Double-layer multi-time-scale scheduling method for off-grid wind-solar complementary hydrogen production system

Abstract

The invention relates to a double-layer multi-time-scale scheduling method for an off-grid type wind-solar complementary hydrogen production system, which is a double-layer scheduling mechanism based on double targets of improving system economy and ensuring equipment safety, an upper layer takes an intelligent agent as a core, the service life of equipment is prolonged by adopting comprehensive consideration, overcharge and overdischarge of an energy storage element are reduced, and the energy storage efficiency is improved. First-stage power distribution of an excitation function for stabilizing the fluctuation of the PEM electrolytic cell array: taking wind and light as the total power of an input system, and distributing the total power to a storage battery, a super capacitor and the PEM electrolytic cell array; the lower-layer second-level PEM electrolytic cell array control depends on the result of the first-level power distribution, the targets of maximizing the hydrogen yield and prolonging the service life of the PEM electrolytic cells are considered at the same time, the PEM electrolytic cells are controlled by adopting a PEM electrolytic cell array optimization scheduling strategy of a double-wheel-value lock multi-time scale, and dynamic distribution of power is completed. Therefore, the system scheduling is higher in adaptability to renewable energy fluctuation on the whole, good in economical efficiency and lower in cost.

Description

Technical Field

[0001] The present invention belongs to the field of flexible application of power system energy storage, and in particular relates to a double-layer multi-time scale scheduling method for an off-grid wind-solar complementary hydrogen production system using renewable energy. Background Art

[0002] In recent years, with the continuous changes in the international environment and energy forms, actively exploring and developing clean, efficient and sustainable energy has become the common goal of most countries in the world. However, the inherent randomness and volatility of renewable energy will lead to low energy utilization, affecting the stability of the hydrogen production system and the hydrogen production. Hybrid energy storage technology includes power-type energy storage with high power demand, transient and rapid load fluctuations, and high energy storage, energy-type energy storage with low self-discharge rate and low energy ratio. The combination of the two can effectively smooth out wind power fluctuations on different time scales. Unreasonable power allocation of the hydrogen production system may lead to increased system costs and frequent switching of charging and discharging of each module, shortening the service life of energy storage equipment, and reducing hydrogen production and purity.

[0003] The existing off-grid wind-solar hybrid hydrogen production system scheduling method is based on model predictive control (MPC). The power allocation adopts variational mode decomposition (VMD) to directly decompose the renewable energy reference power into high-frequency fluctuation component, medium-frequency fluctuation component and low-frequency fluctuation component. The high-frequency fluctuation component and medium-frequency fluctuation component are used as the energy storage reference power, corresponding to the supercapacitor module and the battery module respectively. The low-frequency fluctuation component is used as the reference power of the proton exchange membrane (PEM) electrolyzer array; the PEM electrolyzer array uses a polling mechanism to control and absorb low-frequency variables. The power scheduling based on power decomposition in this scheduling method cannot make real-time adjustments to the PEM electrolyzer array, lithium battery, and supercapacitor power allocation according to the working requirements of the PEM electrolyzer. There are problems such as overcharging and over-discharging of lithium batteries and supercapacitors and the power of the electrolyzer exceeding the limit. The PEM electrolyzer control uses a simple polling mechanism to adjust the PEM electrolyzer in a fixed time period based on a single strategy such as the operating time of the PEM electrolyzer. It lacks the ability to adapt to the volatility of renewable energy and affects the service life of the PEM electrolyzer. Summary of the invention

[0004] In view of the shortcomings of the prior art, the purpose of the present invention is to provide a double-layer multi-time-scale scheduling method for an off-grid wind-solar complementary hydrogen production system. The method is used in an off-grid renewable energy hydrogen production system. On the one hand, an upper-layer scheduling is performed based on an offline actor-critic algorithm. On the other hand, a double-rotation lock is used to control the PEM electrolyzer to achieve economical, safe and stable scheduling of an off-grid renewable energy hydrogen production system.

[0005] In order to achieve the above object, the present invention provides the following scheme: A two-layer multi-time scale scheduling method for an off-grid wind-solar hybrid hydrogen production system, the method is a two-layer scheduling mechanism based on the dual goals of improving system economy and ensuring equipment safety. The upper layer is centered on an intelligent agent, and adopts a primary power allocation of an incentive function that comprehensively considers extending the service life of the equipment, reducing overcharge and overdischarge of energy storage components, and stabilizing the fluctuation of the PEM electrolyzer array. The wind and solar are used as the total power of the input system and are allocated to the battery, supercapacitor, and PEM electrolyzer array. The control of the lower-level secondary PEM electrolyzer array relies on the results of the primary power allocation. Taking into account the goals of maximizing hydrogen production and extending the service life of the PEM electrolyzer, a dual-rotation lock multi-time scale PEM electrolyzer array optimization scheduling strategy is used to control the PEM electrolyzer and complete the dynamic allocation of power. The dual-rotation lock multi-time scale PEM electrolyzer array optimization scheduling strategy is: each electrolyzer has a dual-rotation lock consisting of a fluctuation rotation lock and an operation rotation lock, and the fluctuation rotation lock Used to count the accumulated fluctuating state operation time of the electrolytic cell and run the rotation lock Used to count the cumulative running time of the electrolyzer; in the current time period, the working state of each PEM electrolyzer in the electrolyzer array and the selection of the PEM electrolyzer in the fluctuating power state are determined by the rotation lock of each PEM electrolyzer; When the system starts running, set the initial rotation lock value of each device to 0; enter the long time scale Electrolyzer operation - stop the main scheduling loop, and predict the current time based on the historical data allocated by the upper model The total average power of the electrolyzer array during the time period , using intelligent algorithms to determine the number of electrolytic cells that need to be opened ; Determine whether the operation status needs to be updated. If necessary, sort and select the equipment, and sort and select from high to low priority according to the electrolytic cell with the shortest operation time and the shortest operation time in the fluctuating state, the electrolytic cell with a relatively short operation time and the shortest operation time in the fluctuating state, the electrolytic cell with a relatively short operation time and a relatively short operation time in the fluctuating state, and the electrolytic cell with a relatively long operation time and a relatively long operation time in the fluctuating state. The specific selection process is: a. First press Run Rotation Lock Sort the devices by bucket from smallest to largest; b. On this basis, according to the fluctuation rotation lock Sort the buckets twice from small to large; c. Select the first after sorting A device is used as an operating device. When multiple electrolytic cells have the same ranking, the roulette algorithm is used to select whether to start the electrolytic cells with the same ranking. In determining After the electrolyzer is running for a period of time, it enters the short-term time scale The electrolytic cell fluctuation-steady-state scheduling sub-cycle is used to control the electrolytic cell fluctuation-steady-state operation state. The electrolytic cell fluctuation state operation time is the shortest and the operation time is the shortest, the electrolytic cell fluctuation state operation time is relatively short and the operation time is the shortest, the electrolytic cell fluctuation state operation time is relatively short and the operation time is short, and the electrolytic cell fluctuation state operation time is relatively short and the operation time is short, and the electrolytic cell fluctuation state operation time is relatively long and the operation time is relatively long. The specific selection process is: a. Exclude the last fluctuation device from the running device; b. For the remaining optional devices, press Sort the buckets from small to large, and then sort them according to Sort the buckets twice from small to large; c. Select the first device after sorting as the current fluctuation device; d. Update the wave device , add 1; Repeating short time scales Sub-cycle period z times, where , forming a complete long-time main cycle; at the end of each cycle, the rotation locks of all running devices are updated, and at the same time, it is checked whether any rotation lock exceeds the maximum value If there is a and If the maximum value is exceeded, a scaling factor is used Reduce all rotating locks by equal proportion; Then calculate all and The average value of the rotating lock is obtained by dividing the current value of the rotating lock by its average value to obtain the relative value of the rotating lock of all devices, and the relative value of the rotating lock is used as the initial rotating lock of the next cycle.

[0006] z is an integer greater than 2. Greater than or equal to 5 minutes, .

[0007] Furthermore, taking wind power and solar power as the input power of hybrid energy storage, the upper layer takes the ACE model as the core, comprehensively considering the goals of suppressing overcharge and over-discharge of equipment, maximizing hydrogen production, and smoothing the input power of the electrolyzer array; The upper-level ACE model makes a first-level power allocation to the supercapacitor, battery, and PEM electrolyzer array based on the incentive function. When the supercapacitor and battery are about to be overcharged, the intelligent agent is encouraged to regulate the discharge of the supercapacitor and battery; when the supercapacitor and battery are about to be over-discharged, the intelligent agent is encouraged to regulate the charging of the supercapacitor and battery. At the same time, the intelligent agent is encouraged to stabilize the power of the PEM electrolyzer array on a long time scale. In the lower-level second-level PEM electrolyzer array control, according to the results of the upper-level first-level power allocation, a double-rotation lock multi-time-scale PEM electrolyzer array optimization scheduling strategy is used to control the run-stop state of each electrolyzer in the electrolyzer array on a long time scale, and to control the steady-state-fluctuating operation state on a short time scale.

[0008] Furthermore, the upper-level power allocation includes a comprehensive definition of the state space, action space and incentive function of the hydrogen production system that can be produced from renewable energy, which transforms the optimal scheduling problem of the off-grid wind-solar complementary hydrogen production system into a Markov decision process (MDP).

[0009] Furthermore, the state space is defined as:

[0010] in , for The input power of wind turbine and photovoltaic at all times; , , They are the capacity of the supercapacitor, the capacity of the battery, and the capacity of the hydrogen storage tank; Action Space is defined as:

[0011] in , , They are The power of the battery, supercapacitor, and electrolyzer array controlled by the agent at all times; when the agent controls the power, the following constraints are met:

[0012]

[0013]

[0014]

[0015] in and Indicates the minimum operating power and maximum operating power of the electrolyzer; and Respectively represent the maximum charging and discharging power of the supercapacitor, and and Indicates the maximum charging and discharging power of the battery; Activation Function Satisfy the leveling criterion, expressed as:

[0016] in The corresponding reward is the hydrogen production amount. For electrolyzer array The hydrogen production at the moment is calculated based on the scheduling of the lower electrolyzer array. It is the incentive factor for hydrogen production; , They are the incentives obtained when the agent takes measures to alleviate battery overcharge and overdischarge, and when the battery Status of the moment When the battery is too high, that is, when overcharging is likely to occur, discharging the battery will be encouraged. When the over-discharge phenomenon is likely to occur, charging the battery will be motivated. , It is the excitation coefficient for the battery to take action when it is overcharged or over-discharged. is the upper limit of the battery's safe SOC state, It is the lower limit of the battery safety SOC state; , They are the incentives obtained when the agent takes measures to alleviate the overcharge and over-discharge of the supercapacitor. Status of the moment When the overcharge phenomenon may occur, the supercapacitor will be discharged to obtain incentives. When the over-discharge phenomenon is likely to occur, charging the supercapacitor will be stimulated. , It is the excitation coefficient of the supercapacitor when it is overcharged or over-discharged. is the upper limit of the supercapacitor's safe SOC state. It is the lower limit of the supercapacitor's safe SOC state; It is the agent's incentive to maximize the utilization of the system input power. With load The closer you are, the higher the incentive you get. , , It is the incentive obtained by adjusting the system input power and load balance under different conditions. , It is the boundary value describing the difference between input power and load; When the agent chooses inappropriate behavior, the environment will impose penalties. The penalty functions include battery\supercapacitor SOC over-limit penalty, failure to meet electrical load balance penalty, battery SOC failure to return to the initial state after the scheduling period ends, long-term volatility penalty assigned to the electrolyzer array, and instantaneous power over-limit penalty.

[0017] The intelligent algorithm described in the present invention is a snake optimization algorithm, a genetic algorithm or a particle swarm algorithm.

[0018] Compared with the prior art, the double-layer multi-time scale scheduling method for the off-grid wind-solar hybrid hydrogen production system of the present invention has the following beneficial effects: The method of the present invention is based on the dual goals of improving system economy and ensuring equipment safety, and establishes a two-layer scheduling mechanism. The upper layer is centered on the intelligent body, and adopts a first-level power allocation that comprehensively considers the incentive functions of extending the service life of the equipment, reducing overcharge and overdischarge of energy storage components, and stabilizing the fluctuation of the PEM electrolyzer array. The wind and solar power are used as the total power of the input system and allocated to the battery, supercapacitor, and PEM electrolyzer array; the lower-level secondary PEM electrolyzer array control relies on the result of the first-level power allocation, while considering the goals of maximizing hydrogen production and extending the service life of the PEM electrolyzer, and adopts a double-rotation lock multi-time scale PEM electrolyzer array optimization scheduling strategy to control the PEM electrolyzer. This makes the system scheduling more adaptable to the fluctuation of renewable energy as a whole, with good economy and lower cost.

[0019] The present invention combines the two-layer multi-time scale scheduling based on the offline actor-critic algorithm (ACE) and the double-rotating lock to reasonably allocate wind and solar power, smooth out the fluctuation of renewable energy, and obtain the optimal real-time scheduling result that comprehensively considers the equipment life, hydrogen production, and balances the operating time of each electrolyzer and the fluctuation time.

[0020] The upper-level power allocation in the method of the present invention includes a comprehensive definition of the state space, action space and incentive function of the renewable energy hydrogen production system, which effectively transforms the optimal scheduling problem of the renewable energy hydrogen production system into a Markov decision process (MDP).

[0021] The method of the present invention controls the lower-level secondary electrolyzer array, aims at the problem of balancing the operating time and fluctuation time of each electrolyzer in the PEM electrolyzer array, and proposes a dual-rotation lock multi-time scale PEM electrolyzer array optimization scheduling strategy, which utilizes the dual priorities of comprehensive fluctuation state time and operating time to balance the operating time and fluctuation time of each electrolyzer in the electrolyzer array, maximizes the use of the fast start and stop characteristics of the electrolyzer, increases hydrogen production, and minimizes the damage to the life of the electrolyzer caused by power fluctuations, thereby extending the service life of the electrolyzer. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 This is a schematic diagram of the off-grid wind-solar complementary hydrogen production system of the present invention.

[0023] Figure 2 This is the power-efficiency curve of the electrolyzer.

[0024] Figure 3 This is a forecast diagram of the electrolyzer array for the next 15 minutes.

[0025] Figure 4 The figure is a flowchart of a scheduling method according to an embodiment of the present invention. DETAILED DESCRIPTION

[0026] In order to better understand the purpose, structure and function of the present invention, the real-time scheduling of off-grid renewable energy of the present invention is further described in detail below with reference to the accompanying drawings.

[0027] The present invention belongs to the field of flexible application of power system energy storage, and is a double-layer multi-time scale scheduling method based on reinforcement learning and rotation lock. Hydrogen energy storage has gradually demonstrated its advantages in energy density and environmental protection in recent years, and can be added to traditional hybrid energy storage systems to obtain a safer, more stable and economical energy storage solution. However, wind power and photovoltaic power have strong randomness and volatility, and electrolyzers are sensitive to power volatility. Excessive frequency may cause equipment damage, so research on scheduling methods for renewable energy production hydrogen production systems is essential. By using renewable energy hydrogen production scheduling based on offline actor-critic algorithm (ACE) and double rotation lock multi-time scale, the system scheduling can regulate the wind and photovoltaic power in supercapacitors, batteries, and PEM electrolyzer arrays in real time at the upper layer, and the lower layer uses double rotation locks in the PEM electrolyzer array to reasonably complete the dynamic distribution of power for each PEM electrolyzer at multiple time scales. The method of this embodiment can use wind power and light power as the input power of hybrid energy storage. The upper layer takes the ACE model as the core, comprehensively considering the goals of suppressing overcharging and over-discharging of equipment, maximizing hydrogen production, and smoothing the input power of the electrolyzer array; the lower layer adopts a dual rotation lock based on operation rotation lock and fluctuation rotation lock, which comprehensively considers the priority of the electrolyzer fluctuation state and operation state in the long time scale and short time scale respectively, and sets the electrolyzer operation-stop state and fluctuation-steady state operation state. The specific process is: The wind and solar power is used as the total power of the input dispatching system. The upper ACE model makes a first-level power allocation to the supercapacitor, battery, and PEM electrolyzer array based on the incentive function. When the supercapacitor and battery are about to be overcharged, the intelligent agent is encouraged to regulate the discharge of the supercapacitor and battery. When the supercapacitor and battery are about to be over-discharged, the intelligent agent is encouraged to regulate the charging of the supercapacitor and battery. At the same time, the intelligent agent is encouraged to stabilize the power of the PEM electrolyzer array on a long time scale. In the lower-level secondary PEM electrolyzer array control, according to the results of the upper-level first-level power allocation, a dual-rotation lock multi-time scale PEM electrolyzer array optimization scheduling strategy is used to control the operation-stop state of each electrolyzer in the electrolyzer array on a long time scale, and to control the steady-state-fluctuating operation state on a short time scale.

[0028] First, the upper-level real-time scheduling framework of the current scheduling uses reinforcement learning to allocate power to various parts of the system. The standard framework of reinforcement learning in Markov decision process is described as:

[0029] In the formula, is the state space, that is, the overall state description of the system. is the action space, For state transition dynamics, is the discount factor. The goal of the reinforcement learning model is to learn an optimal policy to maximize the expected discounted reward sum The ACE model is an off-policy maximum entropy deep reinforcement learning algorithm based on soft policy iteration, maximizing the entropy-enhanced policy objective function. Its goal includes a policy entropy regularization term in the objective function, with the goal of performing more diverse actions for each given state and visiting states with higher entropy for better exploration, as described below:

[0030] in It’s a strategy The objective function of Indicated in strategy Next, the state-action pair expectations.

[0031] yes The discount factor at the moment, where , used to balance immediate rewards and future rewards. is Status at all times Execute an action Instant rewards; is the weighted term of policy entropy, is the temperature parameter, controlling the importance of entropy. It’s a strategy In Status The entropy under .

[0032] During the operation of the entire system, the state space variables correspond to the state variables of the system under study. The definition of the state space is:

[0033] in , for The input power of wind turbine and photovoltaic at all times; , , They are the capacity of the supercapacitor, the capacity of the battery, and the capacity of the hydrogen storage tank.

[0034] In the current upper-level scheduling model, when the agent observes the states of various parts of the system operation state, it will select the corresponding action from the action space according to the policy function. The definition of the action space is as follows

[0035] in , , They are The battery power, supercapacitor power, and electrolyzer array power controlled by the agent at all times should meet the following constraints when the agent controls the power:

[0036]

[0037]

[0038] in and Indicates the minimum operating power and maximum operating power of the electrolyzer; and Respectively represent the maximum charging and discharging power of the supercapacitor, and and Represents the maximum charging and discharging power of the battery. The actions scheduled by the agent also need to meet the overall constraints of the system power, that is, the load power cannot exceed the input power:

[0039] After the agent selects a set of actions, the environment rewards or punishes them. Reinforcement learning maximizes long-term rewards through continuous trial and error feedback evaluation. The cost function is set according to the goal of the system at runtime, where the cost function consists of an incentive function, a penalty function, and an operating cost.

[0040] The activation function satisfies the leveling criterion as follows:

[0041] in The corresponding reward is the hydrogen production amount. For electrolyzer array The hydrogen production at the moment is calculated based on the scheduling of the lower electrolyzer array. It is the incentive factor for hydrogen production; , They are the incentives obtained when the agent takes measures to alleviate battery overcharge and overdischarge, and when the battery Status of the moment When the battery is too high, that is, when overcharging is likely to occur, discharging the battery will be encouraged. When the over-discharge phenomenon is likely to occur, charging the battery will be motivated. , It is the excitation coefficient for the battery to take action when it is overcharged or over-discharged. is the upper limit of the battery's safe SOC state, It is the lower limit of the battery safety SOC state; , They are the incentives obtained when the agent takes measures to alleviate the overcharge and over-discharge of the supercapacitor. Status of the moment When the overcharge phenomenon may occur, the supercapacitor will be discharged to obtain incentives. When the over-discharge phenomenon is likely to occur, charging the supercapacitor will be stimulated. , It is the excitation coefficient of the supercapacitor when it is overcharged or over-discharged. is the upper limit of the supercapacitor's safe SOC state. It is the lower limit of the supercapacitor's safe SOC state; is the agent's incentive to maximize the utilization of the system's input power. ) and load ) is closer, the higher the incentive is. , , It is the incentive obtained by adjusting the system input power and load balance under different conditions. , It is the boundary value describing the interpolation of input power and load.

[0042] When the agent chooses an inappropriate behavior, the environment will impose a penalty. The penalty function includes battery\supercapacitor SOC over-limit penalty , unable to meet the electrical load balancing penalty , Penalty for battery SOC not returning to initial state after the scheduling period ends , and the long-term volatility penalty assigned to the electrolyzer array , Instantaneous power over limit penalty The penalty function is expressed as follows.

[0043]

[0044] in , represents the instantaneous power over-limit penalty of the battery and supercapacitor, where , Indicates the maximum capacity of batteries and supercapacitors. , Indicates the minimum capacity of batteries and supercapacitors. , Indicates the capacity status of the battery and supercapacitor at the current moment. , , , is the penalty coefficient, is the time interval for power regulation; It is the long-term unbalanced charge and discharge penalty of the battery. is the battery charging efficiency coefficient, is the discharge efficiency coefficient of the battery, is the penalty coefficient, For the current moment, To sum the index, It is at the moment The power of the battery array, is an indicator function, which is 1 when the condition is met; is the volatility penalty for the power allocated to the electrolyzer array, is the moving weighted average, To sum the index, It is at the moment The power of the battery array, is a smoothing parameter that adjusts the focus on historical data and is calculated as: Where N is the half-life; is current The average power predicted in the minute time period, is the maximum safe power of the electrolytic cell. , is the penalty coefficient; is the penalty for not meeting the load balance requirement, where yes The difference between the system input power and the system load power at the moment, is the acceptable boundary of system equilibrium, when Give the agent an appropriate penalty value ,when When the system power imbalance is large, a more severe penalty is given to the agent. , is the penalty coefficient.

[0045] After the upper model completes the real-time power distribution of supercapacitors, batteries, and electrolyzer arrays, the lower layer is responsible for further scheduling the total power of the electrolyzer array. The lower layer scheduling is a dual-rotation multi-time scale PEM electrolyzer array optimization scheduling strategy: the hydrogen production equipment is optimized and scheduled from multiple time scales, including long-term, short-term, and real-time. In the electrolyzer operation (long-term optimization) scheduling stage, the operation scheduling time scale (long-term time scale) is recorded as ,in Based on the historical data assigned to the electrolyzer array by the upper-level dispatch, the ARIMA (average autoregressive model) is used to The time period is season, get the future The power distribution of the electrolyzer array in the time period is predicted, and the future is determined based on this result. The number of electrolytic cells that need to be turned on within the time period is determined by sorting the start and stop status of the electrolytic cells in the main loop of the electrolytic cell run-stop scheduling based on the shortest running time as the first priority and the fluctuating running time as the second priority.

[0046] However, for determining the optimal stable power of the actual operation of the PEM electrolyzer, the operation scheduling time scale (long-term time scale) The gap with the actual situation is still not negligible. In order to reduce the impact of wind and solar power fluctuations on the life of the electrolyzer, it is necessary to Further corrections are made internally. Here, the electrolyzer fluctuation-steady-state scheduling (short-term rolling optimization scheduling) is used to further adjust it. The fluctuation-steady-state scheduling time scale (short-term time scale) is recorded as ,exist During the time period The electrolytic cell is rolled at intervals of minutes. The state switching polling control of the fluctuation-steady state operation state is satisfy , .

[0047] Proton exchange membrane electrolyzers have smaller size and weight, lower power consumption, inherent ability to handle transient electrical power changes, high purity of gas, and the possibility of compressing air at higher pressures and higher safety levels within the device. The voltages of the electrolyzers are as follows:

[0048] in is the open circuit voltage; is the activation polarization voltage; is the ohmic polarization voltage, the Nernst equation for the electrolytic cell is shown below:

[0049] in is the standard electrode potential, i.e. the theoretical minimum electrolysis voltage under standard conditions; is the universal gas constant; is the operating temperature of the electrolyzer (unit: Kelvin); is the Faraday constant; is the partial pressure of hydrogen is the partial pressure of oxygen; is the activity of water.

[0050] The above formula describes the relationship between the reversible electromotive force of the battery at a certain temperature and the activity of each component participating in the reaction. The standard voltage can be defined as:

[0051] in is the standard Gibbs free energy change.

[0052] The activation polarization voltage can be defined as:

[0053] in is the working current of the electrolytic cell, Standard current The expression of ohmic polarization voltage is:

[0054] in is the membrane resistance, and the expression of membrane resistance is:

[0055] in is the thickness of the film, is the conductivity of the membrane, expressed as:

[0056] in For hydration.

[0057] Electrolyzer efficiency It refers to the efficiency of the electrolyzer in converting electrical energy into hydrogen. The efficiency of PEM electrolyzers is usually between 60% and 80%, depending on the type of electrolyzer, the operating conditions of the electrolyzer and the maintenance of the electrolyzer.

[0058]

[0059]

[0060]

[0061] in is the current efficiency, is the voltage efficiency, is the electrolyzer efficiency, is the thermal neutral voltage, is the actual electrolysis voltage.

[0062] The relationship between electrolyzer efficiency and electrolyzer input power is shown in Figure 2. It reflects the changing trend of the electrolyzer efficiency with the input power. It can be seen that as the input power increases, the electrolyzer efficiency first rises to a peak value, reaching a maximum of about 80% at about 0.5 MW, and then begins to decline. This is because excessive power will cause the electrolysis reaction rate to exceed the tolerance of the electrolyte membrane, and part of the electrical energy will be converted into heat energy and no longer participate in electrolysis. When the input power continues to rise to 1.5 MW, although the efficiency decreases, the hydrogen production is the largest. Therefore, considering the efficiency, hydrogen production and hydrogen purity, the optimal working range of the electrolyzer is 0.5~1.5MW.

[0063] In order to determine each The number of electrolyzers that need to be opened within a minute period needs to be predicted The average power of the electrolyzer array in the minute time period. Here, the difference integrated moving average autoregressive model (ARIMA) model is used to predict the average power. The general model of ARIMA is as follows

[0064]

[0065]

[0066] in: is a backshift operator that satisfies , is the moment of input to the ARIMA model power, is the difference operator, is a white noise process, is an autoregressive polynomial, is the moving average polynomial, By aggregating the second-level data into the average value per minute, the fluctuation of the time series is further eliminated, which improves the future The accuracy of minute prediction, the specific prediction formula is as follows:

[0067]

[0068]

[0069] in It's in time For the future The predicted value of the step, is a data point, It represents the actual observed value in the past. The future predicted value, represents the order of the autoregressive part, Indicates the order of the sliding average part.

[0070] The future is obtained through the ARIMA model Average power forecast for the minute , combined with the constraints of the optimal stable power range of the PEM electrolyzer and the safe operating power of the PEM electrolyzer, the snake optimization algorithm is used to maximize the hydrogen production to determine the future The number of electrolyzers working in a minute. electrolytic cell, in In the minute time period The electrolytic cells are in operation, and one of the electrolytic cells in operation is in a fluctuating operation state, that is, .

[0071] The optimal stable power operation range constraint of PEM electrolyzer is as follows:

[0072] , They represent the upper and lower limits of the optimal working range of the electrolyzer based on the power-efficiency model, Indicates the first Optimal stable power for each PEM electrolyser.

[0073] Power constraint formula for safe operation of PEM electrolyzer caused by fluctuations in hydrogen production equipment:

[0074] , They represent the upper and lower power limits that a PEM electrolyzer with fluctuating power must comply with due to safety restrictions. Indicates time period Neidi Safe operating power of an EM electrolyzer.

[0075] The power of all electrolyzers in the electrolyzer array needs to satisfy the following constraint, that is, the total power allocated to all electrolyzers cannot exceed the total power allocated to the electrolyzers by the upper model, where yes Moment The operating power of an electrolyzer, is the total number of electrolytic cells.

[0076]

[0077] Determine the electrolytic cell Number of jobs in time scale Taking the maximum hydrogen production as the optimization target, the objective function is proposed by taking into account the power efficiency characteristics of the PEM electrolyzer. The objective function of the PEM electrolyzer power-efficiency model is as follows:

[0078] Indicates the time period of hydrogen production equipment The total amount of hydrogen produced in Indicates the number of PEM electrolyzers; Indicates The PEM electrolyzer is The total volume of hydrogen produced in a time period is equal to the hydrogen production per hour of the PEM electrolyzer at the operating power. The product of time.

[0079] PEM electrolyzer at operating power Hydrogen production per hour As shown in the formula, the unit is cubic meter / hour, which means the flow rate when the temperature is 24° and the pressure is 1MPa.

[0080]

[0081] In the formula, 22.4 is the molar volume of the gas under standard conditions, is the rated power of the PEM electrolyzer, is the rated voltage of the PEM electrolyzer, It is the higher calorific value of hydrogen.

[0082] Polling control is based on the improvement of the rotation lock. Each electrolyzer (off state, running state (steady-state operation and fluctuating operation)) has a double rotation lock, a fluctuation rotation lock and an operation rotation lock. The fluctuation rotation lock is used to count the accumulated fluctuating state running time of the electrolyzer, and the operation rotation lock is used to count the accumulated running time of the electrolyzer. In the current time period, the working state of each PEM electrolyzer in the electrolyzer array and the selection of the PEM electrolyzer in the fluctuating power state are determined by the rotation lock of each PEM electrolyzer. The digital rotation lock is a control bit and is only used for the polling control of this application. The fluctuation rotation lock has the following characteristics: (1) All volatility locks are positive integers; (2) The larger the value displayed on the fluctuation rotation lock of a single PEM electrolyzer, the longer the cumulative operation time of the PEM electrolyzer in the fluctuation state is, which also proves that the probability of the PEM electrolyzer being in the fluctuation state in the future is smaller; (3) Based on the second characteristic, the initial value of the fluctuation rotation lock of each PEM electrolyzer is set to 0; The running rotation lock has the following characteristics: (1) All running rotation locks are positive integers; (2) The larger the value displayed on the operation rotation lock of a single PEM electrolyzer, the longer the cumulative operation time of the PEM electrolyzer, which also proves that the probability of the PEM electrolyzer being in the open operation state is smaller; (3) Based on the second characteristic, the initial value of the operation rotation lock of each PEM electrolyzer is set to 0; The rotation update process of the volatility rotation lock is: by is a time period, and at the end of the scheduling time period, the fluctuation rotation lock of the electrolytic cell array is updated. The electrolytic cells are in a fluctuating operation state Minutes, Fluctuation lock for each electrolyzer Plus one, to avoid overflow, when the volatility lock reaches the maximum value, it will be scaled, where is the scaling factor:

[0083] The rotation update process of running the rotation lock is: by is a time period, and at the end of the scheduling time period, the operation rotation lock of the electrolytic cell array is updated. Electrolyzers in operation Minutes, Electrolyzer operation rotation lock Plus one, to avoid overflow, when the volatility lock reaches the maximum value, it will be scaled:

[0084] According to the update process of these two rotating locks, the optimal scheduling strategy of PEM electrolyzer array with dual rotating locks and multiple time scales is described as follows: When the system starts running, set the initial rotation lock value of each device to 0. Enter the long time scale Electrolyzer operation - stop the main scheduling loop, and start from the current moment according to the historical data allocated by the upper model Average power of the total electrolyzer array over a period of minutes , according to the modeling constraints, the snake optimization algorithm is used to determine the number of electrolytic cells that need to be opened . Determine whether the operation status needs to be updated. If necessary, sort and select the equipment. Select from high to low priority according to the electrolytic cell with the shortest operation time and the shortest operation time in the fluctuating state, the electrolytic cell with a relatively short operation time and the shortest operation time in the fluctuating state, the electrolytic cell with a relatively short operation time and a relatively short operation time in the fluctuating state, and the electrolytic cell with a relatively long operation time and a relatively long operation time in the fluctuating state. The specific selection process is as follows: a. First press Run Rotation Lock Sort the devices into buckets from smallest to largest.

[0085] b. On this basis, according to the fluctuation rotation lock Sort the files twice from smallest to largest.

[0086] c. Select the first after sorting A device is used as an operating device. When multiple electrolytic cells have the same ranking, the roulette algorithm is used to select whether to start the electrolytic cells with the same ranking.

[0087] In determining After the electrolyzer is running in the minute time period, it enters the short-term time scale The electrolytic cell fluctuation-steady-state scheduling sub-cycle performs the electrolytic cell fluctuation-steady-state operation state control, and selects them from high to low priority according to the electrolytic cell with the shortest fluctuation state operation time and the shortest operation time, the electrolytic cell fluctuation state operation time is relatively short and the shortest operation time, the electrolytic cell fluctuation state operation time is relatively short and the operation time is relatively short, and the electrolytic cell fluctuation state operation time is relatively long and the operation time is relatively long. The specific selection process is as follows.

[0088] a. Exclude the last fluctuation device from the running devices.

[0089] b. For the remaining optional devices, press Sort the buckets from small to large, and then sort them according to Perform secondary bucket sort from small to large.

[0090] c. Select the first device after sorting as the current fluctuation device.

[0091] d. Update the wave device , add 1.

[0092] Repeating short time scales Sub-cycle period z times, , forming a complete long-time scale main cycle. At the end of each cycle, the rotation locks of all running devices are updated, and at the same time, it is checked whether any rotation lock exceeds the maximum value. If a and If the maximum value is exceeded, a scaling factor is used Reduce all rotating locks by equal proportion; Then calculate all and The average value of the rotation lock is obtained by dividing the current value of the rotation lock by its average value to obtain the relative value of the rotation lock of all devices. The relative value of the rotation lock is used as the initial rotation lock of the next cycle to provide a basis for the decision of the next cycle.

[0093] In this embodiment 15min, is 5min, z is 3.

[0094] Any matters not described in the present invention are applicable to the prior art.

Claims

1. A double-layer multi-time scale scheduling method for an off-grid wind-solar hybrid hydrogen production system, characterized in that: The method adopts a two-layer scheduling mechanism based on the dual goals of improving system economy and ensuring equipment safety. The upper layer is centered on the intelligent agent, and adopts a first-level power allocation of an incentive function that comprehensively considers extending the service life of the equipment, reducing overcharge and overdischarge of energy storage components, and stabilizing the fluctuation of the PEM electrolyzer array. The wind and solar power are used as the total power of the input system and allocated to the battery, supercapacitor, and PEM electrolyzer array. The control of the lower-level secondary PEM electrolyzer array relies on the results of the primary power allocation. Taking into account the goals of maximizing hydrogen production and extending the service life of the PEM electrolyzer, a dual-rotation lock multi-time scale PEM electrolyzer array optimization scheduling strategy is used to control the PEM electrolyzer and complete the dynamic allocation of power. The dual-rotation lock multi-time scale PEM electrolyzer array optimization scheduling strategy is: each electrolyzer has a dual-rotation lock consisting of a fluctuation rotation lock and an operation rotation lock, and the fluctuation rotation lock Used to count the accumulated fluctuating state operation time of the electrolytic cell and run the rotation lock Used to count the cumulative running time of the electrolyzer; in the current time period, the working state of each PEM electrolyzer in the electrolyzer array and the selection of the PEM electrolyzer in the fluctuating power state are determined by the rotation lock of each PEM electrolyzer; When the system starts running, set the initial rotation lock value of each device to 0; enter the long time scale Electrolyzer operation - stop the main scheduling loop, and predict the current time based on the historical data allocated by the upper model The total average power of the electrolyzer array during the time period , using intelligent algorithms to determine the number of electrolytic cells that need to be opened ; Determine whether the operation status needs to be updated. If necessary, sort and select the equipment. Sorting and selecting are carried out from high to low priority according to the electrolytic cell with the shortest operation time and the shortest operation time in the fluctuating state, the electrolytic cell with a relatively short operation time and the shortest operation time in the fluctuating state, the electrolytic cell with a relatively short operation time and a relatively short operation time in the fluctuating state, and the electrolytic cell with a relatively long operation time and a relatively long operation time in the fluctuating state. The specific selection process is: a. First press Run Rotation Lock Sort the devices by bucket from smallest to largest; b. On this basis, according to the fluctuation rotation lock Sort the buckets twice from small to large; c. Select the first after sorting A device is used as an operating device. When multiple electrolytic cells have the same ranking, the roulette algorithm is used to select whether to start the electrolytic cells with the same ranking. In determining After the electrolyzer is running for a period of time, it enters the short-term time scale The electrolytic cell fluctuation-steady-state scheduling sub-cycle is used to control the electrolytic cell fluctuation-steady-state operation state. The electrolytic cell fluctuation state operation time is the shortest and the operation time is the shortest, the electrolytic cell fluctuation state operation time is relatively short and the operation time is the shortest, the electrolytic cell fluctuation state operation time is relatively short and the operation time is short, and the electrolytic cell fluctuation state operation time is relatively short and the operation time is short, and the electrolytic cell fluctuation state operation time is relatively long and the operation time is relatively long. The specific selection process is: a. Exclude the last fluctuation device from the running device; b. For the remaining optional devices, press Sort the buckets from small to large, and then sort them according to Sort the buckets twice from small to large; c. Select the first device after sorting as the current fluctuation device; d. Update the wave device , add 1; Repeating short time scales Sub-cycle period z times, where , forming a complete long-time main cycle; at the end of each cycle, the rotation locks of all running devices are updated, and at the same time, it is checked whether any rotation lock exceeds the maximum value If there is a and If the maximum value is exceeded, a scaling factor is used Reduce all rotating locks by equal proportion; Then calculate all and The average value of the rotating lock is obtained by dividing the current value of the rotating lock by its average value to obtain the relative value of the rotating lock of all devices, and the relative value of the rotating lock is used as the initial rotating lock of the next cycle.

2. The method according to claim 1, characterized in that: Taking wind power and solar power as the input power of hybrid energy storage, the upper layer takes the ACE model as the core, comprehensively considering the goals of suppressing overcharge and over-discharge of equipment, maximizing hydrogen production, and smoothing the input power of the electrolyzer array; The upper ACE model allocates primary power to supercapacitors, batteries, and PEM electrolyzer arrays based on the incentive function. When supercapacitors and batteries are about to be overcharged, the agent is encouraged to regulate the discharge of supercapacitors and batteries; when supercapacitors and batteries are about to be over-discharged, the agent is encouraged to regulate the charging of supercapacitors and batteries. At the same time, the agent is encouraged to stabilize the power of the PEM electrolyzer array over a long time scale. In the lower-level secondary PEM electrolyzer array control, according to the results of the upper-level primary power allocation, a double-rotation lock multi-time scale PEM electrolyzer array optimization scheduling strategy is used to control the run-stop state of each electrolyzer in the electrolyzer array on a long time scale, and to control the steady-state-fluctuating operation state on a short time scale.

3. The method according to claim 1, characterized in that ,The upper level power allocation includes a comprehensive definition of the ,state space, action space and incentive function of the hydrogen production system that ,can produce hydrogen from renewable energy, and converts the optimal scheduling ,problem of off-grid wind-solar complementary hydrogen production system into a Markov ,decision process (MDP).

4. The method according to claim 3, characterized in that: State Space is defined as: in , for The input power of wind turbine and photovoltaic at all times; , , They are the capacity of the supercapacitor, the capacity of the battery, and the capacity of the hydrogen storage tank; Action Space is defined as: in , , They are The power of the battery, supercapacitor, and electrolyzer array controlled by the agent at all times; when the agent controls the power, the following constraints are met: in and Indicates the minimum operating power and maximum operating power of the electrolyzer; and Respectively represent the maximum charging and discharging power of the supercapacitor, and and Indicates the maximum charging and discharging power of the battery; Activation Function Satisfy the leveling criterion, expressed as: in The corresponding reward is the hydrogen production amount. For electrolyzer array The hydrogen production at the moment is calculated based on the scheduling of the lower electrolyzer array. It is the incentive factor for hydrogen production; , They are the incentives obtained when the agent takes measures to alleviate battery overcharge and overdischarge, and when the battery Status of the moment When the battery is too high, that is, when overcharging is likely to occur, discharging the battery will be encouraged. When the over-discharge phenomenon is likely to occur, charging the battery will be motivated. , It is the excitation coefficient for the battery to take action when it is overcharged or over-discharged. is the upper limit of the battery's safe SOC state, It is the lower limit of the battery safety SOC state; , They are the incentives obtained when the agent takes measures to alleviate the overcharge and over-discharge of the supercapacitor. Status of the moment When the overcharge phenomenon may occur, the supercapacitor will be discharged to obtain incentives. When the over-discharge phenomenon is likely to occur, charging the supercapacitor will be stimulated. , It is the excitation coefficient of the supercapacitor when it is overcharged or over-discharged. is the upper limit of the supercapacitor's safe SOC state. It is the lower limit of the supercapacitor's safe SOC state; It is the agent's incentive to maximize the utilization of the system input power. With load The closer you are, the higher the incentive you get. , , It is the incentive obtained by adjusting the system input power and load balance under different conditions. , is the boundary value describing the input power and load interpolation; When the agent chooses inappropriate behavior, the environment will impose penalties. The penalty functions include battery\supercapacitor SOC over-limit penalty, failure to meet electrical load balance penalty, battery SOC failure to return to the initial state after the scheduling period ends, long-term volatility penalty assigned to the electrolyzer array, and instantaneous power over-limit penalty.

5. The method according to claim 4, characterized in that ,Battery\Supercapacitor SOC over-limit penalty , Penalty for battery SOC not returning to initial state after the scheduling period ends , a long-term volatility penalty assigned to the electrolyzer array , and instantaneous power over-limit penalty and failure to meet the electrical load balancing penalty The penalty functions are expressed as: in , represents the instantaneous power over-limit penalty of the battery and supercapacitor, where , Indicates the maximum capacity of batteries and supercapacitors. , Indicates the minimum capacity of batteries and supercapacitors. , Indicates the capacity status of the battery and supercapacitor at the current moment. , , , is the penalty coefficient, is the time interval for power regulation; It is the long-term unbalanced charge and discharge penalty of the battery. is the battery charging efficiency coefficient, is the discharge efficiency coefficient of the battery, is the penalty coefficient, For the current moment, To sum the index, It is at the moment The power of the battery array, is an indicator function, which is 1 when the condition is met; is the volatility penalty for the power allocated to the electrolyzer array, is the moving weighted average, To sum the index, It is at the moment The power of the battery array, is the smoothing parameter; is current The average power predicted in the minute time period, is the maximum safe power of the electrolytic cell. , is the penalty coefficient; is the penalty for not meeting the electrical load balance, where yes The difference between the system input power and the system load power at the moment, is the acceptable boundary of system equilibrium, when Give the agent an appropriate penalty value ,when When the system power imbalance is large, a more severe penalty is given to the agent. , is the penalty coefficient.

6. The method according to claim 3, characterized in that The fluctuation rotation lock has the following characteristics: (1) All volatility locks are positive integers; (2) The larger the value displayed on the fluctuation rotation lock of a single PEM electrolyzer, the longer the cumulative operation time of the PEM electrolyzer in the fluctuation state is, which also proves that the probability of the PEM electrolyzer being in the fluctuation state in the future is smaller; (3) Based on the second characteristic, the initial value of the fluctuation rotation lock of each PEM electrolyzer is set to 0; The running rotation lock has the following characteristics: (1) All running rotation locks are positive integers; (2) The larger the value displayed on the operation rotation lock of a single PEM electrolyzer, the longer the cumulative operation time of the PEM electrolyzer, which also proves that the probability of the PEM electrolyzer being in the open operation state is smaller; (3) Based on the second characteristic, the initial value of the operation rotation lock of each PEM electrolyzer is set to 0; The rotation update process of the volatility rotation lock is: by is a period, and at the end of the scheduling period, the fluctuation rotation lock of the electrolytic cell array is updated. The electrolytic cells are in a fluctuating operation state , No. Fluctuation lock for each electrolyzer Add one when the volatility rotation lock reaches the maximum value , scaling is performed, where is the scaling factor: ； The rotation update process of running the rotation lock is: by is a period of time. At the end of the scheduling period, the operation rotation lock of the electrolytic cell array is updated. Electrolyzers in operation Minutes, Electrolyzer operation rotation lock Plus one, when the volatility lock reaches the maximum value, it will be scaled: 。 7. The method according to claim 1, characterized in that ,The intelligent algorithm is a snake optimization algorithm, a genetic algorithm, or a particle swarm algorithm.

8. The method according to claim 1, characterized in that , z is an integer greater than 2.

Images