Multi energy complementary system is a new method of solving the problem of renewable energy consumption. This paper proposes a wind -pumped storage-hydrogen storage combined operation system based on deep learning and intelligent optimization, which introduces deep neural network to predict wind power generation. With the goal of minimizing power fluctuation and maximizing economic benefits, the system is optimized by multi-objective genetic. Multi energy complementary system is a new method of solving the problem of renewable energy consumption. This paper proposes a wind -pumped storage-hydrogen storage combined operation system based on deep learning and intelligent optimization, which introduces deep neural network to predict wind power generation. With the goal of minimizing power fluctuation and maximizing economic benefits, the system is optimized by multi-objective genetic algorithm for the basic parameters of wind turbine arrangement, electrolyzer and pumped storage power station. After getting the Pareto frontier solutions, we use Technique for Order Preference by Similarity to Ideal Solution(TOPSIS) to select the best scheme. Taking a specific case study for example, the system reduces the daily power fluctuation from 104.20 MW to 23.37 MW, a drop of 77.60%, and produces daily economic benefit of 139,638.5 yuan. Finally, by comparing the system with 3 and 9 wind turbines, we confirm the flexibility and universality of our system.••••Wind power generation model based on deep neural learning.••Generation optimization of combined operation of wind power-pumped storage-hydrogen energy storage.••The simultaneous optimization of control and design of the combined system.Renewable energy consumptionPumped storageDeep learningIntelligent optimizationRenewable energy power generation is an indispensable part of building a clean and low-carbon energy system. At present, the mature and widely used new energy is wind power, photovoltaic, etc. However, due to the inherent intermittent and uncontrollability of wind power, as well as other factors, the problem of renewable energy consumption has been very prominent. In particular, with the rapid expansion of the scale of grid connection, the contradiction between the risk of curtailment and the stable operation of the high proportion of clean energy systems has become more prominent, so we expect a power supply system with greater flexibility. Pumped storage is one of the feasible and effective ways to build a flexible power supply system. The core is to improve the quality and reliability of power grid operation by peak shaving, valley filling, standby, frequency modulation and phase modulation.The existing theoretical and methodological research on the complementary dispatching operation of pumped storage and wind power generation at home and abroad can be generally divided into two categories. The first type is to describe the wind power generation by using the uncertain description method, and then build a joint dispatching model with conventional power sources such as hydropower. The optimization criteria of the model usually includes three types: the first is the clean energy consumption criteria, such as minimizing a. DNN forecast wind power generation model driven by big data is based on Flow Redirection and Induction in Steady State (FLORIS) wind farm simulation platform jointly developed by National Renewable Energy Laboratory (NREL) and Delft University of Technology. FLORIS model is a parameterized model driven by calibration data, which is applied to real-time optimization to improve the performance of wind farms. The power of wind generation is related to the pitch angle and yaw angle of turbine blades, the arrangement position of wind turbines, wind speed, wind direction and other factors. This paper mainly studies the influence of these factors, and establishes the following functional relationship:(1)PWF=FWF(X),X=[x,y,v,d]where, PWF is the total power generated by the wind farm, MW. FWF(X) is the functional relation of X=[x,y,v,d]. x=[x1,x2,x3,x4. xn] is the x-axis coordinate of the wind turbine in the selected coordinate system. y=[y1,y2,y3,y4. yn] is the y-axis coordinate of the wind turbine in the selected coordinate system. v is real-time wind speed, m/s. d is the real-time wind direction.The FLORIS wind farm simulation platform is used to simulate the wind farms with different unit arrangements and different wind speeds and directions (Fig. 1). As is shown in Fig. 1, there are three wind turbines arranged on the given x, y coordinates, which operate under the set wind speed and wind direction. Th.
