HVAC control system based on multi-step prediction deep reinforcement learning algorithm

Abstract

The invention relates to an intelligent control method of a control system of an HVAC (Heating, Ventilation, Air-conditioning and Cooling) control system, in particular to an HVAC control system based on an LSTM (Long Short Term Memory) neural network of a GC (generalized correntropy) loss function and a DRL (deep reinforcement learning) algorithm. The method comprises the following steps: acquiring outdoor environment temperature, indoor environment temperature and electricity price information of a power grid, preprocessing the acquired data, and predicting the outdoor environment temperature of multiple steps in the future by using historical data of the outdoor environment temperature; on the basis of the future outdoor temperature value, the indoor environment temperature and the power grid electricity price information, controlling the power output of the HVAC system by utilizing a DDPG (Deep Deterministic Policy Gradient) algorithm of the DRL. According to the invention, the HVAC system can be intelligently controlled in real time, so that the user cost is reduced, the satisfaction degree of the user is ensured, and the method has high practical engineering application value.

Description

technical field [0001] The invention relates to a method for intelligently optimizing and controlling an HVAC system, in particular to a research method for intelligently controlling an HVAC system based on a GC-LSTM neural network and a DRL algorithm. Background technique [0002] Household users are the end users of the power grid. The user's electricity consumption habits and the addition of distributed renewable energy will directly lead to the emergence of peaks and troughs in the power grid; it has brought serious impacts on the power grid and caused serious threats. With the development of the smart grid and the implementation of the "demand response" strategy in recent years, residential users have changed from being passive to actively joining the grid; Two-way exchange of information. Among household users, the power consumption of the air-conditioning system accounts for about 35% of the total power consumption of the entire user. Therefore, on the premise of sat...

AI Technical Summary

Problems solved by technology

[0003] At present, the HVAC system mainly adopts the traditional control method closed-loop control and model predictive control algorithm. There is a temperature sensor inside the closed-loop control system. When it detects that the indoor temperature reaches the set value, the HVAC system will stop working. Based on the closed-loop control method The HVAC system is simple to operate and easy to implement, but in the environment of smart grid and corresponding demand strategy, it is difficult to perform power conversion according to the dynamic electricity price to meet the standards of energy saving and emission reduction; the model predictive control algorithm is based on the establishment of indoor temperature changes Accurate models can then control the HVAC system, however, the complexity of indoor ambient temperature changes will affect the accuracy of the modeling

With the development of intelligent algorithms, researchers have also proposed to use particle swarm optimization algorithm and genetic algorithm optimization to optimize the control of HVAC system. This type of algorithm optimizes the power output of HVAC system under the mechanism of real-time electricity price to reduce the cost of users. , the algorithm has the characteristics of difficult parameter adjustment, and does not consider the delay of the power output of the HVAC system to the indoor temperature change, and does not really guarantee the user's comfort

Images