Wind-storage joint frequency regulation control method based on model predictive control and deep learning
By combining model prediction control and deep learning, a combined frequency regulation control method of wind storage is constructed, and the multi-constraint processing problem of wind storage system and the problem of long-term solution time is solved, faster and more economical wind storage joint frequency regulation control is achieved, and the power system is improved stability.
Patent Information
- Application Number
- CN202410422413.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-04-09
- Publication Date
- 2024-11-19
- Estimated Expiration
- 2044-04-09
AI Technical Summary
The existing frequency modulation control method of wind storage joint system is difficult to deal with multiple constraints simultaneously, and the traditional model prediction control is long solution in high-dimensional conditions, making it difficult to meet the real-time requirements.
Using the combined frequency modulation control method of wind storage based on model prediction control (MPC) and deep learning, we construct energy storage equivalent model and discrete system prediction model suitable for wind storage joint frequency modulation, and map the optimal power instruction sequence through DNN training to optimize Frequency control of wind storage system.
It improves the MPC solution speed, reduces the control cycle and frequency regulation cost, enhances the frequency regulation capability of the wind storage joint system, and adapts to the power system stability needs after the increase in wind power permeability.
Smart Images

Figure CN118353032B_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the field of frequency regulation control strategy of hybrid energy storage participating in wind power grid in the direction of new energy, and specifically is a wind-storage joint frequency regulation control method based on model predictive control and deep learning. Background Art
[0002] With the increasing proportion of renewable energy in the energy structure, especially the rapid development of new energy such as wind power and solar power, my country has made important achievements in the energy field. By the end of June 2023, my country's installed capacity of renewable energy power generation was about 2.71 billion kilowatts, accounting for about 48.8%, exceeding traditional coal-fired power. The installed capacity of wind power was about 390 million kilowatts, an increase of 13.7% year-on-year, becoming an important part of the power system. However, as the proportion of new energy in the power grid continues to increase, its volatility and intermittency have brought challenges to the stability of the system. Large-scale wind power access will weaken the frequency regulation inertia capacity of the system, making it difficult for the system to cope with load fluctuations. In order to ensure the stable operation of the system, the development of energy storage technology has become crucial. It is expected that by 2030, the installed capacity of wind power and solar power generation will exceed coal-fired power and become the main body. The advancement of energy storage technology will greatly increase the installed capacity of energy storage, which is expected to reach more than 30 million kilowatts by 2025.
[0003] At present, the frequency regulation of wind-storage system is mostly based on traditional control methods, especially the method of filtering out the high-frequency component of the power command through a filter and issuing it to the energy storage system, and the remaining low-frequency component is widely used. However, the frequency regulation scenario of wind-storage joint system has many constraints compared with conventional control, especially when multiple groups or multiple energy storage are configured, and the above methods are difficult to handle many constraints in the wind-storage system at the same time and are not optimal. Model predictive control can solve these problems better. It is a model-based optimal control technology. MPC belongs to optimal control. Generally, it is solved by optimization solvers such as Cplex and Gurobi. However, when solving MPC problems with higher dimensions, the time required for solving using optimization solvers increases exponentially with the increase of decision variables and constraint dimensions. For application scenarios with short control cycles and high variable dimensions of wind-storage joint frequency regulation, it may be difficult to meet the system requirements. Summary of the invention
[0004] The present invention proposes a wind-storage joint frequency regulation control method based on model predictive control and deep learning.
[0005] The technical solution to achieve the purpose of the present invention is: a wind-storage joint frequency modulation control method based on model predictive control and deep learning, comprising:
[0006] Construct an energy storage equivalent model suitable for wind-storage combined frequency regulation application scenarios;
[0007] Construct a discrete system prediction model for wind-storage combined system participating in frequency regulation, predict the system state from step k+1 to step k+Np, set corresponding constraints and objective functions, traverse different system load fluctuations, system frequency deviations, and energy storage system SOC conditions offline, respectively obtain the optimal effective inequality constraint set, and number the system working modes according to the different inequality constraints satisfied;
[0008] With system load fluctuation, system frequency deviation, and energy storage system SOC as input and system working mode number as output, DNN is trained so that it can obtain the current optimal effective inequality constraint set and map the optimal power instruction sequence according to the current system load fluctuation, system frequency deviation, energy storage system SOC state and the predicted system load fluctuation, system frequency deviation, and energy storage system SOC state in the predicted time domain;
[0009] The first one in the optimal power instruction sequence is selected as the wind-storage system power instruction and issued to the wind-storage system.
[0010] Preferably, the energy storage equivalent model suitable for the wind-storage combined frequency regulation application scenario is constructed as follows:
[0011] P battery =u lb i lb
[0012] Where P battery is the energy storage output power, u lb =E lb -R in_lb ·i lb ,
[0013]
[0014] In the formula, E lb Indicates no-load voltage; E 0_lb represents constant voltage; K represents polarization voltage; Q lb Indicates capacity; i lb represents the current; A represents the amplitude in the exponential region; B represents the inverse of the time constant in the exponential region in constant current discharge;
[0015] R in_lb =R 1-lb ·N s_lb / N p_lb
[0016] Where N s_lb N is the number of energy storage series, p_lb is the number of energy storage in parallel, R 1-lb It is the single resistance value.
[0017] Preferably, a discrete system prediction model for the wind-storage combined system participating in frequency modulation is constructed, and the specific process of predicting the system state from step k+1 to step k+Np is as follows:
[0018] Construct the system state equation group in the primary frequency regulation process of the wind-storage combined system:
[0019]
[0020] In the formula, G gov (s) is the speed regulator transfer function, G gen (s) = G gen1 (s)G gen2 (s) is the turbine transfer function, G gen1 (s) is the non-reheat steam turbine transfer function, G gen2 (s) is the transfer function G of the main inlet steam chamber of the reheat steam turbine sc (s) is the supercapacitor transfer function, G battery (s) is the transfer function of lithium battery, G wind (s) is the wind turbine pitch control transfer function, P v is the output power of the speed regulator, P m is the turbine output power, ΔP g is the synchronous machine output power increment, ΔP sc is the supercapacitor output power increment, ΔP scref is the supercapacitor output power increment command value, ΔS sc is the supercapacitor SOC change, T is the control period, E sc is the supercapacitor capacity, ΔP battery is the output power increment of the lithium battery, ΔP batteryref is the lithium battery output power increment command value, ΔS battery is the change in SOC of the lithium battery, E battery is the capacity of lithium battery, ΔP wind is the wind turbine output power increment, ΔP windref is the incremental command value of wind turbine output power, Δf is the system frequency deviation, H is the system inertia time constant, D is the load regulation coefficient, P load is the system load fluctuation, 1 / R is the droop coefficient; P load For load fluctuation.
[0021] The system state equations are discretized, and the discrete system prediction model is obtained as follows:
[0022]
[0023] The k or k+1 in brackets after the variable indicates the variable value of the corresponding step. The k in brackets indicates the predicted value or actual value of the kth step. The k+1 in brackets indicates the predicted value of the k+1th step. v is the output power of the speed regulator, P m is the turbine output power, ΔP g is the synchronous machine output power increment, ΔP sc is the supercapacitor output power increment, ΔP scref is the supercapacitor output power increment command value, S sc is the supercapacitor SOC, T is the control period, E sc is the supercapacitor capacity, ΔP battery is the output power increment of the lithium battery, ΔP batteryref is the lithium battery output power increment command value, S battery is the lithium battery SOC, E battery is the capacity of lithium battery, ΔP wind is the wind turbine output power increment, ΔP windref is the incremental command value of wind turbine output power, Δf is the system frequency deviation, T G is the speed regulator time constant, 1 / R is the droop coefficient, T CH is the main steam inlet chamber time constant, T RH is the reheater time constant, T SC , T B , T WIND are the time constants of supercapacitor, lithium battery and wind turbine respectively, H is the system inertia time constant, D is the load regulation coefficient, P load is the system load wave.
[0024] Preferably, the objective function is set as:
[0025]
[0026] Where N p is the prediction time domain step size, C wind (k+i|k) is the fan frequency regulation cost of the k+i-th step predicted in the k-th step; C battery (k+i|k) is the k+i-step lithium battery frequency modulation cost predicted in the k-step; C sc (k+i|k) is the k+i-th step supercapacitor frequency modulation cost predicted in the k-th step.
[0027] Preferably, the fan frequency regulation cost of the k+ith step predicted in the kth step is specifically:
[0028] C wind (k+i|k)=k wind ΔP wind (k+i|k) 2
[0029] Among them, k wind is the wind power frequency regulation cost coefficient, ΔP wind (k+i|k) is the fan frequency modulation power of the k+ith step predicted in the kth step.
[0030] Preferably, the k+i-th step lithium battery frequency modulation cost predicted in the k-th step is specifically:
[0031] C battery (k+i|k)=k battery ΔP battery (k+i|k) 2
[0032] Among them, k battery is the lithium battery frequency modulation cost coefficient, ΔP battery (k+i|k) is the k+i-th step lithium battery frequency modulation power predicted in the k-th step.
[0033] Preferably, the k+i-th step supercapacitor frequency modulation cost predicted in the k-th step is specifically:
[0034] C sc (k+i|k)=k sc ΔP sc (k+i|k) 2
[0035] Among them, k sc is the supercapacitor frequency modulation cost coefficient, ΔP sc (k+i|k) is the supercapacitor frequency modulation power of the k+ith step predicted in the kth step.
[0036] Preferably, the set constraints include:
[0037] -P W ≤ΔP wind (k+i|k)≤P W
[0038] -P B ≤ΔP battery (k+i|k)≤P B
[0039] S batterymin ≤S battery (k+i|k)≤S batterymax
[0040] -P S ≤ΔP sc (k+i|k)≤P S
[0041] S scmin ≤S sc(k+i|k)≤S scmax
[0042] P(k+i|k)=K·Δf(k+i|k)
[0043] Among them, P W is the upper limit of wind power output, -P W is the lower limit of wind power output, ΔP wind (k+i|k) is the predicted value of wind power frequency regulation output at the current sampling time k for k+i steps, P B The upper limit of lithium battery output, -P B is the lower limit of lithium battery output, ΔP battery (k+i|k) is the predicted value of the k+i-step lithium battery frequency modulation output at the current sampling time k, P S is the upper limit of supercapacitor output, -P is the lower limit of supercapacitor output, ΔP sc (k+i|k) is the predicted value of the supercapacitor frequency modulation output at the current sampling time k for k+i steps, S batterymin is the lower limit of SOC of lithium battery, S batterymax is the upper limit of SOC of lithium battery, S battery (k+i|k) is the predicted value of the lithium battery SOC for k+i steps at the current sampling time k, S scmin is the lower limit of supercapacitor SOC, S scmax is the upper limit of supercapacitor SOC, S sc (k+i|k) is the predicted value of the supercapacitor SOC for the k+i step at the current sampling moment k, P(k+i|k) is the predicted value of the frequency regulation output demand of the wind storage system for the k+i step at the current sampling moment k, K is the coefficient, Δf(k+i|k) is the predicted value of the system frequency deviation for the k+i step at the current sampling moment k.
[0044] Compared with the prior art, the present invention has the following significant advantages: in the future, when the wind power penetration rate in the power system is further improved, in order to ensure the effective absorption of wind power and eliminate the impact of wind power volatility and uncertainty on the power system, the energy storage configuration of wind farms will inevitably follow up synchronously and the scale of energy storage will continue to expand. Therefore, an energy storage equivalent mathematical model suitable for wind-storage joint frequency regulation control is constructed, and the MPC method that can better solve problems under multiple constraints is used to solve the wind-storage joint frequency regulation control; in view of the shortcomings of the traditional MPC solution method that is slow and large-scale solution is difficult to meet the real-time requirements, an improved wind-storage joint frequency regulation control solution method based on deep learning is proposed, which greatly improves the MPC solution speed and reduces the control cycle and frequency regulation cost.
[0045] The present invention is described in further detail below in conjunction with the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS
[0046] Figure 1It is the structure diagram of the simulation verification system.
[0047] Figure 2 It is the comparison of system frequency under different strategies.
[0048] Figure 3 It is a comparison of the frequency modulation output of each part under different control strategies.
[0049] Figure 4 It is a comparison of system frequency regulation costs under different control strategies.
[0050] Table 1 is the specific parameter configuration in the simulation verification. DETAILED DESCRIPTION
[0051] The present invention is further described below in conjunction with specific embodiments.
[0052] A wind-storage joint frequency regulation control method based on model predictive control and deep learning comprises the following steps:
[0053] Step 1: Considering the nonlinear characteristics of the battery and requiring a simple circuit structure, a universal energy storage equivalent mathematical model that can better describe various types of batteries is constructed;
[0054] In a further embodiment, the energy storage equivalent mathematical model is specifically:
[0055]
[0056] In the formula, E lb Indicates no-load voltage; E 0_lb represents constant voltage; K represents polarization voltage; Q lb Indicates capacity; i lb represents the current; A represents the amplitude in the exponential region; B represents the inverse of the time constant in the exponential region in constant current discharge.
[0057] Consider the number of energy storage series is N s_lb , the number of parallel connections is N p_lb , the monomer resistance is R 1-lb , then the energy storage internal resistance is:
[0058] R in_lb =R 1-lb ·N s_lb / N p_lb
[0059] Available energy storage output power
[0060] P battery =u lb i lb
[0061] in,
[0062] ulb =E lb -R in_lb ·i lb
[0063] Step 2: Construct a discrete system prediction model for the wind-storage combined system participating in frequency regulation, and predict the system state from step k+1 to step k+Np;
[0064] Construct the system state equation group in the primary frequency regulation process of the wind-storage combined system:
[0065]
[0066] In the formula, G gov (s) is the speed regulator transfer function, G gen (s) = G gen1 (s)G gen2 (s) is the turbine transfer function, G gen1 (s) is the non-reheat steam turbine transfer function, G gen2 (s) is the transfer function G of the main inlet steam chamber of the reheat steam turbine sc (s) is the supercapacitor transfer function, G battery (s) is the transfer function of lithium battery, G wind (s) is the wind turbine pitch control transfer function, P v is the output power of the speed regulator, P m is the turbine output power, ΔP g is the synchronous machine output power increment, ΔP sc is the supercapacitor output power increment, ΔP scref is the supercapacitor output power increment command value, ΔS sc is the supercapacitor SOC change, T is the control period, E sc is the supercapacitor capacity, ΔP battery is the output power increment of the lithium battery, ΔP batteryref is the lithium battery output power increment command value, ΔS battery is the change in SOC of the lithium battery, E battery is the capacity of lithium battery, ΔP wind is the wind turbine output power increment, ΔP windref is the incremental command value of wind turbine output power, Δf is the system frequency deviation, H is the system inertia time constant, D is the load regulation coefficient, P load is the system load fluctuation, 1 / R is the droop coefficient; P load For load fluctuation.
[0067] The system state equations are discretized, and the discrete system prediction model is obtained as follows:
[0068]
[0069] The k or k+1 in brackets after the variable indicates the variable value of the corresponding step. The k in brackets indicates the predicted value or actual value of the kth step. The k+1 in brackets indicates the predicted value of the k+1th step. v is the output power of the speed regulator, P m is the turbine output power, ΔP g is the synchronous machine output power increment, ΔP sc is the supercapacitor output power increment, ΔP scref is the supercapacitor output power increment command value, S sc is the supercapacitor SOC, T is the control period, E sc is the supercapacitor capacity, ΔP battery is the output power increment of the lithium battery, ΔP batteryref is the lithium battery output power increment command value, S battery is the lithium battery SOC, E battery is the capacity of lithium battery, ΔP wind is the wind turbine output power increment, ΔP windref is the incremental command value of wind turbine output power, Δf is the system frequency deviation, T G is the speed regulator time constant, 1 / R is the droop coefficient, T CH is the main steam inlet chamber time constant, T RH is the reheater time constant, T SC , T B , T WIND are the time constants of supercapacitor, lithium battery and wind turbine respectively, H is the system inertia time constant, D is the load regulation coefficient, P load is the system load fluctuation. Then, the discrete equations of the system can be used to predict the k+1 to k+N p The system status within the step.
[0070] Set the objective function to the frequency regulation cost of the wind-storage system in the forecast time domain
[0071]
[0072] C wind (k+i|k) is the fan frequency regulation cost of the k+i-th step predicted in the k-th step:
[0073] C wind (k+i|k)=k wind ΔP wind (k+i|k) 2
[0074] Among them, k wind is the wind power frequency regulation cost coefficient, ΔP wind(k+i|k) is the fan frequency modulation power of the k+i-th step predicted in the k-th step
[0075] C battery (k+i|k) is the k+i-step lithium battery frequency modulation cost predicted in the k-step:
[0076] C battery (k+i|k)=k battery ΔP battery (k+i|k) 2
[0077] Among them, k battery is the lithium battery frequency modulation cost coefficient, ΔP battery (k+i|k) is the k+i-step lithium battery frequency modulation power predicted in the k-step
[0078] C sc (k+i|k) is the k+i-step supercapacitor frequency modulation cost predicted in the k-th step:
[0079] C sc (k+i|k)=k sc ΔP sc (k+i|k) 2
[0080] Among them, k sc is the supercapacitor frequency modulation cost coefficient, ΔP sc (k+i|k) is the supercapacitor frequency modulation power predicted in the kth step (k+i)
[0081] The constraints include:
[0082] -P W ≤ΔP wind (k+i|k)≤P W
[0083] -P B ≤ΔP battery (k+i|k)≤P B
[0084] S batterymin ≤S battery (k+i|k)≤S batterymax
[0085] -P S ≤ΔP sc (k+i|k)≤P S
[0086] S scmin ≤S sc (k+i|k)≤S scmax
[0087] P(k+i|k)=K·Δf(k+i|k)
[0088] Where P W is the upper limit of wind power output, -P W is the lower limit of wind power output, ΔP wind (k+i|k) is the predicted value of wind power frequency regulation output at the current sampling time k for k+i steps, P B The upper limit of lithium battery output, -P B is the lower limit of lithium battery output, ΔP battery (k+i|k) is the predicted value of the k+i-step lithium battery frequency modulation output at the current sampling time k, P S is the upper limit of supercapacitor output, -P is the lower limit of supercapacitor output, ΔP sc (k+i|k) is the predicted value of the supercapacitor frequency modulation output at the current sampling time k for k+i steps,
[0089] S batterymin is the lower limit of SOC of lithium battery, S batterymax is the upper limit of SOC of lithium battery, S battery (k+i|k) is the predicted value of the lithium battery SOC for k+i steps at the current sampling time k, S scmin is the lower limit of supercapacitor SOC, S scmax is the upper limit of supercapacitor SOC, S sc (k+i|k) is the predicted value of the supercapacitor SOC for the k+i step at the current sampling moment k, P(k+i|k) is the predicted value of the frequency regulation output demand of the wind storage system for the k+i step at the current sampling moment k, K is the coefficient, Δf(k+i|k) is the predicted value of the system frequency deviation for the k+i step at the current sampling moment k.
[0090] According to the predicted system state and the set constraints and objective function, the operating conditions of different system load fluctuations, system frequency deviations, and energy storage system SOC are traversed offline for solution. However, the specific output of the wind turbine and energy storage in the optimal solution is not recorded. Instead, the output of the wind turbine, energy storage, and energy storage SOC are used to verify whether each inequality constraint works on the equality constraint, and then the optimal effective inequality constraint set is found. The system working modes are numbered according to the different inequality constraints that are satisfied, and the system load fluctuations, system frequency deviations, energy storage system SOC, and system working mode number under each working condition are recorded.
[0091] Specifically, by using traditional solution methods to offline traverse different system load fluctuations, system frequency deviations, and SOC states of the energy storage system, the optimal power instructions for fans, lithium batteries, and supercapacitors are obtained. The power instructions here are only used to combine the remaining state variables predicted by the prediction model to judge the inequality constraints one by one, determine whether each inequality constraint works in an equality constraint state or a relaxation state, and select a set of inequality constraints in an equality constraint state, which is the optimal effective inequality constraint set.
[0092] Step 3: Take the load fluctuation, system frequency deviation, and energy storage system SOC under various working conditions as input, and the system working mode number compiled according to the different inequality constraints satisfied as output, and train the DNN so that it can obtain the optimal effective inequality constraint set based on the current system load fluctuation, system frequency deviation, energy storage system SOC state and the predicted system load fluctuation, system frequency deviation, and energy storage system SOC state in the predicted time domain.
[0093] For convex optimization problems, when the global optimal point is outside the constraint range, it can be concluded that the optimal point in the feasible domain is on the equality constraint boundary of the inequality constraint. By finding the boundary conditions (that is, the effective inequality constraint set), the problem can be transformed into a quadratic programming problem with equality constraints and solved to obtain the optimal power instruction sequence.
[0094] Among them, the energy storage system SOC includes lithium battery SOC and supercapacitor SOC;
[0095] Step 4: The obtained control sequence including wind power frequency modulation power and energy storage frequency modulation power Δu(k), Δu(k+1), Δu(k+2)... only takes the first control instruction in the wind power frequency modulation power and energy storage frequency modulation power control sequence and issues it to the wind turbine and energy storage system for execution, and obtains the actual state of the system in the previous step at each step to perform feedback correction on the controller, so as to fix the frequency modulation power of the wind storage system within the feasible domain.
[0096] Example
[0097] The simulation system structure is as follows Figure 1As shown, the system includes wind turbines in the wind farm station, energy storage system in the wind farm station, power control system in the wind farm station, AC public power grid, thermal power synchronous units and AC loads. The thermal power plant G1 has a capacity of 470MW, and the wind farm G2 is a wind farm with a capacity of 150MW (30 * 5MW wind turbines). The wind turbine load reduction rate is set to 10%, and the wind condition is a constant wind speed of 10m / s. The model is selected as the NREL5MW model, with a rated wind speed of 11.4m / s. Under the set working conditions, the standby power is 10.12MW, the lithium battery system is configured as 10MW / 4.167MWh, and the supercapacitor is configured as 10MW / 0.4167MWh. In addition, there is a fluctuating AC load, the base load is 480MW, and the system load suddenly increases by 40MW at 0s.
[0098] Step 1: Comprehensively consider the impact of voltage and capacity on energy storage operation characteristics, and build an energy storage equivalent model suitable for wind-storage combined frequency regulation application scenarios:
[0099] P battery =u lb i lb
[0100] Where P battery is the energy storage output power, u lb =E lb -R in_lb ·i lb ,
[0101]
[0102] In the formula, E lb Indicates no-load voltage; E 0_lb represents constant voltage; K represents polarization voltage; Q lb Indicates capacity; i lb represents the current; A represents the amplitude in the exponential region; B represents the inverse of the time constant in the exponential region in constant current discharge;
[0103] R in_lb =R 1-lb ·N s_lb / N p_lb
[0104] Where N s_lb N is the number of energy storage series, p_lb is the number of energy storage in parallel, R 1-lb It is the single resistance value.
[0105] Step 2: Construct the system state equation group during the primary frequency regulation process of the wind-storage combined system:
[0106]
[0107] In the formula, Ggov (s) is the speed regulator transfer function, G gen (s) = G gen1 (s)G gen2 (s) is the turbine transfer function, G gen1 (s) is the non-reheat steam turbine transfer function, G gen2 (s) is the transfer function G of the main inlet steam chamber of the reheat steam turbine sc (s) is the supercapacitor transfer function, G battery (s) is the transfer function of lithium battery, G wind (s) is the wind turbine pitch control transfer function, P v is the output power of the speed regulator, P m is the turbine output power, ΔP g is the synchronous machine output power increment, ΔP sc is the supercapacitor output power increment, ΔP scref is the supercapacitor output power increment command value, ΔS sc is the supercapacitor SOC change, T is the control period, E sc is the supercapacitor capacity, ΔP battery is the output power increment of the lithium battery, ΔP batteryref is the lithium battery output power increment command value, ΔS battery is the change in SOC of the lithium battery, E battery is the capacity of lithium battery, ΔP wind is the wind turbine output power increment, ΔP windref is the incremental command value of wind turbine output power, Δf is the system frequency deviation, H is the system inertia time constant, D is the load regulation coefficient, P load is the system load fluctuation, 1 / R is the droop coefficient; P load For load fluctuation.
[0108] The system state equations are discretized, and the discrete system prediction model is obtained as follows:
[0109]
[0110] The k or k+1 in brackets after the variable indicates the variable value of the corresponding step. The k in brackets indicates the predicted value or actual value of the kth step. The k+1 in brackets indicates the predicted value of the k+1th step. v is the output power of the speed regulator, P m is the turbine output power, ΔP g is the synchronous machine output power increment, ΔP sc is the supercapacitor output power increment, ΔP scref is the supercapacitor output power increment command value, S sc is the supercapacitor SOC, T is the control period, Esc is the supercapacitor capacity, ΔP battery is the output power increment of the lithium battery, ΔP batteryref is the lithium battery output power increment command value, S battery is the lithium battery SOC, E battery is the capacity of lithium battery, ΔP wind is the wind turbine output power increment, ΔP windref is the incremental command value of wind turbine output power, Δf is the system frequency deviation, T G is the speed regulator time constant, 1 / R is the droop coefficient, T CH is the main steam inlet chamber time constant, T RH is the reheater time constant, T SC , T B , T WIND are the time constants of supercapacitor, lithium battery and wind turbine respectively, H is the system inertia time constant, D is the load regulation coefficient, P load is the system load fluctuation. Then, the discrete equations of the system can be used to predict the k+1 to k+N p The system status within the step.
[0111] Set the objective function to the frequency regulation cost of the wind-storage system in the forecast time domain
[0112]
[0113] C wind (k+i|k) is the fan frequency regulation cost of the k+i-th step predicted in the k-th step:
[0114] C wind (k+i|k)=k wind ΔP wind (k+i|k) 2
[0115] Among them, k wind is the wind power frequency regulation cost coefficient, ΔP wind (k+i|k) is the fan frequency modulation power of the k+i-th step predicted in the k-th step
[0116] C battery (k+i|k) is the k+i-step lithium battery frequency modulation cost predicted in the k-step:
[0117] C battery (k+i|k)=k battery ΔP battery (k+i|k) 2
[0118] Among them, k battery is the lithium battery frequency modulation cost coefficient, ΔP battery(k+i|k) is the k+i-step lithium battery frequency modulation power predicted in the k-step
[0119] C sc (k+i|k) is the k+i-step supercapacitor frequency modulation cost predicted in the k-th step:
[0120] C sc (k+i|k)=k sc ΔP sc (k+i|k) 2
[0121] Among them, k sc is the supercapacitor frequency modulation cost coefficient, ΔP sc (k+i|k) is the supercapacitor frequency modulation power predicted in the kth step (k+i)
[0122] The constraints include:
[0123] -P W ≤ΔP wind (k+i|k)≤P W
[0124] -P B ≤ΔP battery (k+i|k)≤P B
[0125] S batterymin ≤S battery (k+i|k)≤S batterymax
[0126] -P S ≤ΔP sc (k+i|k)≤P S
[0127] S scmin ≤S sc (k+i|k)≤S scmax
[0128] P(k+i|k)=K·Δf(k+i|k)
[0129] Where P W is the upper limit of wind power output, -P W is the lower limit of wind power output, ΔP wind (k+i|k) is the predicted value of wind power frequency regulation output at the current sampling time k for k+i steps, P B The upper limit of lithium battery output, -P B is the lower limit of lithium battery output, ΔP battery (k+i|k) is the predicted value of the k+i-step lithium battery frequency modulation output at the current sampling time k, P Sis the upper limit of supercapacitor output, -P is the lower limit of supercapacitor output, ΔP sc (k+i|k) is the predicted value of the supercapacitor frequency modulation output at the current sampling time k for k+i steps,
[0130] S batterymin is the lower limit of SOC of lithium battery, S batterymax is the upper limit of SOC of lithium battery, S battery (k+i|k) is the predicted value of the lithium battery SOC for k+i steps at the current sampling time k, S scmin is the lower limit of supercapacitor SOC, S scmax is the upper limit of supercapacitor SOC, S sc (k+i|k) is the predicted value of the supercapacitor SOC for the k+i step at the current sampling moment k, P(k+i|k) is the predicted value of the frequency regulation output demand of the wind storage system for the k+i step at the current sampling moment k, K is the coefficient, Δf(k+i|k) is the predicted value of the system frequency deviation for the k+i step at the current sampling moment k.
[0131] According to the predicted system state and the set constraints and objective function, the operating conditions of different system load fluctuations, system frequency deviations, and energy storage system SOC are traversed offline for solution. However, the specific output of the wind turbine and energy storage in the optimal solution is not recorded. Instead, the output of the wind turbine, energy storage, and energy storage SOC are used to verify whether each inequality constraint works on the equality constraint, and then the optimal effective inequality constraint set is found. The system working modes are numbered according to the different inequality constraints that are satisfied, and the system load fluctuations, system frequency deviations, energy storage system SOC, and system working mode number under each working condition are recorded.
[0132] Step 3: Take the system load fluctuation, system frequency deviation, and energy storage system SOC as input, and the mode number compiled according to the different inequality constraints satisfied as output, and train the DNN so that it can obtain the optimal effective inequality constraint set based on the current system load fluctuation, system frequency deviation, energy storage system SOC state and the predicted system load fluctuation, system frequency deviation, and energy storage system SOC state in the predicted time domain, and then map the optimal power instruction sequence.
[0133] Step 4: The obtained control sequence including wind power frequency modulation power and energy storage frequency modulation power Δu(k), Δu(k+1), Δu(k+2)... only takes the first control instruction in the wind power frequency modulation power and energy storage frequency modulation power control sequence and issues it to the wind turbine and energy storage system for execution, and obtains the actual state of the system in the previous step at each step to perform feedback correction on the controller, so as to fix the frequency modulation power of the wind storage system within the feasible domain.
[0134] Table 1 Specific parameter configuration in simulation verification
[0135]
[0136]
[0137] Finally, the traditional PI frequency modulation control method is set as the control group to verify the effectiveness of the strategy proposed in this invention, as shown in Table 1. The wind-storage combined frequency modulation power control designed in this invention should first ensure the system frequency stability and be able to support the system frequency when the load fluctuates. Therefore, the frequency stability of the system operation under different strategies is first compared, and the comparison results are shown in Table 1. Figure 2 As shown in the figure, it can be seen that the proposed strategy is superior to the control strategy in terms of the lowest frequency point and stabilization time, and can effectively and timely ensure the stable operation of the system. Afterwards, by comparing and analyzing the frequency modulation output of each part under the two strategies, it can be found that the strategy proposed by the present invention improves the shortcomings of the traditional PI strategy that it is difficult to fully mobilize the supercapacitor and lithium battery energy storage, and fully utilizes the characteristics of fast charging and discharging of energy storage, greatly improving the frequency support capability, such as Figure 3 As shown. Finally, the economic costs of system frequency regulation under the two strategies are compared, and the results are shown in Figure 4. It can be seen that after applying the strategy proposed in the present invention, the overall system frequency regulation cost decreases, and energy storage devices with different operating characteristics can be more reasonably applied. In summary, it can be explained that the application of the energy storage system power control strategy proposed in the present invention can achieve the improvement of the utilization of various parts of the wind storage system while ensuring the system frequency support capability, improve the system operation stability, reduce equipment operation losses, and improve the economy of the wind storage system application.
[0138] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art can still modify the technical solutions described in the aforementioned embodiments or replace some of the technical features therein by equivalents. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.
Claims
1. A wind-storage joint frequency modulation control method based on model predictive control and deep learning, characterized in that: include: Construct an energy storage equivalent model suitable for wind-storage combined frequency regulation application scenarios, specifically: P battery =in lb and lb Where P battery is the energy storage output power, u lb =E lb -R in_lb ·i lb , In the formula, E lb Indicates no-load voltage; E 0_lb represents constant voltage; K represents polarization voltage; Q lb Indicates capacity; i lb represents the current; A represents the amplitude in the exponential region; B represents the inverse of the time constant in the exponential region in constant current discharge; R in_lb =R 1-lb ·N s_lb / N p_lb Where N s_lb N is the number of energy storage series, p_lb is the number of energy storage in parallel, R 1-lb is the single resistance value; A discrete system prediction model for wind-storage combined system participating in frequency regulation is constructed to predict the system state from step k+1 to step k+Np. The corresponding constraints and objective functions are set. Different system load fluctuations, system frequency deviations, and energy storage system SOC conditions are traversed offline to obtain the optimal effective inequality constraint set. The system working modes are numbered according to the different inequality constraints that are satisfied. The specific process is as follows: Construct the system state equation group in the primary frequency regulation process of the wind-storage combined system: In the formula, G gov (s) is the speed regulator transfer function, G gen (s) = G gen1 (s)G gen2 (s) is the turbine transfer function, G gen1 (s) is the non-reheat steam turbine transfer function, G gen2 (s) is the transfer function G of the main inlet steam chamber of the reheat steam turbine sc (s) is the supercapacitor transfer function, G battery (s) is the transfer function of lithium battery, G wind (s) is the wind turbine pitch control transfer function, P v is the output power of the speed regulator, P m is the turbine output power, ΔP g is the synchronous machine output power increment, ΔP sc is the supercapacitor output power increment, ΔP scref is the supercapacitor output power increment command value, ΔS sc is the supercapacitor SOC change, T is the control period, E sc is the supercapacitor capacity, ΔP battery is the output power increment of the lithium battery, ΔP batteryref is the lithium battery output power increment command value, ΔS battery is the change in SOC of the lithium battery, E battery is the capacity of lithium battery, ΔP wind is the wind turbine output power increment, ΔP windref is the incremental command value of wind turbine output power, Δf is the system frequency deviation, H is the system inertia time constant, D is the load regulation coefficient, P load is the system load fluctuation, 1 / R is the droop coefficient; P load For load fluctuation; The system state equations are discretized, and the discrete system prediction model is obtained as follows: The k or k+1 in brackets after the variable indicates the variable value of the corresponding step. The k in brackets indicates the predicted value or actual value of the kth step. The k+1 in brackets indicates the predicted value of the k+1th step. v is the output power of the speed regulator, P m is the turbine output power, ΔP g is the synchronous machine output power increment, ΔP sc is the supercapacitor output power increment, ΔP scref is the supercapacitor output power increment command value, S sc is the supercapacitor SOC, T is the control period, E sc is the supercapacitor capacity, ΔP battery is the output power increment of the lithium battery, ΔP batteryref is the lithium battery output power increment command value, S battery is the lithium battery SOC, E battery is the capacity of lithium battery, ΔP wind is the wind turbine output power increment, ΔP windref is the incremental command value of wind turbine output power, Δf is the system frequency deviation, T G is the speed regulator time constant, 1 / R is the droop coefficient, T CH is the main steam inlet chamber time constant, T RH is the reheater time constant, T SC , T B , T WIND are the time constants of supercapacitor, lithium battery and wind turbine respectively, H is the system inertia time constant, D is the load regulation coefficient, P load is the system load wave; The objective function is set as: Where N p is the prediction time domain step size, C wind (k+i|k) is the fan frequency regulation cost of the k+i-th step predicted in the k-th step; C battery (k+i|k) is the k+i-step lithium battery frequency modulation cost predicted in the k-step; C sc (k+i|k) is the k+i-th step supercapacitor frequency modulation cost predicted in the k-th step; The fan frequency regulation cost of the k+i step predicted in the kth step is: C wind (k+i∣k)=k wind ·ΔP wind (k+i∣k) 2 Among them, k wind is the wind power frequency regulation cost coefficient, ΔP wind (k+i|k) is the fan frequency modulation power of the k+ith step predicted in the kth step; The specific cost of lithium battery frequency modulation in the k+i step predicted in the kth step is: C battery (k+i∣k)=k battery ·ΔP battery (k+i∣k) 2 Among them, k battery is the lithium battery frequency modulation cost coefficient, ΔP battery (k+i|k) is the k+i-step lithium battery frequency modulation power predicted in the k-step; The specific cost of supercapacitor frequency modulation in the k+i step predicted in the kth step is: C sc (k+i∣k)=k sc ·ΔP sc (k+i∣k) 2 Among them, k sc is the supercapacitor frequency modulation cost coefficient, ΔP sc (k+i|k) is the supercapacitor frequency modulation power of the k+ith step predicted in the kth step; The constraints set include: -P W ≤ΔP wind (k+i∣k)≤P W -P B ≤ΔP battery (k+i∣k)≤P B S batterymin ≤S battery (k+i∣k)≤S batterymax -P S ≤ΔP sc (k+i∣k)≤P S S scmin ≤S sc (k+i∣k)≤S scmax P(k+i|k)=K·Δf(k+i|k) Among them, P W is the upper limit of wind power output, -P W is the lower limit of wind power output, ΔP wind (k+i|k) is the predicted value of wind power frequency regulation output at step k+i, P B The upper limit of lithium battery output, -P B is the lower limit of lithium battery output, ΔP battery (k+i|k) is the predicted value of the k+i lithium battery frequency modulation output in the kth step, P S is the upper limit of supercapacitor output, -P s is the lower limit of supercapacitor output, ΔP sc (k+i|k) is the predicted value of the supercapacitor frequency modulation output for the k+i step in the kth step, S batterymin is the lower limit of SOC of lithium battery, S batterymax is the upper limit of SOC of lithium battery, S battery (k+i|k) is the predicted value of the lithium battery SOC at step k+i, S scmin is the lower limit of supercapacitor SOC, S scmax is the upper limit of supercapacitor SOC, S sc (k+i|k) is the predicted value of the supercapacitor SOC at step k+i, P(k+i|k) is the predicted value of the frequency regulation output demand of the wind storage system at step k+i, K is the coefficient, Δf(k+i|k) is the predicted value of the system frequency deviation at step k+i; With system load fluctuation, system frequency deviation, and energy storage system SOC as input and system working mode number as output, DNN is trained so that it can obtain the current optimal effective inequality constraint set and map the optimal power instruction sequence according to the current system load fluctuation, system frequency deviation, energy storage system SOC state and the predicted system load fluctuation, system frequency deviation, and energy storage system SOC state in the predicted time domain; The first one in the optimal power instruction sequence is selected as the wind-storage system power instruction and issued to the wind-storage system.
