Research On Some Key Issues For Building Electrical Energy Saving Based On Neural Network

Building is an important source of energy consumption. It is essential for long-term development of national economy to study on how to reduce the comprehensive energy consumption of buildings, improve the internal energy system efficiency and enhance the comfort of the environment. Therefore, using the theory and method of artificial intelligence, the research on building electrical energy saving control is carried out to reduce the energy consuming of building electrical, which has both theoretical significance and practical value.The key issues for building electrical energy saving have been studied from three aspects in the dissertation, including the identification of harmonic source and harmonic suppression of the pure capacitance reactive compensation, the predictive control and decoupling control of variable air volume conditioning system, as well as the evaluation of building electrical energy saving.Aiming at reducing the harm of harmonic to building electrical and decreasing the energy consumption, the dissertation is focused on the research of the identification of harmonic source and harmonic suppression of the pure capacitance reactive compensation. However, the key element of harmonic control is to monitor the parameters of harmonic distortion in real-time, then to evaluate the parameters comprehensive. Therefore, the harmonic source diagnosis method based on improved BP neural network is proposed, which takes the aberration rate of measured data as the input feature vectors, to realize the diagnosis and classify of harmonic interference. Results show that the network outputs are near to the expectations, with the maximum error 0.09%. Meanwhile, the harmonic resolution strategy of taking pure capacitance as reactive power compensation is adopted. Testing and analysis show that the defects exist in the traditional pure capacitance reactive compensation strategy when the system contains higher harmonic. Therefore, the method of capacitor in series with inductance reactor is used in reactive power compensation loop to change the impedance characteristics under harmonic and make the reactive power compensation loop not present capacitive to avoid harmonic amplification and resonant generation.Based on the study of harmonic suppression, the research on predictive control and decoupling control of air conditioning system is carried out. It is well known that the central air conditioning system is multivariate, complicated and time-varying. And there are serious nonlinear, large time delay and strong coupling relationships between the process elements, resulting in difficult control of central air conditioning system, which leads to a great waste of electricity energy. Therefore, the predictive controller of neural network structure is designed to create a control model, which is used to adjust the NNC weights. Then a single hidden layer with three inputs and single output structure is achieved. Meanwhile, the fuzzy neural network prediction control method is proposed. The method has a dynamic memory function that the previous moment output of hidden layer is recorded in structure layer, with the purpose of improving the prediction accuracy and strengthening the dynamic memory function. Furthermore, the prediction neural network controller is established. And the control part of the air conditioning system is mainly composed of two parts:a controller and a predictor. Simulation results show that the fuzzy neural network predictive control method improves the overshoot, settling time and steady precision, raises the adaptability and robustness of the system, and eliminates the static error, which makes the system have a good adaptive ability and strong learning ability. In addition, in view of the serious coupled phenomenon in the nonlinear and temperature-humidity control of variable air volume (VAV) air conditioning system, the controller model is established based on the structure of neural network decoupling controller, with the theory of combining neural networks and fuzzy control. The model has the feature of global approximation ability, compact topology structure, separate structure parameters and fast convergence rate. The predictor model is established based on the Elman neural networks, which is able to reflect the dynamic characteristics. A new multivariable neural network decoupling control method is proposed. And the decoupling controller is performed by neural network, which takes the temperature control value, humidity control value and the input of coupling channel as inputs of decoupling controller. Then the inputs will be translated to two control values of single-input single-output to reduce the impact of temperature and humidity coupling. The experimental results show that the method can improve the control effect of the air conditioning system, enhance the stability and dynamic of the VAV control system, and improve the energy efficient effect of air conditioning system.For evaluating the proposed method of harmonic suppression and central air conditioning energy saving control, and solving the accuracy problem of the traditional energy saving evaluation method, an energy saving evaluation index system of building electrical is established, which is based on the establishment principles of architectural electrical energy saving indicator system. The indicator system is described from three aspects:technical index, economic index and function index. In the meantime, using the analytic hierarchy process and the established decision matrix, a complete evaluation index system is constructed, according to the relative weight of each index to superior index and total objective. Based on BP neural network, the building electrical comprehensive evaluation model is established, and the quantitative comprehensive evaluation results are obtained. Moreover, a building electrical energy saving evaluation model based on chaotic neural network is established aiming at solving the following problem, that is the energy of BP algorithm may get into local minimum in solving process, and the optimal solution can't be obtained. The model evaluates sample building with artificial neural networks. The network layer includes input layer, output layer and hidden layer. Input layer node numbers of network correspond to the number of evaluation indexes, which includes three big indexes and twenty sub-indexes. And the node numbers of hidden layer are eight, while the output layer has one neuron. The results show that the model is maneuverable and has steady performance, which simplifies the evaluation process and reduces the error rate.