Hybrid energy storage system output optimization method and system based on dynamic empowerment model predictive control

Abstract

The invention relates to a hybrid energy storage system output optimization method based on dynamic empowerment model predictive control, which comprises the following steps of: 1, establishing a combined model of a photovoltaic-hybrid energy storage system, the prediction values comprise photovoltaic grid-connected power PGrid and an energy state SHESS of an energy storage system, energy storage output power PHESS, photovoltaic power prediction quantity PPV, and prediction values PGridd (k + 1) and SHESS (k + 1) of the grid-connected power and the energy state; step 2, designing a control algorithm of multi-objective model prediction, and designing a multi-objective optimization function through output of a joint model prediction system; and step 3, designing a dynamic weighting method of a weight coefficient, introducing a Harris eagle algorithm, and solving a multi-objective function weight coefficient in real time. According to the method, actual engineering application is considered, the weight coefficient of each control sub-target is dynamically adjusted, the problem that the charging and discharging states of the energy storage system are frequently switched due to the fact that a traditional algorithm is prone to overcompensation is solved, adaptability is enhanced, and the method has a better stabilizing effect on power fluctuation changing in real time.

Description

Technical Field

[0001] This invention belongs to the field of photovoltaic energy storage technology, and relates to a method and system for optimizing the output of a hybrid energy storage system based on dynamic weighted model predictive control. Background Technology

[0002] The increasing penetration of renewable energy sources such as photovoltaics in the power system has led to a rise in the proportion of clean energy, but it has also caused serious power fluctuation problems. Because photovoltaic output is significantly affected by meteorological factors and is unpredictable and prone to sudden changes, if it is directly connected to the grid without regulation, it can easily cause problems such as abnormal voltage frequency and bus voltage disturbances, thereby affecting the safety and stability of grid operation.

[0003] Therefore, energy storage systems, as regulating devices for photovoltaic output, are widely used to suppress power fluctuations. However, traditional control methods such as low-pass filtering (LPF) and moving average (MA) algorithms are widely used in engineering, but their control strategies are relatively passive, processing only the signal itself and failing to consider both the energy state of the energy storage system and its operating efficiency. Furthermore, these methods have poor adaptability to varying fluctuation amplitudes and changing weather conditions, easily leading to problems such as frequent charging and discharging of the energy storage system, overcompensation, and reduced lifespan. In recent years, model predictive control (MPC) has gradually gained attention due to its excellent ability to handle multi-objective optimization problems. By introducing a state prediction mechanism, MPC can balance photovoltaic fluctuation mitigation and energy storage system state regulation; however, these two control objectives often conflict, requiring reasonable setting of weighting coefficients.

[0004] Currently, there are three main problems:

[0005] (1) The problem that traditional filtering and averaging algorithms cannot actively regulate the energy storage state:

[0006] Traditional filtering and averaging algorithms have significant limitations in controlling the output of hybrid energy storage systems. They primarily rely on passive adjustments based on photovoltaic power fluctuations, failing to sense or intervene in the internal state of the energy storage system. This passive control approach easily leads to overcharging, over-discharging, overcompensation, or undercompensation during actual operation, resulting in frequent charge / discharge switching, energy accumulation imbalances, and other problems that severely impact system stability and the lifespan of energy storage equipment.

[0007] (2) The problem that traditional low-pass filtering algorithms cannot effectively smooth out fluctuations in real time:

[0008] Traditional low-pass filtering algorithms suffer from poor real-time performance and delayed response when suppressing photovoltaic (PV) power fluctuations. Their control strategies rely on historical data to tune filter parameters, lacking sensitivity to rapid changes in PV output. This results in an inability to respond effectively in a timely manner when PV power fluctuates drastically or abruptly, easily leading to fluctuations exceeding limits. Furthermore, low-pass filtering control results typically exhibit significant delays, affecting the continuity and stability of grid-connected power. In complex and variable lighting environments, this fixed-parameter, single-strategy filtering method has poor adaptability and is difficult to meet the high requirements of grid-connected power fluctuation control in applications such as ports.

[0009] (3) Fixed weight control lacks adaptability to operating conditions:

[0010] Fixed-weight control strategies cannot flexibly adjust the importance of each control objective according to actual operating conditions in multi-objective optimization, resulting in a lack of adaptability of the system under different photovoltaic fluctuation intensities or energy storage states. For example, maintaining the original weight configuration when photovoltaic fluctuations are severe may cause the energy storage system to over-respond, exacerbating the charging and discharging burden; conversely, if the weights are not adjusted when the energy storage state is nearing its limit, it is easy to cause energy overflow or control failure. Due to the inability to dynamically balance the contradiction between grid-connected power stability and energy storage health, fixed-weight control has poor adaptability in real-world complex scenarios and is difficult to achieve overall optimization control effects. Summary of the Invention

[0011] This invention addresses the problems of existing technologies by proposing a method and system for optimizing the output of a hybrid energy storage system based on dynamic weighted model predictive control. The hybrid energy storage system combines flywheel energy storage and lithium battery energy storage. This invention is achieved through the following technical solutions:

[0012] A method for optimizing the output of a hybrid energy storage system based on dynamic weighted model predictive control includes the following steps:

[0013] Step 1: Establish a joint model of the photovoltaic-hybrid energy storage system, which includes the grid-connected photovoltaic power P. Grid Energy State S of Energy Storage System HESS Energy storage output power P HESS Photovoltaic power forecast P PV Predicted values ​​of grid-connected power and energy state P Grid (k+1) and S HESS (k+1);

[0014] Step 2: Design a control algorithm for multi-objective model prediction. By jointly predicting the system output through the model, design a multi-objective optimization function and introduce weight coefficients to combine the multiple objectives into a unified optimization cost function.

[0015] Step 3: Design a dynamic weighting method for the weight coefficients, introduce the Harris Eagle algorithm, solve the weight coefficients of the multi-objective function in real time, and obtain and execute the optimal energy storage output sequence based on the rolling optimization principle.

[0016] Furthermore, in step one, the specific method for establishing the joint model of the photovoltaic-hybrid energy storage system is as follows:

[0017] According to the principle of energy conservation, at any time t, the system satisfies the following power balance relationship:

[0018] P Grid (t+1)=P PV (t)+P HESS (t)

[0019] In the formula: P PV (t) represents the actual output power of the port photovoltaic power plant at time t, P HESS (t) represents the total output power of the hybrid energy storage system. HESS (t) > 0 indicates energy storage and discharge, < 0 indicates charging; P Grid (t+1) represents the grid-connected photovoltaic power at the next moment.

[0020] Introducing state variable S HESS (t) represents the percentage of the energy storage system's current remaining capacity relative to its rated capacity, and its evolution follows the following law:

[0021]

[0022] Among them, E N S represents the rated energy capacity of the energy storage system, Δt represents the time step of the control cycle, and S represents the rated energy capacity of the energy storage system. HESS (t) is typically limited to [0.1, 0.9] or [0.2, 0.8] (normalized to 10%–90% or 20%–80%), with the specific range determined based on battery type and lifespan requirements. HESS,t+1 This is the state quantity for the next moment.

[0023] In summary, the system has two core state variables: x1(t) = P Grid Photovoltaic grid-connected power, x2(t) = S HESS (t), the energy state of the energy storage system; a control variable: u(t) = P HESS (t), energy storage output power; a disturbance input: d(k) = P PV (t), the photovoltaic power prediction; two outputs: P Grid (k+1) and S HESS (k+1), the predicted values ​​of grid-connected power and energy state;

[0024] Convert to a discrete state-space model:

[0025]

[0026] Furthermore, in step two, the method for multi-objective model prediction is as follows:

[0027] Using the discrete state-space model of the photovoltaic-hybrid energy storage system as the prediction model, we obtain:

[0028]

[0029] In the formula, k is the sampling time, x(k) is the system state variable, y(k) is the system output variable, u(k) is the control variable, d(k) is the disturbance variable, A is the state matrix, B is the control matrix, C is the output matrix, and D is the disturbance matrix.

[0030] Based on the coupling relationship between variables, equation (4) is iteratively calculated, and the predicted system output y(k+i) at time k+i is:

[0031]

[0032] In the formula, y(k+i) is the predicted value of the system output at time k+i by the model prediction stage at time k, that is, the grid-connected photovoltaic power P at time k+i. Grid (k+i) and the energy state S of the hybrid energy storage system HESS (k+i).

[0033] Furthermore, in step two, the multi-objective optimization function is designed as follows:

[0034] With the primary objective of smoothing grid-connected power and minimizing the deviation between the grid-connected power and the ideal target curve, the objective function J1 is established as follows:

[0035]

[0036] In the formula, P Grid (k+i) represents the grid-connected power of the port's photovoltaic system at time k+i, N is the prediction time domain length of the model predictive control, and k is the current control time.

[0037] To minimize the output power of the energy storage system, the objective function J2 is established as follows:

[0038]

[0039] To maintain the energy state near the median and preserve the system's ability to continuously regulate, the objective function J3 is established as follows:

[0040]

[0041] Furthermore, in step two, the method of introducing weighting coefficients to combine multiple objectives into a unified optimization cost function is as follows:

[0042] We introduce weighting coefficients α(k), β(k), and λ(k) to characterize the relative importance of each control sub-objective at each time step. We then use a weighted summation method to combine the three optimization objectives into a unified optimization cost function.

[0043]

[0044] These correspond to the output power constraints of the hybrid energy storage system, the energy state constraints of the hybrid energy storage system, and the power fluctuation constraints of the photovoltaic grid connection, respectively:

[0045]

[0046] In the formula P HESS_max S represents the maximum output power of the hybrid energy storage system. min and S max These are the lower and upper limits of the State of Energy (SOE) for a hybrid energy storage system, respectively; maxP Grid_1min (k) and minP Grid_1min (k) represents the maximum and minimum values ​​of the photovoltaic grid-connected power at the 1-minute scale, respectively, and γ is the grid-connected volatility limit.

[0047] Furthermore, in step three, a dynamic weighting method for the weight coefficients is designed, introducing the Harris Eagle algorithm, which is divided into three parts: a global search phase, a transition phase, and a local development phase. The global search and transition phase methods are as follows:

[0048] (1) Global search phase

[0049] The initial iterative process for optimizing the weight coefficient parameters involves randomly generating a series of weight coefficient combinations and continuously selecting the relatively optimal solution from them.

[0050] (2) Transition Phase

[0051] The concept of escape energy E is introduced to facilitate the transition from the global search phase to the local exploitation phase. The formula for calculating escape energy E is as follows:

[0052]

[0053] In the formula, E0 is the initial escape energy of the prey, which is a random number in the range of [-1,1]; T is the maximum number of iterations; when |E| < 1, the algorithm enters the subsequent local development stage, otherwise it remains in the global search stage.

[0054] Furthermore, the partial development phase approach is as follows:

[0055] When |E|<1, four siege methods will be used to perform a local search based on |E| and a random number r in the range [0,1].

[0056] When |E|≥0.5 and r≥0.5, the position update formula is as follows:

[0057] X i (t+1)=ΔX i (t)-E|J i X p (t)-X i (t)|

[0058] ΔX i (t)=X p (t)-X i (t)

[0059] In the formula, t represents the current iteration number; X i (t) represents the position of the optimization objective at the t-th iteration, i.e., the weight coefficient combination (α(t), β(t), λ(t)) corresponding to this iteration. i X is a random number in the range [0,2]. p (t) represents the position of the prey in the t-th iteration, and ΔX represents the optimal weight coefficient up to the t-th iteration; i (t) represents the distance between the actual value and the target value at the t-th iteration, and represents the difference between the weight coefficient obtained in this iteration and the current optimal weight coefficient.

[0060] When |E| < 0.5 and r ≥ 0.5, the position update formula is as follows:

[0061] X i (t+1)=X p (t)-E|ΔX i (t)|.

[0062] When |E|≥0.5 and r<0.5, the position update formula is as follows:

[0063]

[0064] Y i (t)=X p (t)-E|J i X p (t)-X i (t)|

[0065] Z i (t)=Y i (t)+S i ·LF(D)

[0066] In the formula, F(·) is the objective function used to sort the frontier values; D is the dimension of the problem to be solved; S represents a D-dimensional random row vector, where LF is the Levy function, and its expression is:

[0067]

[0068] Where u and v are both random numbers, β is a constant, and has

[0069]

[0070] When |E| < 0.5 and r < 0.5, the position update formula is as follows:

[0071]

[0072] Y i (t)=X p (t)-E|J i X p (t)+X m (t)|

[0073] Z i (t)=Y i (t)+S i ·LF(D)

[0074] Finally, the optimal solution set of the control quantity sequence is obtained in each control cycle, and the first item of the solution set is executed to achieve rolling optimization.

[0075] The present invention also relates to a hybrid energy storage system based on dynamic weighted model predictive control, comprising a computer module, wherein the computer module utilizes a method for optimizing the output of the hybrid energy storage system based on dynamic weighted model predictive control.

[0076] Beneficial effects

[0077] This invention considers the randomness of photovoltaic fluctuations and the dynamic changes in the operating state of hybrid energy storage systems. It proposes a hybrid energy storage output optimization control method based on multi-objective model predictive control and a dynamic weighting mechanism to improve the grid connection performance of photovoltaic systems and the operational reliability of energy storage systems. The main contents include:

[0078] (1) Based on the analysis of the power output fluctuation characteristics of the photovoltaic system and the energy change law of the hybrid energy storage system, a joint discrete state space model of the photovoltaic-hybrid energy storage system is established.

[0079] Unlike the traditional approach of simplifying the relationship between photovoltaics and energy storage by directly using empirical formulas, this invention establishes a joint discrete state-space model for a photovoltaic-hybrid energy storage system. This model can accurately reflect the temporal changes in photovoltaic output and energy storage status, providing a foundation for multi-objective model predictive control.

[0080] (2) To replace the traditional filtering and averaging algorithms, a control algorithm based on multi-objective model prediction is constructed.

[0081] To address the issue that traditional filtering and averaging algorithms cannot adjust the energy storage state and suffer from severe overcompensation, which seriously affects control performance and stability, this invention proposes a control algorithm based on multi-objective model prediction. It constructs an optimization function with three control objectives: minimizing grid connection deviation, minimizing energy storage output, and optimizing energy storage state.

[0082] (3) Dynamically assigning weights to the three control targets can maintain a good energy storage state while stabilizing and suppressing photovoltaic power fluctuations, effectively solving the problem of overcompensation that traditional algorithms are prone to.

[0083] This invention takes into account practical engineering applications and dynamically adjusts the weight coefficients of each control sub-objective, so that while meeting grid connection requirements, it also takes into account the active regulation of the state of the hybrid energy storage system. This not only solves the problem of frequent switching of charging and discharging states of the energy storage system caused by overcompensation in traditional algorithms, but also enhances adaptability and has a better smoothing effect in the face of real-time power fluctuations. Attached Figure Description

[0084] Figure 1 This is a schematic diagram illustrating the principle of the hybrid energy storage system of the present invention for suppressing photovoltaic fluctuations.

[0085] Figure 2 This is a flowchart of the optimization process for the control sub-objective weight coefficients based on the Harris Eagle algorithm of this invention;

[0086] Figure 3 This is a schematic diagram of the original photovoltaic output power of a port according to the present invention;

[0087] Figure 4 This is a schematic diagram comparing the 1-minute fluctuation of photovoltaic grid-connected power under typical daily conditions according to the present invention;

[0088] Figure 5 This is a schematic diagram comparing the 1-minute fluctuation of photovoltaic grid-connected power under extreme daily conditions according to the present invention;

[0089] Figure 6 This is a schematic diagram comparing the output power of the hybrid energy storage system under the two algorithms of this invention;

[0090] Figure 7 This is a schematic diagram comparing the throughput energy of the hybrid energy storage system under the two algorithms of this invention;

[0091] Figure 8 This is a schematic diagram comparing the 1-minute fluctuation of grid-connected power before and after the dynamic weighted MPC control of this invention.

[0092] Figure 9 This is a schematic diagram comparing the cumulative probability distribution of the 1-minute fluctuation of the grid-connected power before and after the control of this invention;

[0093] Figure 10 This is a schematic diagram illustrating the state regulation effect of the hybrid energy storage system under dynamic weighted MPC control according to the present invention.

[0094] Figure 11 This is a schematic diagram comparing the cumulative probability distribution of the output power of the hybrid energy storage system under different methods of the present invention;

[0095] Figure 12 This is a schematic diagram comparing the 1-minute fluctuation of grid-connected power before and after dynamic weighted MPC control when the grid-connected volatility limit is 5% according to the present invention.

[0096] Figure 13 This is a schematic diagram illustrating the adjustment effect of the dynamic weighted MPC method for controlling a hybrid energy storage system when the grid connection volatility limit is 5%.

[0097] Figure 14 This is a schematic diagram of the 1-minute fluctuation of grid-connected power before and after control by the extreme day-time dynamic weighted MPC method of the present invention;

[0098] Figure 15 This is a schematic diagram illustrating the regulation effect of a hybrid energy storage system under the control of the extreme day-time dynamic weighting MPC method of the present invention. Detailed Implementation

[0099] The following is in conjunction with the appendix Figures 1 to 15 The present invention further details the method and system for optimizing the output of a hybrid energy storage system based on dynamic weighted model predictive control.

[0100] A method for optimizing the output of a hybrid energy storage system based on dynamic weighted model predictive control includes the following steps:

[0101] Step 1: Establish a joint model of the photovoltaic-hybrid energy storage system, which includes the grid-connected photovoltaic power P. Grid Energy State S of Energy Storage System HESS Energy storage output power P HESS Photovoltaic power forecast P PV Predicted values ​​of grid-connected power and energy state P Grid (k+1) and S HESS (k+1);

[0102] According to the principle of energy conservation, at any time t, the system satisfies the following power balance relationship:

[0103] P Grid (t+1)=P PV (t)+P HESS (t)

[0104] In the formula: P PV (t) represents the actual output power of the port photovoltaic power plant at time t, P HESS (t) represents the total output power of the hybrid energy storage system. HESS (t) > 0 indicates energy storage and discharge, < 0 indicates charging; P Grid (t+1) represents the grid-connected photovoltaic power at the next moment.

[0105] Introducing state variable S HESS (t) represents the percentage of the energy storage system's current remaining capacity relative to its rated capacity, and its evolution follows the following law:

[0106]

[0107] Among them, E N S represents the rated energy capacity of the energy storage system, Δt represents the time step of the control cycle, and S represents the rated energy capacity of the energy storage system. HESS (t) is usually limited to a certain range; S HESS,t+1 This is the state quantity for the next moment.

[0108] Convert to a discrete state-space model:

[0109]

[0110] Step 2: Design a control algorithm for multi-objective model prediction. By jointly predicting the system output through the model, design a multi-objective optimization function and introduce weight coefficients to combine the multiple objectives into a unified optimization cost function.

[0111] The method for multi-objective model prediction is as follows:

[0112] Using the discrete state-space model of the photovoltaic-hybrid energy storage system as the prediction model, we obtain:

[0113]

[0114] In the formula, k is the sampling time, x(k) is the system state variable, y(k) is the system output variable, u(k) is the control variable, d(k) is the disturbance variable, A is the state matrix, B is the control matrix, C is the output matrix, and D is the disturbance matrix.

[0115] Based on the coupling relationship between variables, equation (4) is iteratively calculated, and the predicted system output y(k+i) at time k+i is:

[0116]

[0117] In the formula, y(k+i) is the predicted value of the system output at time k+i by the model prediction stage at time k, that is, the grid-connected photovoltaic power P at time k+i. Grid(k+i) and the energy state S of the hybrid energy storage system HESS (k+i).

[0118] The multi-objective optimization function is designed as follows:

[0119] With the primary objective of smoothing grid-connected power and minimizing the deviation between the grid-connected power and the ideal target curve, the objective function J1 is established as follows:

[0120]

[0121] In the formula, P Grid (k+i) represents the grid-connected power of the port's photovoltaic system at time k+i, N is the prediction time domain length of the model predictive control, and k is the current control time.

[0122] To minimize the output power of the energy storage system, the objective function J2 is established as follows:

[0123]

[0124] To maintain the energy state near the median and preserve the system's ability to continuously regulate, the objective function J3 is established as follows:

[0125]

[0126] The method of combining multiple objectives into a unified optimization cost function by incorporating weighting coefficients is as follows:

[0127] We introduce weighting coefficients α(k), β(k), and λ(k) to characterize the relative importance of each control sub-objective at each time step. We then use a weighted summation method to combine the three optimization objectives into a unified optimization cost function.

[0128]

[0129] These correspond to the output power constraints of the hybrid energy storage system, the energy state constraints of the hybrid energy storage system, and the power fluctuation constraints of the photovoltaic grid connection, respectively:

[0130]

[0131] In the formula P HESS_max S represents the maximum output power of the hybrid energy storage system. min and S max These are the lower and upper limits of the State of Energy (SOE) for a hybrid energy storage system, respectively; maxP Grid_1min (k) and minP Grid_1min (k) represents the maximum and minimum values ​​of the photovoltaic grid-connected power at the 1-minute scale, respectively, and γ is the grid-connected volatility limit.

[0132] Step 3: Design a dynamic weighting method for the weight coefficients, introduce the Harris Eagle algorithm, solve the weight coefficients of the multi-objective function in real time, and obtain and execute the optimal energy storage output sequence based on the rolling optimization principle.

[0133] A dynamic weighting method for the weight coefficients is designed, introducing the Harris Eagle algorithm, which is divided into three parts: a global search phase, a transition phase, and a local development phase. The global search and transition phase methods are as follows:

[0134] (1) Global search phase

[0135] The initial iterative process for optimizing the weight coefficient parameters involves randomly generating a series of weight coefficient combinations and continuously selecting the relatively optimal solution from them.

[0136] (2) Transition Phase

[0137] The concept of escape energy E is introduced to facilitate the transition from the global search phase to the local exploitation phase. The formula for calculating escape energy E is as follows:

[0138]

[0139] In the formula, E0 is the initial escape energy of the prey, which is a random number in the range of [-1,1]; T is the maximum number of iterations; when |E| < 1, the algorithm enters the subsequent local development stage, otherwise it remains in the global search stage.

[0140] Furthermore, the partial development phase approach is as follows:

[0141] When |E|<1, four siege methods will be used to perform a local search based on |E| and a random number r in the range [0,1].

[0142] When |E|≥0.5 and r≥0.5, the position update formula is as follows:

[0143] X i (t+1)=ΔX i (t)-E|J i X p (t)-X i (t)|

[0144] ΔX i (t)=X p (t)-X i (t)

[0145] In the formula, t represents the current iteration number; X i (t) represents the position of the optimization objective at the t-th iteration, i.e., the weight coefficient combination (α(t), β(t), λ(t)) corresponding to this iteration. iThe random number is in the range [0,2]; Xp(t) is the position of the prey in the t-th iteration, representing the optimal weight coefficient up to the t-th iteration; ΔX i (t) represents the distance between the actual value and the target value at the t-th iteration, and represents the difference between the weight coefficient obtained in this iteration and the current optimal weight coefficient;

[0146] When |E| < 0.5 and r ≥ 0.5, the position update formula is as follows:

[0147] X i (t+1)=X p (t)-E|ΔX i (t)|.

[0148] When |E|≥0.5 and r<0.5, the position update formula is as follows:

[0149]

[0150] Y i (t)=X p (t)-E|J i X p (t)-X i (t)|

[0151] Z i (t)=Y i (t)+S i ·LF(D)

[0152] In the formula, F(·) is the objective function used to sort the frontier values; D is the dimension of the problem to be solved; S represents a D-dimensional random row vector, where LF is the Levy function, and its expression is:

[0153]

[0154] Where u and v are both random numbers, β is a constant, and has

[0155]

[0156] When |E| < 0.5 and r < 0.5, the position update formula is as follows:

[0157]

[0158] Y i (t)=X p (t)-E|J i X p (t)+X m (t)|

[0159] Z i (t)=Yi (t)+S i ·LF(D)

[0160] Finally, the optimal solution set of the control quantity sequence is obtained in each control cycle, and the first item of the solution set is executed to achieve rolling optimization.

[0161] Example

[0162] 1. Joint Discrete State-Space Model of Photovoltaic-Hybrid Energy Storage System

[0163] To suppress photovoltaic fluctuations while controlling the state of the energy storage system, the key state parameter of the photovoltaic-hybrid energy storage system joint model is the grid-connected photovoltaic power P. Grid And the State of Energy (SOE) of hybrid energy systems.

[0164] The principle of hybrid energy storage systems in suppressing port photovoltaic fluctuations is as follows: Figure 1 As shown.

[0165] To suppress photovoltaic fluctuations, even if the grid-connected photovoltaic power P at the port... Grid To meet the volatility requirements, precise control of the energy storage system's output power P is necessary. HESS Regarding the actual output power P of the photovoltaic system PV Compensation is performed. According to the principle of energy conservation, at any time t, the system satisfies the following power balance relationship:

[0166] P Grid (t+1)=P PV (t)+P HESS (t) (1)

[0167] In the formula, P PV (t) represents the actual output power of the port photovoltaic power plant at time t, P HESS (t) represents the total output power of the hybrid energy storage system. HESS (t) > 0 indicates energy storage and discharge, < 0 indicates charging. P Grid (t+1) represents the grid-connected photovoltaic power at the next moment.

[0168] In the process of power compensation, hybrid energy storage systems essentially transfer and regulate energy through charging and discharging processes. However, simply describing the instantaneous change in output power cannot reflect the overall operating state of the energy storage system. Therefore, a second state parameter, SOE, is introduced.

[0169] The value of SOE is expressed by the state variable S. HESS (t) represents the percentage of the energy storage system's current remaining capacity relative to its rated capacity, reflecting the power output capability of the hybrid energy storage system and its potential to suppress photovoltaic fluctuations. HESS (t) changes over time, evolving

[0170] The law of transformation satisfies:

[0171]

[0172] Among them, E N S represents the rated energy capacity of the energy storage system, Δt represents the time step of the control cycle, and S represents the rated energy capacity of the energy storage system. HESS,t Typically limited to [0.1, 0.9] or [0.2, 0.8] (normalized to 10%–90% or 20%–80%), the specific range is set according to the battery type and lifespan requirements. HESS,t+1 This is the state quantity for the next moment.

[0173] Considering the discrete sampling and digital implementation of the controller in practical engineering applications, it is necessary to convert the system description into a discrete state-space model. Based on the above variable definitions, the following state-space model is constructed:

[0174]

[0175] This is the discrete state-space model of the photovoltaic-hybrid energy storage system, and also the prediction model for the subsequent multi-objective model predictive control algorithm, providing input for obtaining the optimal output power of the hybrid energy storage system.

[0176] 2. Control Algorithm Based on Multi-Objective Model Prediction

[0177] Using the discrete state-space model of the photovoltaic-hybrid energy storage system as the prediction model, we can obtain:

[0178]

[0179] In the formula, k is the sampling time, x(k) is the system state variable, y(k) is the system output variable, u(k) is the control variable, d(k) is the disturbance variable, A is the state matrix, B is the control matrix, C is the output matrix, and D is the disturbance matrix.

[0180] Further iterative calculation of equation (4) based on the coupling relationship between variables yields the predicted system output y(k+i) at time k+i:

[0181]

[0182] In the formula, y(k+i) is the predicted value of the system output at time k+i by the model prediction stage at time k, that is, the grid-connected photovoltaic power P at time k+i. Grid (k+i) and the energy state S of the hybrid energy storage system HESS (k+i).

[0183] Next, three optimization objectives are designed as follows:

[0184] (1) Minimum grid-connected power deviation

[0185] Photovoltaic power exhibits significant randomness and uncertainty at different times. Without adjustment, this will lead to fluctuations in grid-connected power P. Grid Significant fluctuations severely impact the safe operation of the port power grid. Therefore, the primary objective of MPC (Mean Power Regulation) should be to smooth out grid-connected power, specifically by minimizing the deviation between the grid-connected power and the ideal target curve. The objective function J1 is established as follows:

[0186]

[0187] In the formula, P Grid (k+i) represents the grid-connected power of the port's photovoltaic system at time k+i, N is the prediction time domain length of the model predictive control, and k is the current control time.

[0188] (2) Energy storage output is the lowest

[0189] To extend the lifespan of the energy storage system and reduce the number of battery charge-discharge cycles and energy throughput, its output power should be controlled to be as small as possible. The objective function J2 is established as follows:

[0190]

[0191] (3) The energy storage state is maintained at the ideal median value.

[0192] During long-term operation, maintaining the energy storage state of energy (SOE) around 0.5 (the median value) maximizes the charge / discharge margin of the hybrid energy storage system, maximizes its ability to mitigate photovoltaic fluctuations, and minimizes its lifespan loss. Therefore, maintaining the SOE around the median value is crucial for ensuring the system's continuous regulation capability. The objective function J3 is established as follows:

[0193]

[0194] There are conflicts among the three optimization objectives mentioned above, and in practical engineering, excessive suppression of well network power is not pursued. Therefore, weighting coefficients α(k), β(k), and λ(k) are introduced to characterize the relative importance of each control sub-objective at each time step. A weighted summation method is used to combine the three optimization objectives into a unified optimization cost function, which is continuously adjusted according to photovoltaic fluctuations and energy storage status, thereby achieving overall optimization rather than local optimization. Finally, the form of the MPC control problem can be reduced to:

[0195]

[0196]

[0197] This includes multiple constraints on energy storage output, power status, and grid connection volatility to ensure the practical feasibility of the control strategy and the safe operation of the system.

[0198] 3. Dynamic weighting of three control objectives

[0199] In multi-objective model predictive control strategies, the importance of different control objectives changes continuously with the system's operating state and external disturbances. To address this, this invention introduces a dynamic weighting mechanism to continuously adjust the weight coefficients to cope with the complex stochastic changes in the port. Specifically, it can be divided into three parts: a global search phase, a transition phase, and a local development phase.

[0200] (1) Global search phase

[0201] The initial iterative process of optimizing the weight coefficient parameters involves randomly generating a series of weight coefficient combinations and continuously selecting the relatively optimal solution from them, laying the foundation for subsequent local development.

[0202] (2) Transition Phase

[0203] The concept of escape energy E is introduced to facilitate the transition from the global search phase to the local exploitation phase. The formula for calculating escape energy E is as follows:

[0204]

[0205] In the formula, E0 is the initial escape energy of the prey, which is a random number in the range of [-1,1]; T is the maximum number of iterations; when |E| < 1, the algorithm enters the subsequent local development stage, otherwise it remains in the global search stage.

[0206] (3) Partial Development Phase

[0207] When |E|<1, four siege methods will be used to conduct a local search based on |E| and a random number r in the range [0,1].

[0208] 1) When |E|≥0.5 and r≥0.5, the position update formula is as follows:

[0209] X i (t+1)=ΔX i (t)-E|J i X p (t)-X i (t)| (12)

[0210] ΔX i (t)=X p (t)-X i (t) (13)

[0211] In the formula, t represents the current iteration number; X i (t) represents the position of the optimization objective at the t-th iteration, i.e., the weight coefficient combination (α(t), β(t), λ(t)) corresponding to this iteration.i A random number in the range [0,2]; ΔX i (t) represents the distance between the actual value and the target value at the t-th iteration, and represents the difference between the weight coefficient obtained in this iteration and the current optimal weight coefficient.

[0212] 2) When |E| < 0.5 and r ≥ 0.5, the position update formula is as follows:

[0213] X i (t+1)=X p (t)-E|ΔX i (t)| (14)

[0214] 3) When |E|≥0.5 and r<0.5, the position update formula is as follows:

[0215]

[0216] Y i (t)=X p (t)-E|J i X p (t)-X i (t)| (16)

[0217] Z i (t)=Y i (t)+S i ·LF(D) (17)

[0218] In the formula, F(·) is the objective function used to sort the frontier values; D is the dimension of the problem to be solved; S represents a D-dimensional random row vector, where LF is the Levy function, and its expression is:

[0219]

[0220] Where u and v are both random numbers, β is a constant, and has

[0221]

[0222] 4) When |E| < 0.5 and r < 0.5, the position update formula is as follows:

[0223]

[0224] Y i (t)=X p (t)-E|J i X p (t)+X m (t)| (21)

[0225] Z i (t)=Yi (t)+S i ·LF(D) (22)

[0226] The optimization process for the weight coefficients of the control sub-objectives of a hybrid energy storage system based on dynamic weighting is expressed as follows: Figure 2 As shown in the figure.

[0227] Effect verification

[0228] This invention utilizes raw photovoltaic data from a 4MW photovoltaic power plant at a port to verify the effectiveness of the proposed hybrid energy storage system output optimization control strategy. Key parameters of the energy storage unit are shown in Table 1. Output power data from the port's photovoltaic power plant was collected for 90 consecutive days, with a sampling period from 6:00 AM to 6:00 PM (12 hours) and a sampling interval of 1 minute. Data showing the most representative fluctuations was selected as typical daily data, while data showing the most severe fluctuations was selected as extreme daily data, as shown below. Figure 3 As shown.

[0229] Table 1

[0230]

[0231] Since the required capacity of a hybrid energy storage system is closely related to the effectiveness of its output control, this invention focuses solely on the optimization and control of the hybrid energy storage system's output. Therefore, the capacity of the hybrid energy storage system is set to be sufficient.

[0232] 1. Control effects of traditional filtering algorithms and averaging algorithms

[0233] The 1-minute fluctuation limit of photovoltaic grid-connected power is set to 10%, i.e., no more than 0.4 MW / min. The filtering parameter 'a' of the low-pass filter (LPF) algorithm is set to 0.6. The length L of the data window of the moving average (MA) algorithm is initially 3 and can not exceed 9.

[0234] Low-pass filtering algorithm:

[0235]

[0236] The principle of the moving average algorithm is as follows:

[0237]

[0238] To address the typical daily power output of photovoltaic systems in a port, two algorithms were designed and implemented based on equations (23) and (24). The comparison results of the grid-connected power fluctuations before and after control are as follows: Figure 4 As shown.

[0239] Switching the original photovoltaic output from a typical day to an extreme day, while keeping other parameters unchanged, and retaining the controller parameters with the typical day's setpoints, the control effects of the two algorithms are as follows: Figure 5 As shown.

[0240] The following analysis examines the state of the hybrid energy storage system under two control algorithms. The comparison results of output power and throughput energy under typical and extreme solar photovoltaic output conditions are as follows: Figure 6 , Figure 7 As shown.

[0241] The results show that both methods can mitigate short-term fluctuations in photovoltaic power to some extent, but they exhibit response lag when photovoltaic output changes rapidly, easily leading to overcompensation or undercompensation, and causing grid-connected power fluctuations to exceed the set threshold. The low-pass filtering algorithm performs well in suppressing high-frequency fluctuations, but is insufficient in mitigating low-frequency fluctuations; the moving average algorithm has strong ability to smooth slow fluctuations, but poor adaptability to sudden changes in operating conditions. Furthermore, neither method considers the dynamic changes in the energy state of energy storage, which can easily lead to frequent charging and discharging of energy storage, increasing operational stress.

[0242] Based on the analysis of the limitations of traditional control methods, five key indicators are extracted for evaluating the control performance of multi-objective model prediction algorithms:

[0243] 1) Average fluctuation rate of grid-connected power γ M

[0244]

[0245] In the formula, M represents the number of sampling points within the grid connection cycle; P PV_N Indicates the total installed capacity of the port's photovoltaic power plant; γ M This represents the average volatility of grid-connected photovoltaic power at the port, reflecting the overall level of energy storage in mitigating photovoltaic power fluctuations throughout the entire grid connection cycle.

[0246] 2) Total cumulative energy throughput of energy storage E T

[0247]

[0248] In the formula, T represents the control period; E T This represents the total amount of energy stored during the entire control cycle. The smaller the value, the less power the system outputs during the control cycle, and the less lifespan loss it experiences.

[0249] 3) Energy storage output capacity evaluation coefficient C HESS

[0250] The output capacity of energy storage can be quantified by the energy state (SOE) of the energy storage, and its expression is:

[0251]

[0252] In the formula, T represents the control period; C HESS This represents the system's output capacity evaluation coefficient. The smaller the value, the closer the SOE of the energy storage is to 50%, and the stronger its ability to handle power throughput and smooth photovoltaic fluctuations.

[0253] 4) Maximum energy demand for energy storage operation E H

[0254] E H =max(E(k))-min(E(k)) (28)

[0255] In the formula E H This represents the peak energy level of the energy storage system during the entire control cycle, reflecting the peak energy throughput of the system throughout its entire operating cycle.

[0256] 5) Maximum power requirement P for energy storage operation H

[0257] P H =max(|P(k)|) (29)

[0258] In the formula P H This represents the peak output power of the energy storage system throughout the entire control cycle, reflecting the output capacity required by the energy storage system throughout its entire operating cycle.

[0259] 2. Control effect of the dynamically weighted MPC method

[0260] The prediction time domain for model predictive control is 5 minutes, and the control period is 1 minute. The 1-minute fluctuation results of the well network power before and after control are as follows: Figure 8 Compared with the cumulative probability distribution of fluctuations in traditional methods, for example... Figure 9 .

[0261] Comparing the fluctuation amounts of the four methods before and after control, the dynamic weighted MPC method of this invention, while meeting the grid connection fluctuation requirements, more closely approximates the fluctuation amount before control, demonstrating its ability to significantly reduce overcompensation. Taking a cumulative fluctuation probability of 90% as an example, the corresponding... Figure 9 Among the five points A, B, C, D, and E, the 0.33MW of the dynamic weighted MPC method is the closest to the 0.36MW before control, while the other three methods are 0.14MW, 0.18MW, and 0.20MW respectively.

[0262] Based on the comparison of grid-connected power fluctuations, this paper further compares the effects of state regulation of energy storage systems from the perspectives of output control and energy throughput. The output power and energy throughput of the hybrid energy storage systems before and after control are as follows: Figure 10 Compared with the cumulative probability distribution of the output power of energy storage systems using traditional methods, for example... Figure 11 .

[0263] Based on the proposed key evaluation metrics, the comparison results of the two multi-objective MPC methods, as well as the traditional low-pass filtering algorithm and the moving average algorithm, are shown in Table 1.

[0264] Table 2

[0265]

[0266] As can be seen from Table 3-4: 1) From the perspective of suppressing overcompensation, under the premise of meeting the grid connection volatility requirements, the average grid connection volatility γ of the multi-objective MPC control method is... M The higher value indicates that multi-objective MPC control can effectively reduce overcompensation, with the dynamic weighted MPC method proposed in this invention achieving 2.67%, the highest among the four control methods; 2) From the perspective of energy storage output, the cumulative total charge and discharge energy E of the dynamic weighted MPC method is higher. T The method proposed in this invention reduces the burden on the energy storage system during the process of smoothing photovoltaic fluctuations by 71.5%, 56.9%, and 47.9% respectively, compared to the other three methods. This indicates that the method reduces the lifespan loss of the energy storage system. 3) From the perspective of energy storage output capacity, the energy storage output evaluation coefficient C of the dynamic weighted MPC method is the lowest. HESS The minimum value is far lower than that of traditional control methods, indicating that the method of the present invention can effectively maintain the optimal energy storage state to cope with possible drastic fluctuations in photovoltaic power in the future; 4) From the perspective of energy storage power and capacity requirements, the method of the present invention has the lowest peak output power and the narrowest range of energy variation in the process of suppressing photovoltaic fluctuations in ports. The power configuration requirements are reduced by 42.8%, 26.2% and 23.7% respectively, and the energy configuration requirements are reduced by 67.6%, 37.5% and 18.6% respectively, thereby significantly reducing the energy storage configuration cost and making it more economical.

[0267] Adaptability Validation

[0268] The grid-connected volatility limit was reduced to 5%, while keeping other parameters unchanged, to verify the effectiveness of the method of the present invention under different volatility limits.

[0269] To further verify the adaptability of the method of the present invention under different photovoltaic output conditions, the control effect was tested under extreme daily photovoltaic output conditions. The comparison results of the 1-minute fluctuation before and after control are as follows: Figure 14 As shown, the comparison results of the output power and throughput energy of the hybrid energy storage system are as follows: Figure 15 As shown.

[0270] The control indicators for the two different situations are shown in Table 2:

[0271] Table 3

[0272]

[0273] Based on the above analysis, the method proposed in this study has good adaptability and can achieve effective control under different volatility limits and different photovoltaic output conditions. While ensuring that the 1-minute volatility of the grid-connected power meets the requirements, it also takes into account the state regulation of the hybrid energy storage system, reducing its energy changes and lifespan loss.

[0274] The above description of the present invention is only a preferred embodiment of the present invention and is not intended to limit the implementation of the present invention. Those skilled in the art can easily make corresponding modifications or alterations based on the main concept and spirit of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of protection claimed in the claims.

Claims

1. A method for optimizing the output of a hybrid energy storage system based on dynamic weighted model predictive control, characterized in that, Includes the following steps: Step 1: Establish a joint model of the photovoltaic-hybrid energy storage system, which includes the grid-connected photovoltaic power P. Grid Energy State S of Energy Storage System HESS Energy storage output power P HESS Photovoltaic power forecast P PV Predicted values ​​of grid-connected power and energy state P Grid (k+1) and S HESS (k+1); Step 2: Design a control algorithm for multi-objective model prediction. By jointly predicting the system output through the model, design a multi-objective optimization function and introduce weight coefficients to combine the multiple objectives into a unified optimization cost function. Step 3: Design a dynamic weighting method for the weight coefficients, introduce the Harris Eagle algorithm, solve the weight coefficients of the multi-objective function in real time, and obtain and execute the optimal energy storage output sequence based on the rolling optimization principle.

2. The method for optimizing the output of a hybrid energy storage system based on dynamic weighted model predictive control as described in claim 1, characterized in that, In step one, the specific method for establishing the joint model of the photovoltaic-hybrid energy storage system is as follows: According to the principle of energy conservation, at any time t, the system satisfies the following power balance relationship: P Grid (t+1)=P PV (t)+P HESS (t) In the formula: P PV (t) represents the actual output power of the port photovoltaic power plant at time t, P HESS (t) represents the total output power of the hybrid energy storage system. HESS (t) > 0 indicates energy storage and discharge, < 0 indicates charging; P Grid (t+1) represents the grid-connected photovoltaic power at the next moment. Introducing state variable S HESS (t) represents the percentage of the energy storage system's current remaining capacity relative to its rated capacity, and its evolution follows the following law: Among them, E N S represents the rated energy capacity of the energy storage system, Δt represents the time step of the control cycle, and S represents the rated energy capacity of the energy storage system. HESS,t Typically limited to [0.1, 0.9] or [0.2, 0.8] (normalized to 10%–90% or 20%–80%), the specific range is set according to the battery type and lifespan requirements. HESS,t+1 This is the state quantity for the next moment. Convert to a discrete state-space model:

3. The method for optimizing the output of a hybrid energy storage system based on dynamic weighted model predictive control as described in claim 1, characterized in that, In step two, the method for multi-objective model prediction is as follows: Using the discrete state-space model of the photovoltaic-hybrid energy storage system as the prediction model, we obtain: In the formula, k is the sampling time, x(k) is the system state variable, y(k) is the system output variable, u(k) is the control variable, d(k) is the disturbance variable, A is the state matrix, B is the control matrix, C is the output matrix, and D is the disturbance matrix. Based on the coupling relationship between variables, equation (4) is iteratively calculated, and the predicted system output y(k+i) at time k+i is: In the formula, y(k+i) is the predicted value of the system output at time k+i by the model prediction stage at time k, that is, the grid-connected photovoltaic power P at time k+i. Grid (k+i) and the energy state S of the hybrid energy storage system HESS (k+i).

4. The method for optimizing the output of a hybrid energy storage system based on dynamic weighted model predictive control as described in claim 1, characterized in that, In step two, the multi-objective optimization function is designed as follows: With the primary objective of smoothing grid-connected power and minimizing the deviation between the grid-connected power and the ideal target curve, the objective function J1 is established as follows: In the formula, P Grid (k+i) represents the grid-connected power of the port's photovoltaic system at time k+i, N is the prediction time domain length of the model predictive control, and k is the current control time. To minimize the output power of the energy storage system, the objective function J2 is established as follows: To maintain the energy state near the median and preserve the system's ability to continuously regulate, the objective function J3 is established as follows:

5. The method for optimizing the output of a hybrid energy storage system based on dynamic weighted model predictive control as described in claim 1, characterized in that, In step two, the method of introducing weighting coefficients to combine multiple objectives into a unified optimization cost function is as follows: We introduce weighting coefficients α(k), β(k), and λ(k) to characterize the relative importance of each control sub-objective at each time step. We then use a weighted summation method to combine the three optimization objectives into a unified optimization cost function. These correspond to the output power constraints of the hybrid energy storage system, the energy state constraints of the hybrid energy storage system, and the power fluctuation constraints of the photovoltaic grid connection, respectively: In the formula P HESS_max S represents the maximum output power of the hybrid energy storage system. min and S max These are the lower and upper limits of the State of Energy (SOE) for hybrid energy storage systems, respectively. maxP Grid_1min (k) and minP Grid_1min (k) represents the maximum and minimum values ​​of the photovoltaic grid-connected power at the 1-minute scale, respectively, and γ is the grid-connected volatility limit.

6. The method for optimizing the output of a hybrid energy storage system based on dynamic weighted model predictive control as described in claim 1, characterized in that, In step three, a dynamic weighting method for the weight coefficients is designed, introducing the Harris Eagle algorithm, which is divided into three parts: a global search phase, a transition phase, and a local development phase. The global search and transition phase methods are as follows: (1) Global search phase The initial iterative process for optimizing the weight coefficient parameters involves randomly generating a series of weight coefficient combinations and continuously selecting the relatively optimal solution from them. (2) Transition Phase The concept of escape energy E is introduced to facilitate the transition from the global search phase to the local exploitation phase. The formula for calculating escape energy E is as follows: In the formula, E0 is the initial escape energy of the prey, which is a random number in the range of [-1,1]; t is the current iteration number, and T is the maximum iteration number; when |E| < 1, the algorithm enters the subsequent local development stage, otherwise it remains in the global search stage.

7. The dynamic weighting method for design weight coefficients according to claim 6, characterized in that, The partial development phase method is as follows: When |E|<1, four siege methods will be used to perform a local search based on |E| and a random number r in the range [0,1]. When |E|≥0.5 and r≥0.5, the position update formula is as follows: X i (t+1)=ΔX i (t)-E|J i X p (t)-X i (t)| ΔX i (t)=X p (t)-X i (t) In the formula, X i (t) represents the position of the optimization objective at the t-th iteration, i.e., the weight coefficient combination (α(t), β(t), λ(t)) corresponding to this iteration. i X is a random number in the range [0,2]. p (t) represents the position of the prey in the t-th iteration, and ΔX represents the optimal weight coefficient up to the t-th iteration; i (t) represents the distance between the actual value and the target value at the t-th iteration, and represents the difference between the weight coefficient obtained in this iteration and the current optimal weight coefficient. When |E| < 0.5 and r ≥ 0.5, the position update formula is as follows: X i (t+1)=X p (t)-E|ΔX i (t)|。 8. The dynamic weighting method for design weight coefficients according to claim 7, characterized in that, The partial development phase method also includes: When |E|≥0.5 and r<0.5, the position update formula is as follows: Y i (t)=X p (t)-E|J i X p (t)-X i (t)| Z i (t)=Y i (t)+S i ·LF(D) In the formula, F(·) is the objective function used to sort the frontier values; D is the dimension of the problem to be solved; S represents a D-dimensional random row vector, where LF is the Levy function, and its expression is: Where u and v are both random numbers, β is a constant, and has When |E| < 0.5 and r < 0.5, the position update formula is as follows: Y i (t)=X p (t)-E|J i X p (t)+X m (t)| Z i (t)=Y i (t)+S i ·LF(D) Finally, the optimal solution set of the control quantity sequence is obtained in each control cycle, and the first item of the solution set is executed to achieve rolling optimization.

9. A hybrid energy storage system based on dynamic weighted model predictive control, characterized in that, The invention includes a computer module that utilizes a method for optimizing the output of a hybrid energy storage system based on dynamic weighted model predictive control, as described in any one of claims 1 to 8.

Images