Hydrogen-containing energy storage micro-grid dispatching optimization method combining ARIMA-LSTM and DDPG

Abstract

The invention discloses a hydrogen-containing energy storage micro-grid dispatching optimization method in combination with ARIMA-LSTM and DDPG. Comprising the following steps: firstly, acquiring historical renewable energy productivity data, historical power load data, historical electricity price and historical energy storage equipment state of the hydrogen-containing energy storage microgrid; secondly, historical renewable energy productivity and power load data are predicted through an ARIMA-LSTM model, a complete state space is constructed in combination with the renewable energy productivity, the electricity price, the energy storage state and the like of the current time slot and input into an Actor and Critic network of the DDPG, a DDPG network model is trained, and a reward function model in the DDPG network model aims at optimizing the annual operation cost; the constraints comprise equipment operation constraints, renewable energy source output constraints, power imbalance and the like; and finally, a micro-grid scheduling strategy responding to external environment change is obtained, dynamic scheduling optimization of each device is realized, and adaptability and robustness of the system are improved.

Description

Technical Field

[0001] The present invention relates to a microgrid optimization scheduling method, and in particular to a hydrogen energy storage microgrid scheduling optimization method combining ARIMA-LSTM and DDPG. Background Art

[0002] Microgrids can not only realize autonomous power supply in local areas, but also effectively interact with large power grids to further improve energy efficiency. However, due to the large number of unstable renewable energy sources (such as photovoltaic and wind power) connected to microgrids, their power output has strong volatility and uncertainty, which brings great challenges to the scheduling optimization of microgrids.

[0003] To meet these challenges, traditional dispatch optimization methods usually rely on deterministic models and simple rule-based control methods, but these methods have difficulty in dealing with complex randomness and dynamically changing environments. With the development of artificial intelligence and big data technologies, data-driven prediction and optimization methods have gradually become an effective way to solve microgrid dispatch problems. In recent years, time series prediction models (such as ARIMA, LSTM, etc.) have shown good results in forecasting renewable energy generation and power load, while deep reinforcement learning algorithms (such as DDPG) can achieve optimal control in complex continuous states and action spaces.

[0004] In addition, hydrogen energy storage technology, as an efficient energy storage method, has gradually gained attention in microgrids. By converting electrical energy into hydrogen for storage, when there is excess electrical energy, the excess electrical energy can be converted into hydrogen for storage using water electrolysis hydrogen production technology, and when there is insufficient electrical energy, the hydrogen can be converted back into electrical energy through hydrogen fuel cells. Microgrid systems combined with hydrogen energy storage technology have high flexibility and dispatchability, which helps to further smooth the volatility of renewable energy.

[0005] Although current research has achieved initial results in microgrid scheduling and hydrogen energy storage technology, how to combine efficient time series prediction models with optimized scheduling algorithms to further improve the stability and economy of microgrid systems remains an urgent problem to be solved. Summary of the invention

[0006] In order to solve the problems in the background technology, the present invention provides a method for optimizing the scheduling of hydrogen energy storage microgrids combined with ARIMA-LSTM and DDPG. The method first models the microgrid scheduling problem as a Markov decision process, and establishes a mathematical model of each subsystem device of the microgrid; secondly, the historical renewable energy capacity and power load data are predicted by the ARIMA-LSTM model, and the renewable energy capacity, electricity price, energy storage status, etc. of the current time slot are combined to construct a complete state space, which is input into the Actor (evaluation) and Critic (execution) networks of DDPG to generate a scheduling strategy and a value assessment model; then, with the goal of optimizing the annual operating cost, a reward function model including equipment operation constraints, renewable energy output constraints, power imbalance, etc. is established; finally, the optimized model is trained by real data to obtain a microgrid scheduling strategy that responds to changes in the external environment, realize dynamic scheduling optimization of each device, and improve the adaptability and robustness of the system. The present invention can not only more accurately predict the renewable energy generation and power load, but also realize a dynamic and optimized microgrid scheduling strategy through deep reinforcement learning, effectively improving the operating efficiency and robustness of the system.

[0007] The technical solution adopted by the present invention is as follows:

[0008] 1. A method for optimizing the dispatch of hydrogen energy storage microgrids by combining ARIMA-LSTM and DDPG.

[0009] S1: Obtain historical renewable energy capacity data, historical power load data, historical electricity prices and historical energy storage equipment status of hydrogen energy storage microgrids;

[0010] S2: Use the time series forecasting model to generate the historical renewable energy capacity data, the predicted renewable energy capacity data and the predicted power load data corresponding to the historical power load data, and then generate a complete state space;

[0011] S3: After training the DDPG network model using the complete state space, a trained DDPG network model is obtained;

[0012] S4: Input the real-time collected renewable energy capacity data and power load data into the time series prediction model. The model outputs the predicted renewable energy capacity data and power load data and generates a real-time complete state space. The predicted complete state space is input into the trained DDPG network model. The model outputs the optimized scheduling data to complete the scheduling optimization of the hydrogen energy storage microgrid.

[0013] In S1, the historical renewable energy production capacity data is calculated by constructing a photovoltaic power generation model and a wind turbine power generation model of a hydrogen energy storage microgrid.

[0014] The complete state space includes historical renewable energy capacity data, historical power load data and corresponding predicted renewable energy capacity data, predicted power load data, historical electricity prices, and historical energy storage device status.

[0015] The time series prediction model includes an ARIMA-LSTM model.

[0016] In S3, the state space S of each agent in the DDPG network model t Satisfies the following formula:

[0017]

[0018] in, represents the photovoltaic power generation at time step t, represents the wind turbine power generation at time step t, represents the electrical load power at time step t, represents the photovoltaic power generation at time step t+1, represents the power generated by the wind turbine at time step t+1, represents the step power load power at time step t+1, represents the electricity price at time step t, represents the hydrogen storage tank pressure at time step t;

[0019] The action space A of each agent t Satisfies the following formula:

[0020]

[0021] in, represents the generator output power at time step t, represents the operating power of the electrolysis equipment at time step t, represents the fuel cell operating power at time step t;

[0022] The reward function R for each agent t Satisfies the following formula:

[0023]

[0024] in, represents the generator operating cost at time step t, is the cost of hydrogen production by electrolysis at time step t, represents the cost of hydrogen fuel cell power generation at time step t, represents the carbon pollution cost of all actions, Indicates the power imbalance penalty cost of this action.

[0025] 2. A Hydrogen Energy Storage Microgrid Dispatching Optimization System Combining ARIMA-LSTM and DDPG

[0026] An operation data acquisition unit of a hydrogen energy storage microgrid, used to acquire historical renewable energy production capacity data, historical power load data, historical electricity prices and historical energy storage equipment status of the hydrogen energy storage microgrid;

[0027] A data prediction unit, used to generate predicted renewable energy capacity data and predicted power load data corresponding to historical renewable energy capacity data and historical power load data using a time series prediction model, thereby generating a complete state space;

[0028] The optimization scheduling data generation unit is used to train the DDPG network model using the complete state space, obtain the trained DDPG network model and store it;

[0029] The real-time scheduling output unit is used to input the real-time collected renewable energy production capacity data and power load data into the time series prediction model in the data prediction unit. The model outputs the predicted renewable energy production capacity data and power load data and generates a real-time complete state space. The predicted complete state space is then input into the trained DDPG network model stored in the optimized scheduling data generation unit and outputs the optimized scheduling data.

[0030] 3. A computer device

[0031] The device includes a memory and a processor, the memory stores a computer program, and the processor implements the steps of the hydrogen energy storage microgrid scheduling optimization method combining ARIMA-LSTM and DDPG when executing the computer program.

[0032] 4. A computer-readable storage medium

[0033] The medium stores a computer program, which, when executed by a processor, implements the steps of the method for optimizing the scheduling of a hydrogen energy storage microgrid combining ARIMA-LSTM and DDPG.

[0034] V. A computer program product

[0035] The product includes a computer program / instruction, which, when executed by a processor, implements the steps of the hydrogen energy storage microgrid scheduling optimization method combining ARIMA-LSTM and DDPG.

[0036] The beneficial effects of the present invention are:

[0037] The present invention combines ARIMA-LSTM with DDPG, which can achieve high-precision prediction of renewable energy generation and power load, and improve the accuracy and real-time performance of microgrid dispatching decisions. Compared with traditional methods, the combined model of ARIMA-LSTM better captures the short-term fluctuations and long-term trends of time series data, reduces prediction errors, and provides reliable input data for subsequent dispatching.

[0038] By introducing DDPG, the present invention can realize the optimal scheduling strategy in complex continuous state and action space. DDPG dynamically adjusts the scheduling strategy of the microgrid through the Actor-Critic framework, optimizes the overall operation efficiency of the microgrid, reduces the operation cost, and improves the adaptability and robustness of the system to external environmental changes.

[0039] In addition, the present invention combines hydrogen energy storage technology, allowing the microgrid system to convert electricity into hydrogen for storage when there is excess electricity, and convert hydrogen into electricity through fuel cells when there is insufficient electricity, thereby improving energy utilization efficiency and the energy scheduling flexibility of the microgrid, and effectively solving the challenges brought by the volatility of renewable energy. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 It is an overall block diagram of the present invention;

[0041] Figure 2 is a microgrid structure diagram in one embodiment of the present invention;

[0042] Figure 3 It is a comparison chart of the reward values ​​of the method proposed in the present invention and other methods during the training process. DETAILED DESCRIPTION

[0043] Further description is given below in conjunction with the accompanying drawings.

[0044] The present invention describes the microgrid scheduling problem as a Markov decision process, such as Figure 1 As shown, the specific process of the present invention includes:

[0045] S1: Obtain historical renewable energy capacity data, historical power load data, historical electricity prices and historical energy storage equipment status of hydrogen energy storage microgrids;

[0046] S1 is specifically:

[0047] First, mathematical models of the subsystems of the hydrogen energy storage microgrid are constructed, including photovoltaic power generation model, wind turbine power generation model, hydrogen electrolysis model and hydrogen fuel cell discharge model, as follows:

[0048] Photovoltaic power generation model: Design the output power of the photovoltaic device at time t (i.e. time step t) It is related to the solar radiation intensity G, the photovoltaic panel area A and the conversion efficiency η of the equipment, and the formula is as follows:

[0049]

[0050] Among them, η PV Indicates the conversion efficiency of PV equipment.

[0051] Wind turbine power generation model: The designed wind turbine power generation is related to the wind speed and is divided into multiple working power intervals according to the wind speed. The formula is as follows:

[0052]

[0053] in, Indicates the rated output power of the fan, v ci 、v co 、v r 、v in They represent cut-in wind speed, cut-out wind speed, rated wind speed and wind speed at time t respectively.

[0054] Electrolysis hydrogen model: When there is excess electricity, the microgrid is designed to produce hydrogen by electrolyzing water, thereby converting electrical energy into chemical energy storage. The formula is as follows:

[0055]

[0056] in, represents the production of hydrogen, η elec Indicates the working efficiency of the electrolyzer, Indicates input power, F stands for Faraday constant, V cell Indicates the voltage of the electrolytic cell.

[0057] Hydrogen fuel cell discharge model: Design fuel cells, which convert the chemical energy of hydrogen into electrical energy. Its mathematical model is based on the expression between the input amount of hydrogen and the output electrical energy:

[0058]

[0059] in, Represents the output power, η FC represents the working efficiency of the fuel cell, represents the amount of hydrogen input to the fuel cell, n represents the number of stacks, V cell Indicates the Nernst voltage of the battery stack.

[0060] The hydrogen electrolysis model and hydrogen fuel cell discharge model are used to calculate the cost in the reward function, so that the model can make corresponding rewards and punishments for the decision of the intelligent agent. The power load data, electricity price, and energy storage device status data are directly obtained. The historical renewable energy capacity data is calculated by the photovoltaic power generation model and wind turbine power generation model of the constructed hydrogen energy storage microgrid.

[0061] S2: Use the time series forecasting model to generate the historical renewable energy capacity data, the predicted renewable energy capacity data and the predicted power load data corresponding to the historical power load data, and then generate a complete state space;

[0062] The complete state space includes historical renewable energy capacity data, historical power load data and the corresponding predicted renewable energy capacity data, predicted power load data, historical electricity prices, and historical energy storage equipment status.

[0063] The time series prediction model in this embodiment adopts the ARIMA-LSTM model, which specifically includes:

[0064] Data collection and preprocessing: Collect historical renewable energy capacity data (such as photovoltaic power generation data, wind power generation data) and power load data, eliminate outliers and missing values, and standardize the data to ensure the consistency and accuracy of the data input into the model.

[0065] Assume that the historical data set D is D = {(y1, y2...y t )}, where y t is the observed value at time t. After normalization, the normalized observed value y′ at time t is obtained t .

[0066] ARIMA model predicts short-term trends: The ARIMA model can model the autoregression and moving average processes in the time series, extract the long-term trend and seasonal characteristics of the data, predict the short-term trend of renewable energy capacity data and power load data, and provide a basis for subsequent LSTM predictions.

[0067] For the normalized predicted value y′ at time t t , the formula of the ARIMA model can be expressed as:

[0068]

[0069] Among them, φ1, φ2, φ p denote the first, second and pth sub-regression coefficients, θ1, θ2, θ q Represent the first, second and qth moving average coefficients respectively, ∈ t-1 ,∈ t-2 ,∈ t-qThey represent the white noise error terms at time t-1, time t-2 and time tq respectively.

[0070] LSTM model captures long-term and short-term dependencies: The LSTM model can capture long-term and short-term dependencies of data through its special memory unit and gating mechanism, further improving the prediction accuracy of renewable energy capacity and power load. Its state update formula is as follows:

[0071] Forget gate: determines the memory unit C of the previous moment t-1 Whether to retain the information in:

[0072] f t =σ(W f [h t-1 , x t ]+b f ),

[0073] Input gate: decides whether to store new information into the memory cell:

[0074] i t =σ(W i [h t-1 , x t ]+b i )

[0075]

[0076] h t =o t *tanh(C t )

[0077] Output gate: determines the hidden state h of the final output t :

[0078] Among them, W f , W i , W C , W0 represents the weight matrix, b f , b i , b C , b o represents the bias term, J t represents the output of the forget gate, σ() represents the activation function, x t Represents the input at the current moment, i t represents the output of the input gate, represents the candidate cell state, o t represents the output of the output gate, C t and C t-1 Respectively represent the cell state at the current moment and the previous moment, h t Indicates the hidden state at the current moment.

[0079] Based on the ARIMA-LSTM model and historical data, the forecast values ​​of renewable energy capacity and power load for the next time step are generated and combined into the state space of the DDPG model.

[0080] S3: After training the DDPG network model using the complete state space, a trained DDPG network model is obtained;

[0081] The state space S of each agent in the DDPG network model t Satisfies the following formula:

[0082]

[0083] in, represents the photovoltaic power generation at time step t, represents the wind turbine power generation at time step t, represents the electrical load power at time step t, represents the photovoltaic power generation at time step t+1, represents the power generated by the wind turbine at time step t+1, represents the step power load power at time step t+1, represents the electricity price at time step t, represents the hydrogen storage tank pressure at time step t;

[0084] The action space A of each agent t Contains all the ways in which the agent interacts with the environment and satisfies the following formula:

[0085]

[0086] in, represents the generator output power at time step t, represents the operating power of the electrolysis equipment at time step t, represents the fuel cell operating power at time step t;

[0087] The design goal of the reward function is to optimize the annual operating cost of the microgrid system, that is, the instant reward obtained by the agent based on the interaction between the action and the environment, and to maximize the cumulative reward of the entire learning process as the learning goal of the agent. The reward function in this invention comprehensively considers factors such as equipment operation constraints, renewable energy output constraints, power imbalance constraints, power generation costs and pollution costs. The reward function R of each agent t Satisfies the following formula:

[0088]

[0089] in, represents the generator operating cost at time step t, is the cost of hydrogen production by electrolysis at time step t, represents the cost of hydrogen fuel cell power generation at time step t, represents the carbon pollution cost of all actions, Indicates the power imbalance penalty cost of this action.

[0090] The DDPG (Deep Deterministic Policy Gradient) algorithm is a deep reinforcement learning algorithm based on policy gradient. It combines the Actor-Critic structure, generates policies through the Actor network, evaluates the policies through the Critic network, and guides the update of policies through the value function. It can effectively solve control problems in continuous action space.

[0091] The complete state space is input into the Actor network and Critic network in the DDPG network model to train the DDPG model. The Actor network generates the scheduling strategy, and the Critic network evaluates the value of the strategy. The DDPG network outputs actions according to the state of the agent, and the agent is rewarded after executing the action. The network then learns according to the state and reward. Through multiple rounds of iterations, the network parameters are adjusted until it converges to the optimal strategy, and finally a trained DDPG network model is obtained. The optimal microgrid scheduling strategy that responds to changes in the external environment is generated through the optimized model, and the strategy is used to dynamically schedule each device in the microgrid to improve the adaptability and robustness of the microgrid system. The DDPG network model and the ARIMA-LSTM model together constitute the microgrid scheduling optimization model. The former is responsible for optimizing the scheduling strategy, and the latter predicts renewable energy production capacity and power load to provide input for scheduling optimization.

[0092] Specifically:

[0093] The Actor network is used to t Generate a specific optimal scheduling action a t :

[0094] a t =μ(s t |θ μ ),

[0095] Among them, θ μ represents the network parameters of the Actor, and μ() represents the policy function. At each step of the policy update, the Actor updates the policy based on the policy gradient formula according to the gradient information provided by the Critic network:

[0096]

[0097] in, represents the gradient, represents the expectation, Q(s, a|θ Q )| a＝μ(s) represents the action value function, θ Q Represents the network parameters of Critic, r t represents the immediate reward at time t, and γ represents the discount factor.

[0098] The Critic network estimates the state-action value function Q(s) based on the Bellman formula t , a t ) to evaluate the performance of the current strategy and minimize the loss function L(Q θ ) to update the network parameters:

[0099] Q(s t , a t |θ Q )=r t +γQ(s t+1 ,μ(s t+1 |θ μ )|θ Q )

[0100]

[0101] Through continuous iteration, the Actor network of DDPG gradually generates better scheduling strategies, and the Critic network continuously evaluates and optimizes the strategies, so that the microgrid can perform efficient hydrogen energy storage scheduling optimization in a dynamic environment.

[0102] S4: Input the real-time collected renewable energy capacity data and power load data into the time series prediction model. The model outputs the predicted renewable energy capacity data and power load data and generates a real-time complete state space. The predicted complete state space is input into the trained DDPG network model. The model outputs the optimized scheduling data to complete the scheduling optimization of the hydrogen energy storage microgrid.

[0103] In order to further verify the effectiveness and excellence of the method proposed in the present invention, the following examples are provided for illustration.

[0104] The microgrid dispatch optimization model is trained in the constructed simulation experiment environment. In this example, a microgrid consisting of 4 photovoltaic power stations, 1 wind power station, 1 diesel generator, 1 electrolysis equipment, and 1 fuel cell is used as the simulation experiment environment. The maximum operating power of the diesel generator is set to 100KW. The maximum power of the electrolysis equipment is set to 40KW, and the electrolysis efficiency is set to 0.5. The maximum power of the fuel cell is set to 50KW, and the working efficiency is set to 0.5. In order to encourage the microgrid system to actively dispatch renewable energy, the diesel price is set to 0.4 times the original price. The training set uses data including 8640 hours of radiation intensity, wind speed, power load data, and electricity prices per year.

[0105] In the simulation stage, the parameters of the model are listed in Table 1. The design model starts from t = 1 and ends at t = 8640. In each time step, the ARIMA-LSTM model is first applied to predict the next step based on the historical renewable energy data and power load data, and the obtained results are combined with the state space of the current time step and fed into DDPG as a complete state space for training. In order to evaluate the performance of the algorithm, the reward function is used as a metric.

[0106] Table 1 shows the parameter settings of ARIMA-LSTM and DDPG

[0107]

[0108]

[0109] In order to verify the superiority of the DDPG method combined with ARIMA-LSTM proposed in this paper, it is compared with the standard DDPG. Figure 3 A. Figure 3 B and Figure 3 As shown in C, the DDPG combined with ARIMA-LSTM proposed in the present invention is significantly better than the traditional DDPG in performance. Specifically, after 200 rounds of random action to expand the experience pool, the DDPG combined with ARIMA-LSTM has a faster rising speed, and it is close to convergence in less than 20 rounds of formal training. The reward value is also increasing in subsequent rounds, but the rising trend is slowing down. The final reward value is also significantly greater than the traditional DDPG, and the variance is also smaller. After the formal training starts, the traditional DDPG needs nearly 100 rounds to converge approximately, and the final value is always small.

[0110] The present invention also proposes a hydrogen energy storage microgrid dispatch optimization system combining ARIMA-LSTM and DDPG, including:

[0111] An operation data acquisition unit of a hydrogen energy storage microgrid, used to acquire historical renewable energy production capacity data, historical power load data, historical electricity prices and historical energy storage equipment status of the hydrogen energy storage microgrid;

[0112] A data prediction unit, used to generate predicted renewable energy capacity data and predicted power load data corresponding to historical renewable energy capacity data and historical power load data using a time series prediction model, thereby generating a complete state space;

[0113] The optimization scheduling data generation unit is used to train the DDPG network model using the complete state space, obtain the trained DDPG network model and store it;

[0114] The real-time scheduling output unit is used to input the real-time collected renewable energy production capacity data and power load data into the time series prediction model in the data prediction unit. The model outputs the predicted renewable energy production capacity data and power load data and generates a real-time complete state space. The predicted complete state space is then input into the trained DDPG network model stored in the optimized scheduling data generation unit and outputs the optimized scheduling data.

[0115] Finally, it should be noted that the above embodiments and explanations are only used to illustrate the technical solution of the present invention rather than to limit it. Those skilled in the art should understand that the technical solution of the present invention can be modified or replaced by equivalents without departing from the spirit and scope disclosed in the technical solution of the present invention, which should be included in the scope of protection of the claims of the present invention.

Claims

1. A method for optimizing the scheduling of hydrogen energy storage microgrids combining ARIMA-LSTM and DDPG, characterized in that: The following steps are involved: S1: Obtain historical renewable energy capacity data, historical power load data, historical electricity prices and historical energy storage equipment status of hydrogen energy storage microgrids; S2: Use the time series forecasting model to generate the historical renewable energy capacity data, the predicted renewable energy capacity data and the predicted power load data corresponding to the historical power load data, and then generate a complete state space; S3: After training the DDPG network model using the complete state space, a trained DDPG network model is obtained; S4: Input the real-time collected renewable energy capacity data and power load data into the time series prediction model. The model outputs the predicted renewable energy capacity data and power load data and generates a real-time complete state space. The predicted complete state space is input into the trained DDPG network model. The model outputs the optimized scheduling data to complete the scheduling optimization of the hydrogen energy storage microgrid.

2. According to claim 1, a method for optimizing the scheduling of a hydrogen-containing energy storage microgrid combining ARIMA-LSTM and DDPG, characterized in that: In S1, the historical renewable energy production capacity data is calculated by constructing a photovoltaic power generation model and a wind turbine power generation model of a hydrogen energy storage microgrid.

3. According to claim 1, a method for optimizing the scheduling of a hydrogen-containing energy storage microgrid combining ARIMA-LSTM and DDPG, characterized in that: The complete state space includes historical renewable energy capacity data, historical power load data and corresponding predicted renewable energy capacity data, predicted power load data, historical electricity prices, and historical energy storage device status.

4. According to claim 1, a method for optimizing the scheduling of a hydrogen-containing energy storage microgrid combining ARIMA-LSTM and DDPG, characterized in that: The time series prediction model includes an ARIMA-LSTM model.

5. According to a method for optimizing the scheduling of hydrogen energy storage microgrids combining ARIMA-LSTM and DDPG according to claim 1, it is characterized in that: In S3, the state space S of each agent in the DDPG network model t Satisfies the following formula: in, represents the photovoltaic power generation at time step t, represents the wind turbine power generation at time step t, represents the electrical load power at time step t, represents the photovoltaic power generation at time step t+1, represents the power generated by the wind turbine at time step t+1, represents the step power load power at time step t+1, represents the electricity price at time step t, represents the hydrogen storage tank pressure at time step t; The action space A of each agent t Satisfies the following formula: in, represents the generator output power at time step t, represents the operating power of the electrolysis equipment at time step t, represents the fuel cell operating power at time step t; The reward function R for each agent t Satisfies the following formula: in, represents the generator operating cost at time step t, is the cost of hydrogen production by electrolysis at time step t, represents the cost of hydrogen fuel cell power generation at time step t, represents the carbon pollution cost of all actions, Indicates the power imbalance penalty cost of this action.

6. A hydrogen energy storage microgrid dispatch optimization system combining ARIMA-LSTM and DDPG, characterized in that: include: An operation data acquisition unit of a hydrogen energy storage microgrid, used to acquire historical renewable energy production capacity data, historical power load data, historical electricity prices and historical energy storage equipment status of the hydrogen energy storage microgrid; A data prediction unit, used to generate predicted renewable energy capacity data and predicted power load data corresponding to historical renewable energy capacity data and historical power load data using a time series prediction model, thereby generating a complete state space; The optimization scheduling data generation unit is used to train the DDPG network model using the complete state space, obtain the trained DDPG network model and store it; The real-time scheduling output unit is used to input the real-time collected renewable energy production capacity data and power load data into the time series prediction model in the data prediction unit. The model outputs the predicted renewable energy production capacity data and power load data and generates a real-time complete state space. The predicted complete state space is then input into the trained DDPG network model stored in the optimized scheduling data generation unit and outputs the optimized scheduling data.

7. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, the steps of the hydrogen energy storage microgrid scheduling optimization method combining ARIMA-LSTM and DDPG as described in any one of claims 1 to 5 are implemented.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method for optimizing the scheduling of a hydrogen energy storage microgrid combining ARIMA-LSTM and DDPG as described in any one of claims 1 to 5 are implemented.

9. A computer program product comprising a computer program / instructions, characterized in that When the computer program / instruction is executed by a processor, the steps of the method for optimizing the scheduling of a hydrogen energy storage microgrid combining ARIMA-LSTM and DDPG as described in any one of claims 1 to 5 are implemented.

Images