function of inertia weight parameter in particle swarm algorithm, and changing position of the random number in the speed update equation. The improved particle swarm algorithm can avoid "premature" or fall into local optimal solution, so it could get a more accurate solution set. Using improved particle swarm algorithm to solve PI controller optimization objective function, to get a more accurate PI control parameters. Simulation and experimental results show that mutations in the case load, the overshooting of an capacitor DC voltage compare with the traditional PI control reduces 5.5%, compared with Particle Swarm Optimization reduces 1.5%. Based on the simulation analyses and experimental results, it has been proved that this method can realize a comprehensive optimization.

Because the traditional Web service lacks of the description of the semantic level. Construction semantic Web service has become the development trend of traditional Web service and based on the semantic web, the service composition technology is also evolved with demand. On the basis of analyzing the traditional Web service composition technology, the paper researches the related critical technology of constructing semantic Web service，and then uses the method by combining the ontology modeling with the OWL-S to improve the traditional service composition model. Finally, the practicality and convenience of the improved model have been proved by the experiment.

The study of the application of the sliding mode observer method that estimates the state of charge. Based on the state space model of battery established on the model of improved EMF equivalent circuit, a sliding mode state observer is designed to help improve the jitter problem. Considering the nonlinear terms in the model for the analysis of the stability of the observer and the characteristics of the industry under its derivative, and using Lagrange mean value theory to guarantee the convergence conditions of the observer, the design parameters of the observer can thus be determined .Then, this thesis compares the simulation of this method under Matlab environment with the extended Kalman filter method. The results show that the method has higher estimation accuracy in the case of the same battery modeling errors. Therefore, the SOC estimation of the sliding mode observer can effectively reduce the state of charge estimation error introduced by the model error.