A data-driven control system for battery energy storage in islanded microgrids

Abstract

A data-driven control system for battery energy storage in islanded microgrids, a data-driven grid support (DDGS) and data-driven method for tuning parameters, using a data-driven method to adjust voltage and frequency controllers, can be implemented or retrofitted without changing existing power source internal control loops; the control system allows the frequency dependence of the control gain, so that the droop control has been significantly improved, the droop control only uses fixed control gain, which is applied to the error signal at all frequencies.

Description

Technical Field

[0001] This invention belongs to the field of microgrids and relates to a data-driven control system for battery energy storage in islanded microgrids. Background Technology

[0002] Islanded microgrids are prone to voltage and frequency instability due to their low generation capacity and small synchronization inertia. Installing excess generation capacity and increasing spinning reserves can reduce voltage and frequency instability. However, this increases costs, reduces the efficiency of fossil fuel generators, and increases fuel consumption and greenhouse gas emissions. Alternatively, battery energy storage systems can help maintain microgrid stability by providing power reserves. These systems connect to the microgrid via a power electronic controller, enabling them to participate in voltage and frequency control, or an improved secondary control system, which provides faster voltage and frequency recovery. However, microgrid stability relies more heavily on the primary control system. Therefore, improving the control system is necessary to enhance the stability of islanded microgrids.

[0003] Grid-connected inverters are an alternative that use power-based synchronization instead of phase-locked loops (PLLs). Grid-connected inverters can replicate the inertial behavior of a motor more accurately than machine-based inverters. In most cases, the control systems of grid-connected inverters are designed to replicate the behavior of synchronous motors, although some also replicate induction motors. Machine-based control systems have several advantages: they are intuitively understandable to electrical engineers, they can provide instantaneous active power responses to frequency changes, and they can sometimes operate independently.

[0004] Another, more rigorous approach to designing distributed master control systems for grid-connected inverters involves using a dynamic phasor model of the microgrid to adjust the master controllers of both the battery storage system and the generator. This system-wide approach offers several advantages, as it allows for joint optimization of the battery storage system control system and the generator control system. Furthermore, performance and stability are guaranteed compared to machine-based control systems.

[0005] Moving away from dynamic model-based controller design, data-driven control design methods have been applied in power electronics and power systems. Data-driven methods can be used to tune control systems for electromagnet power supplies. A data-driven adaptive voltage control system for DC microgrids also exists. These data-driven control methods reduce the discrepancy between theoretical and practical performance, lower modeling requirements, reduce the complexity of control structures, and simultaneously provide guaranteed performance and stability.

[0006] In grid-connected inverter solutions, gate-forming inverters require extensive modifications to the inverter control system compared to more typical grid-connected inverters. This can significantly increase costs, which does not incentivize manufacturers to invest in commercialization. Furthermore, the methods for tuning grid-connected inverter simulation machine parameters are heuristic or use simplified dynamic power system models, requiring simplification of power system behavior and complete neglect of voltage dynamics. This can lead to oscillating response and poor performance in practice, and cannot guarantee small-signal stability and performance.

[0007] There are also centralized control schemes, where a centralized master control system exhibits good performance. However, only simulation results are presented, and concerns remain regarding the practical implementation and reliability of centralized master control systems in practice due to communication delays and reliability issues. Alternatively, a more broad discussion can be made of distributed master control systems using droop control, including inertial damping methods and grid-connected inverters. Inertial damping methods use frequency rate of change measurements and inertial gain to increase the frequency drop controller. However, in practice, frequency rate of change measurements require strong filtering with a time constant of approximately 1 second to avoid unstable interactions with the converter control loop and the converter / grid interface, which can severely degrade the performance of low-inertia systems. Furthermore, inertial damping methods only consider the active power control loop. This can lead to undesirable interactions with the reactive power controller and instability, and may negatively impact small-signal stability. Summary of the Invention

[0008] To overcome the shortcomings of existing technologies, this invention provides a data-driven control system for battery energy storage in islanded microgrids. Data-driven islanded microgrid support (DDGS) and a data-driven method are used to adjust parameters. The voltage and frequency controllers are adjusted using the data-driven method, which can be implemented or modified without changing the existing power supply internal control loop. This control system allows for frequency dependence of control gain, resulting in a significant improvement in droop control. Droop control uses only a fixed control gain and applies to error signals at all frequencies.

[0009] The technical solution adopted by this invention to solve its technical problem is:

[0010] A data-driven control system for battery energy storage in islanded microgrids, characterized in that the data-driven islanded microgrid supports the DDGS control system, x i，…j =0, 1, ..., p is the molecular parameter, y i，…j =0, 1, ..., q-1 are the denominator parameters. The DDGS control system has 2(p+q-1) parameters. In order to adjust the DDGS controller, a calculation-based optimization method is used, and the tuning is performed directly in the discrete time domain.

[0011] First, identify the frequency response data (FRD) of the power system. The FRD uses time-domain input and output data obtained from a real-time power system simulator. The data is collected on the same system that the controller is designed for.

[0012] The microgrid is excited using an injected pseudo-random binary sequence (PRBS), which is a repeating discrete binary sequence using only two values. Its power spectral density (PSD) is close to that of white noise. Since the PRBS contains multiple frequency components, the system output is measured by injecting the PRBS. The PRBS is generated using a linear feedback shift register of length n, with the maximum sequence length LL = 2 defined by the following formula: n -1;

[0013] Before applying PRBS to the system, it is set to bipolar with an amplitude of a and an identification sampling period T. s，id Determine the maximum excitation frequency, while the excitation amplitude of ±a is defined by applying the amplitude of the control input;

[0014] In order to perform identification, BESS, P grid,ref and Q grid,ref The control input is sequentially excited via PRBS, while the measured output ω is recorded. t and v t,rms The response; this process is repeated twice, using different T values. s,id The value is used to improve the resolution of the frequency response in the low-frequency range; the resulting data is used to adjust the DDGS controller, with a data sampling time step of T. s,sample =T s,id Take m PRBS periods, and then calculate the number of samples in each dataset, i.e., the record length N. rec =m(2 n -1);

[0015] The time-domain identification dataset collected using PRBS signal injection is processed to find the system’s a 2×2 MIMO frequency response data system, which correlates the control input u of the ESS with the system output measured by the inverter y. The frequency response includes the behavior of the SRF power controller, measurement sensors and microgrid system.

[0016] To find the MIMO frequency response data and G, the collected dataset was processed using spectral analysis methods, and the transfer function is shown below:

[0017]

[0018] Among them, R uu It is the autocorrelation of the input PRBS, R uyIt is the cross-correlation between the input PRBS and the measurement system output; DFT is the Discrete Fourier Transform operator. This is the corresponding element of the MIMO frequency response data G; this results in two sets of MIMO frequency response data, set 1 and set 2, each with a frequency point of L / 2, and the linear interval between the frequency points of set 1 and set 2 is 2 / (LT). s,id,1 )Hz and 2 / (LT s,id,2 )Hz;

[0019] Finally, a set of higher-resolution frequency response data is constructed, with array 1 covering a wider spectrum, although the frequency resolution is lower than that of array 2; therefore, the points extracted from array 2 are truncated to cover only the 0-1Hz range and appended to array 1 to increase the resolution of the frequency response data at low frequencies, thereby constructing a complete frequency response dataset for system G.

[0020] The initial stabilizing controller used in the DDGS design process is the conventional IDGS controller described in Section C, namely:

[0021]

[0022] In a closed-loop system with a DDGS controller, and They are K DD1 (z) and K DD2 The frequency response of (z) is given by r, which is a vector of reference values. Vector d represents the interference caused by load changes on the microgrid to the system voltage and frequency. The DDGS controller K controls the frequency and voltage errors based on the following conditions:

[0023]

[0024] The control objective of this control system is to suppress disturbances to the system frequency and voltage. Therefore, the optimization objective is to minimize the H∞ norm of the weighted system sensitivity, i.e.:

[0025]

[0026] X is the molecular parameter x i Y is K DD1 (z) and K DD2 The denominator parameter y of (z) i S is the sensitivity function: S = (I + GK) -1 ;

[0027] The DC gain of the controller is constrained to achieve steady-state active and reactive power sharing, i.e.

[0028] This constraint is repeated for k = 1, 2, where the subscript k represents an element of the controller input vector e, i.e., e ω and e v ;x i,k ,i=0,1,…,p are molecular parameters, y j,k j = 0, 1, ..., q-1 are the denominator parameters of the DDGS controller, k init,k This is the gain of the initial IDGS controller;

[0029] Finally, in order to use MIMO FRD to solve the optimization problem, the optimization problem is convexized around it, the optimization problem is solved for a set of frequency points ω, and then the objective is checked for convergence.

[0030] Compared to traditional grid-supported inverters, this system does not require changes to the actual power and reactive power control loops within the inverter. Adjusting the data-driven controller does not require a dynamic model of the microgrid. Instead, it identifies the frequency response of the microgrid and directly optimizes and tunes the controller to meet performance and robustness standards. Furthermore, in many cases, it is not possible to perform joint optimization of the entire microgrid main control system. Therefore, it is more advantageous to adopt a control method specifically for battery energy storage systems.

[0031] The beneficial effects of this invention are mainly reflected in the following aspects: Data-driven islanded microgrid support (DDGS) and data-driven methods are used to adjust parameters. The voltage and frequency controllers are adjusted using data-driven methods, which can be implemented or modified without changing the existing power supply internal control loop. The control system allows for frequency dependence of control gain, which significantly improves droop control. Droop control only uses a fixed control gain and is applied to error signals at all frequencies. Attached Figure Description

[0032] Figure 1 This is a block diagram of the data-driven islanded microgrid support control system.

[0033] Figure 2 This is a diagram illustrating the design process of a data-driven, islanded microgrid support control system.

[0034] Figure 3 This is a block diagram of a closed-loop system with a DDGS controller. Detailed Implementation

[0035] The present invention will now be further described with reference to the accompanying drawings.

[0036] Reference Figures 1-3 A data-driven control system for battery energy storage in islanded microgrids, wherein the data-driven islanded microgrid supports DDGS control systems such as... Figure 1 As shown, x i,…j =0,1,…,p are molecular parameters, yi,…j =0,1,…,q-1 are the denominator parameters. Compared to the IDGS control system, which has two parameters, the DDGS control system has 2(p+q-1) parameters. The integers p and q are explicitly defined by the designer, allowing for the selection of smaller orders where appropriate. Compared to the drooping controller, the higher order of the fixed-structure controller leads to improved response of the lithium battery energy storage system to frequency and voltage transients when well tuned. A computation-based optimization method is used to tune the DDGS controller; furthermore, tuning is performed directly in the discrete time domain. Therefore, errors due to controller discretization are not introduced.

[0037] First, identify the frequency response data (FRD) of the power system. The FRD uses time-domain input and output data obtained from a real-time power system simulator. The data is collected on the same system that the controller is designed for. If a real microgrid system is available, it should be used for data acquisition and controller implementation. Therefore, there will be no inaccuracies in the FRD used to adjust the controller parameters. This is one of the advantages of the DDGS control system.

[0038] To collect data, the microgrid is excited using an injected pseudo-random binary sequence (PRBS). The PRBS is a repeating discrete binary sequence, employing only two values, with a power spectral density (PSD) close to that of white noise. Since the PRBS contains multiple frequency components, the system output is measured by injecting the PRBS. The PRBS is generated using a linear feedback shift register of length n, with the maximum sequence length LL = 2 defined by the following formula. n -1;

[0039] Before applying PRBS to the system, it is set to bipolar with an amplitude of a and an identification sampling period T. s,id Determine the maximum excitation frequency, while the excitation amplitude of ±a is defined by applying the amplitude of the control input;

[0040] In order to perform identification, BESS, P grid,ref and Q grid,ref The control input is sequentially excited via PRBS, while the measured output ω is recorded. t and v t,rms The response; this process is repeated twice, using different T values. s,id The value is used to improve the resolution of the frequency response in the low-frequency range; the resulting data is used to adjust the DDGS controller, with a data sampling time step of T. s,sample =T s,id To improve noise suppression, m PRBS periods are taken, and then the number of samples in each dataset is calculated, i.e., the record length N. rec =m(2 n -1);

[0041] The time-domain identification dataset collected using PRBS signal injection is processed to find the system's 2×2 MIMO frequency response data system, which correlates the control input u of the ESS with the system output measured by the inverter y. The frequency response includes the behavior of the SRF power controller, measurement sensors, and the microgrid system.

[0042] To find the MIMO frequency response data and G, the collected dataset was processed using spectral analysis methods, and the transfer function is shown below.

[0043]

[0044] Among them, R uu It is the autocorrelation of the input PRBS, R uy It is the cross-correlation between the input PRBS and the measurement system output; DFT is the Discrete Fourier Transform operator. This is the corresponding element of the MIMO frequency response data G; this results in two sets of MIMO frequency response data, set 1 and set 2, each with a frequency point of L / 2, and the linear interval between the frequency points of set 1 and set 2 is 2 / (LT). s,id,1 )Hz and 2 / (LT s,id,2 )Hz;

[0045] Finally, a set of higher-resolution frequency response data is constructed, with array 1 covering a wider spectrum, although the frequency resolution is lower than that of array 2; therefore, the points extracted from array 2 are truncated to cover only the 0-1Hz range and appended to array 1 to increase the resolution of the frequency response data at low frequencies, thereby constructing a complete frequency response dataset for system G.

[0046] Figure 2 The design process of a data-driven support control system for islanded microgrids is presented. The initial stabilizing controller used in the DDGS design process is the conventional IDGS controller described in Section C, namely:

[0047]

[0048] A block diagram of a closed-loop system with a DDGS controller is shown below. Figure 3 As shown, where and They are K DD1 (z) and K DD2 The frequency response of (z) is given by r, which is a vector of reference values. Vector d represents the interference caused by load changes on the microgrid to the system voltage and frequency. The DDGS controller K controls the frequency and voltage errors based on the following conditions:

[0049]

[0050] The primary control objective of this control system is to suppress disturbances to the system frequency and voltage. Therefore, the optimization objective is to minimize the H∞ norm of the weighted system sensitivity, i.e.:

[0051]

[0052] X is the molecular parameter x i Y is K DD1 (z) and K DD2 The denominator parameter y of (z) i S is the sensitivity function;

[0053] S = (I + GK) -1

[0054] The DC gain of the controller is constrained to achieve steady-state active and reactive power sharing, i.e.

[0055]

[0056] This constraint is repeated for k = 1, 2, where the subscript k represents an element of the controller input vector e, i.e., e ω and e v ;x i,k ,i=0,1,…,p are molecular parameters, y j,k j = 0, 1, ..., q-1 are the denominator parameters of the DDGS controller, k init,k This is the gain of the initial IDGS controller;

[0057] Finally, in order to use MIMO FRD to solve the optimization problem, the optimization problem is convexized around it, the optimization problem is solved for a set of frequency points ω, and then the convergence of the objective is checked. At this point, the controller design is complete.

[0058] The embodiments described in this specification are merely examples of implementations of the inventive concept and are for illustrative purposes only. The scope of protection of this invention should not be considered limited to the specific forms described in these embodiments; rather, it extends to equivalent technical means conceived by those skilled in the art based on the inventive concept.

Claims

1. A data-driven control system for battery energy storage in isolated microgrids, characterized in that, Data-driven islanded microgrids support DDGS control systems, x i,…j =0,1,…,p are molecular parameters, y i,…j =0,1,…,q-1 are the denominator parameters. The DDGS control system has 2(p+q-1) parameters. In order to adjust the DDGS controller, a calculation-based optimization method is used, and the tuning is performed directly in the discrete time domain. First, identify the frequency response data (FRD) of the power system. The FRD uses time-domain input and output data obtained from a real-time power system simulator. The data is collected on the same system that the controller is designed for. The microgrid is excited using an injected pseudo-random binary sequence (PRBS), which is a repeating discrete binary sequence using only two values. Its power spectral density (PSD) is close to that of white noise. Since the PRBS contains multiple frequency components, the system output is measured by injecting the PRBS. The PRBS is generated using a linear feedback shift register of length n, with the maximum sequence length LL = 2 defined by the following formula: n -1; Before applying PRBS to the system, it is set to bipolar with an amplitude of a and an identification sampling period T. s,id Determine the maximum excitation frequency, while the excitation amplitude of ±a is defined by applying the amplitude of the control input; In order to perform identification, BESS, P grid,ref and Q grid,ref The control input is sequentially excited via PRBS, while the measured output ω is recorded. t and v t,rms The response; this process is repeated twice, using different T values. s,id The value is used to improve the resolution of the frequency response in the low-frequency range; the resulting data is used to adjust the DDGS controller, with a data sampling time step of T. s,sample =T s,id Take m PRBS periods, and then calculate the number of samples in each dataset, i.e., the record length N. rec =m(2 n -1); The time-domain identification dataset collected using PRBS signal injection is processed to find the system’s a 2×2 MIMO frequency response data system, which correlates the control input u of the ESS with the system output measured by the inverter y. The frequency response includes the behavior of the SRF power controller, measurement sensors and microgrid system. To find the MIMO frequency response data and G, the collected dataset was processed using spectral analysis methods, and the transfer function is shown below: Among them, R uu It is the autocorrelation of the input PRBS, R uy It is the cross-correlation between the input PRBS and the measurement system output; DFT is the Discrete Fourier Transform operator. This is the corresponding element of the MIMO frequency response data G; this results in two sets of MIMO frequency response data, set 1 and set 2, each with a frequency point of L / 2, and the linear interval between the frequency points of set 1 and set 2 is 2 / (LT). s,id,1 )Hz and 2 / (LT s,id,2 )Hz; Finally, a set of higher-resolution frequency response data is constructed, with array 1 covering a wider spectrum, although the frequency resolution is lower than that of array 2; therefore, the points extracted from array 2 are truncated to cover only the 0-1Hz range and appended to array 1 to increase the resolution of the frequency response data at low frequencies, thereby constructing a complete frequency response dataset for system G. The initial stable controller used in the DDGS design process is the conventional IDGS controller described in Section C, namely: In a closed-loop system with a DDGS controller, and They are K DD1 (z) and K DD2 The frequency response of (z) is given by r, which is a vector of reference values. Vector d represents the interference caused by load changes on the microgrid to the system voltage and frequency. The DDGS controller K controls the frequency and voltage errors based on the following conditions: The control objective of the control system is to suppress disturbances to the system frequency and voltage. Therefore, the optimization objective is to minimize the H∞ norm of the weighted system sensitivity, i.e.: X is the molecular parameter x i Y is K DD1 (z) and K DD2 The denominator parameter y of (z) i S is the sensitivity function: S = (I + GK) -1 ; The DC gain of the controller is constrained to achieve steady-state active and reactive power sharing, i.e. This constraint is repeated for k = 1, 2, where the subscript k represents an element of the controller input vector e, i.e., e ω and e v ;x i,k ,i=0,1,…,p are molecular parameters, y j,k j = 0, 1, ..., q-1 are the denominator parameters of the DDGS controller, k init,k This is the gain of the initial IDGS controller; Finally, in order to use MIMO FRD to solve the optimization problem, the optimization problem is convexized around it, the optimization problem is solved for a set of frequency points ω, and then the objective is checked for convergence.

Images