A network construction converter coordinated control method for water-based organic liquid flow energy storage battery

Abstract

The application discloses a network construction converter coordination control method for water-based organic liquid flow energy storage batteries. The method first builds a state space model of the coupling of the converter and the energy storage battery, a 7-dimensional composite state vector, divides an operation mode label, establishes a discrete prediction model, generates a system matrix and a frozen matrix; determines the operation mode according to relevant information and keeps the mode stable within the prediction time domain. Then, a quadratic programming rolling optimization objective function is built, a smooth reference sequence is generated through exponential transition, equivalent disturbances are estimated online and feedforward compensation is introduced. After the basic control quantity is obtained by solving, correction is completed with the help of a switching inertia compensation matrix, and finally, double-rate prediction control is adopted and cyclic iteration is performed. The method can adapt to the slow time-varying characteristics of the water-based organic liquid flow battery, effectively weaken the transient impact caused by mode switching, and improve the control stability of the converter.

Description

Technical Field

[0001] This invention belongs to the field of power electronics and energy storage system control technology, specifically relating to a grid converter coordinated control method for aqueous organic liquid flow energy storage batteries. Background Technology

[0002] Aqueous organic flow batteries are gaining increasing attention in the energy storage field due to their advantages such as high safety, long lifespan, and independently expandable capacity. However, unlike lithium-ion batteries or inorganic flow batteries, the electrolyte viscosity, pH value, and concentration of organic active materials in aqueous organic flow batteries change slowly with temperature and flow rate, resulting in significant slow time-varying characteristics in the equivalent output impedance and polarization properties, with time constants typically on the order of seconds to minutes. As the core equipment for interaction with the grid, grid-connected converters need to provide voltage and frequency support for weak grids or islanded systems while overcoming the effects of these slowly varying disturbances. Existing technologies have the following shortcomings: 1) Large impact during mode switching: Common control strategies, such as droop control and virtual synchronous machine control, are independently designed. The control law changes abruptly at the moment of switching, resulting in transient impacts of voltage and current several times the rated value. Existing literature has pointed out that the peak impact current in hard switching mode can reach 2.5 times.

[0003] 2) Multi-machine parallel resonant circulating current: Due to the difference in line impedance and the coupling of control parameters, it is difficult to suppress the circulating current. According to literature reports, the droop control parallel circulating current can reach more than 15% of the rated current.

[0004] 3) Conflict between dynamic response and steady-state accuracy: Fixed parameter control makes it difficult to balance fast response and steady-state accuracy.

[0005] 4) Existing limitations of MPC applications: Existing MPC-based converter control mainly focuses on output voltage tracking, without incorporating mode switching, multi-machine coordination, and the gradual change characteristics of flow batteries into a unified optimization framework. Furthermore, the prediction model is fixed and has poor adaptability to changes in battery state.

[0006] Therefore, a coordinated control method for grid converters is needed that can organically integrate smooth mode transition, multi-constraint rolling optimization and disturbance compensation, and enhance the slow time-varying characteristics of aqueous organic flow batteries. Summary of the Invention

[0007] The present invention aims to solve the problems of large voltage and frequency surges, uneven power distribution, unsmooth mode switching, and poor adaptability to the slow-changing operating conditions of flow batteries in grid-connected converters during grid-connected / off-grid switching and multi-unit parallel operation.

[0008] To achieve the above objectives, the present invention adopts the following technical solution: A grid-connected converter coordinated control method for aqueous organic liquid flow energy storage batteries includes the following: Step S1: Construct a state-space model of the coupled grid converter and the water-based organic liquid flow energy storage battery, and construct a 7-dimensional composite state vector; divide the operation into four types of operation mode labels, with the mode labels being independent of the composite state vector; combine the state-space model with a fixed sampling period to establish a discrete prediction model, and generate a complete system matrix and a frozen matrix. Step S2: Determine the current operating mode label based on external instructions, local electrical quantity detection results, and switching timing information; keep the operating mode label constant within the model prediction time domain of the discrete prediction model; Step S3: Based on the 7-dimensional composite state vector and the current operating mode label, construct a rolling optimization objective function in the form of quadratic programming; Step S4: Generate a smooth reference sequence using an exponential transition function; based on the output residual method and state observer, estimate the equivalent additive perturbation in the discrete prediction model online, and input the perturbation value as a feedforward quantity into the discrete prediction model; Step S5: Using the discrete prediction model, rolling optimization objective function, smoothed reference sequence and disturbance value as input, complete the state recursion and quadratic programming solution in the model prediction time domain, output the optimal control increment sequence, and take the first increment as the basic control quantity. Step S6: Calculate the switching inertia compensation matrix based on the operating modes of the current time and the previous time; use the compensation matrix to perform feedforward correction on the control increment of the previous time step, and superimpose the correction result into the basic control quantity; Step S7: Combining the model's prediction of state recursion in the time domain, the complete system matrix and the frozen matrix are called to perform dual-rate predictive control, and the state is updated alternately according to a preset period and the results are sent back to step S5 to adapt to the slow time-varying conditions of the aqueous organic flow battery.

[0009] Preferably, the method further includes the following steps: In step S8, when multiple grid-connected converters are running in parallel, each device uses peer-to-peer distributed communication to exchange active power and frequency reference data. The average value of all device data is used as a common reference frequency, which is then used for the processing flow in steps S2 and S4. When a communication anomaly is detected, the operating mode is downgraded, and each parallel grid-connected converter automatically switches to an independent single-unit operating mode and enables a fixed reference frequency. In single-unit mode, all control processes from steps S1 to S7 are continuously executed to achieve overall coordinated operation of multiple units.

[0010] Preferably, in step S1, the 7-dimensional composite state vector includes voltage deviation, frequency deviation, active power deviation, reactive power deviation, virtual inertia state, and dq-axis switching memory continuous state, wherein the calculation formula for the dq-axis switching memory continuous state is: ; in, To switch the memory continuous state for the d-axis, at time k; The q-axis is switched to a continuous state of memory at time k; To switch the memory continuous state for the d-axis, at time k-1; Switch the memory continuous state for the q-axis at time k-1; The exponential decay coefficient; For a fixed sampling period; It is a natural constant; The d-axis control increment at time k-1; The increment for q-axis control at time k-1.

[0011] Preferably, in step S2, the logic for determining the current operating mode label based on the local electrical quantity detection results and the switching timing information is as follows: The effective voltage value and grid frequency at the grid connection point of the grid-connected converter are continuously monitored. If any parameter exceeds ±5% of the corresponding rated value, it is judged as an over-limit event. If an over-limit event occurs for three consecutive power frequency cycles, the system is determined to be in an off-grid state, and the corresponding current operating mode label is off-grid virtual synchronous machine mode; otherwise, the system is determined to be in a grid-connected state, and the corresponding current operating mode label is grid-connected droop mode. If the newly determined grid-connected or off-grid status is inconsistent with the currently effective operating mode label, a mode switch is triggered, and the current operating mode label is switched to a transient switching transition mode. A fixed transition duration is timed according to the switching timing information in step S2. After the timing ends, the system automatically switches to the target stable operating mode label to prevent frequent mode jumps.

[0012] Preferably, in step S3, the rolling optimization objective function includes three types of independent cost terms: the first type is the output deviation weighted square term composed of voltage, frequency, active power, and reactive power deviations; the second type is the control increment weighted square term; and the third type is the virtual inertia state change weighted square term. The virtual inertia state change weighted square term is used to constrain the smooth change of the virtual inertia trajectory and suppress the voltage and frequency secondary oscillations caused by mode switching and load sudden changes.

[0013] Preferably, in step S4, an exponential transition function is used to perform self-shaping processing on the target reference value to generate a smooth reference sequence, wherein the calculation formula of the function used for self-shaping processing is:

[0014] in, This is the smoothed reference value for the i-th step in the time domain at the current sampling time k. The weighting coefficients of the exponential transition function vary with the number of prediction steps. Exponential decay; The target reference value vector before the switch; This is the target reference value vector after the switch; The attenuation coefficient; For predicting the time-domain incremental step index; For a fixed sampling period.

[0015] Preferably, in step S5, when performing the quadratic programming solution, the rolling optimization objective function of the quadratic programming form constructed in step S3 is configured with multiple linear inequality constraints such as electrical quantity amplitude, control increment, voltage frequency, and virtual inertia; the interior point method is used as the solution algorithm, and a fixed maximum number of iterations is set to ensure that the overall solution time is less than the fixed sampling period in step S1, so as to meet the real-time operation requirements of the controller of the grid converter.

[0016] Preferably, in step S6, the formula for obtaining the switching inertia compensation matrix is: ; in, The source mode number for the switching operation; The target mode number is assigned to the switching operation; The adjustable coefficient for switching the inertial compensation intensity; The system matrix for the target mode; The system matrix for the source mode; The control input matrix for the source mode; I is the regularization parameter; I is the AND matrix. An identity matrix of the same dimension.

[0017] Preferably, in step S7, the specific execution rule of the dual-rate predictive control is as follows: taking the fixed sampling period of step S1 as the timing reference, the dual-rate period multiple is set to 10; every 10 fixed sampling periods, the complete system matrix is ​​called to complete the update of all state variables; in other regular sampling periods, the frozen matrix is ​​called to carry out state prediction calculation; the frozen matrix is ​​obtained by modifying the complete system matrix generated in step S1, and the specific modification method is: replacing the matrix rows corresponding to the virtual inertia state and the dq axis switching memory continuous state in the complete system matrix with standard basis vectors, so that the two types of slow dynamic state variables, the virtual inertia state and the dq axis switching memory continuous state, remain numerically constant during the state prediction process, adapting to the inherent slow time-varying operating conditions of the aqueous organic flow battery.

[0018] Preferably, in step S8, the peer-to-peer distributed communication is set to a fixed communication period of 10ms; each parallel-connected converter in the network periodically broadcasts its own active power and frequency reference data, and calculates the average value of all parallel unit data as the common reference frequency; when the communication interruption time exceeds 30ms, all parallel-connected converters in the network synchronously switch to independent single-machine operation mode and enable the fixed reference frequency; after the communication link is restored to normal, each converter re-establishes peer-to-peer distributed communication, synchronously calculates the common reference frequency again, and switches back to multi-machine parallel coordinated operation mode.

[0019] The present invention provides a grid-connected converter coordinated control method for aqueous organic liquid flow energy storage batteries, which achieves several technical advantages: 1. Shock suppression by switching the inertial compensation matrix: With the reference self-shaping mechanism, the voltage transient overshoot during grid-connected / off-grid switching is reduced compared to traditional hard switching.

[0020] 2. Multi-machine parallel current sharing: Under the rolling optimization framework of this invention, in conjunction with the peer-to-peer distributed communication structure, under the same line impedance difference conditions, the steady-state circulating current is less than 3% of the rated current.

[0021] 3. Adapting to slowly varying operating conditions of flow batteries: Online disturbance estimation It can compensate for the equivalent output impedance shift caused by the slow changes in electrolyte viscosity, conductivity, and active substance concentration with temperature / flow rate, thus reducing the offline identification frequency.

[0022] 4. Unified Coordination Framework: Incorporate electrical quantities and slow energy states into the same prediction model to avoid multi-loop control parameter tuning coupling. Attached Figure Description

[0023] Figure 1 This is a flowchart of steps S1-S7 of a grid converter coordinated control method for an aqueous organic liquid flow energy storage battery according to the present invention.

[0024] Figure 2 This is a diagram showing the overall architecture of the control method of the present invention.

[0025] Figure 3 This is a schematic diagram of the switching inertia compensation logic of the present invention.

[0026] Figure 4 This is a flowchart of steps S1-S8 of a grid converter coordinated control method for an aqueous organic liquid flow energy storage battery according to the present invention. Detailed Implementation

[0027] The following detailed implementation of a grid-connected converter coordinated control method for an aqueous organic liquid flow energy storage battery according to the present invention will be described in detail with reference to specific embodiments. The embodiments of the present invention are only used to explain the present invention and are not intended to limit the scope of protection of the present invention.

[0028] Example 1: Implementation of a grid converter coordinated control method for an aqueous organic liquid flow energy storage battery.

[0029] Combined with appendix Figures 1-3 As shown, the present invention provides a grid converter coordinated control method for an aqueous organic liquid flow energy storage battery.

[0030] like Figure 1 As shown, Figure 1 This is a flowchart of steps S1-S7 of a grid converter coordinated control method for an aqueous organic liquid flow energy storage battery according to Embodiment 1 of the present invention. Figure 2 This is a diagram showing the overall architecture of the control method of the present invention. Figure 2 The overall architecture of the coordinated control method for grid-connected converters is demonstrated. After the sensors collect voltage and current signals, they are sent to the state observer and the mode scheduler, respectively. The MPC optimizer integrates state estimation, mode determination, reference shaping, and disturbance feedforward information to calculate the optimal control command and drive the converter through pulse width modulation output.

[0031] Step S1: Establish the coupled state-space model of the grid converter-flow battery.

[0032] Step S1: Construct a state-space model of the coupled grid converter and the water-based organic liquid flow energy storage battery, and construct a 7-dimensional composite state vector; divide the operation into four types of operation mode labels, with the mode labels being independent of the composite state vector; combine the state-space model with a fixed sampling period to establish a discrete prediction model, and generate a complete system matrix and a frozen matrix.

[0033] In step S1, the 7-dimensional composite state vector includes voltage deviation, frequency deviation, active power deviation, reactive power deviation, virtual inertia state, and dq-axis switching memory continuous state. The calculation formula for the dq-axis switching memory continuous state is as follows: ; in, To switch the memory continuous state for the d-axis, at time k; The q-axis is switched to a continuous state of memory at time k; To switch the memory continuous state for the d-axis, at time k-1; Switch the memory continuous state for the q-axis at time k-1; The exponential decay coefficient; For a fixed sampling period; It is a natural constant; The d-axis control increment at time k-1; The q-axis control increment is set at time k-1. The specific process is as follows: Define a composite state vector (continuous real number field, ): ; in, Voltage deviation (pu); Frequency deviation ; , Active / reactive power deviation (pu); The internal state of virtual inertia simulates the difference variables in the equation of motion of a synchronous generator rotor; , Switch between continuous memory states (one for each of the d and q axes), defined as a first-order exponential moving average (IIR low-pass filter): ; in, To switch the memory continuous state for the d-axis, at time k; The q-axis is switched to a continuous state of memory at time k; To switch the memory continuous state for the d-axis, at time k-1; Switch the memory continuous state for the q-axis at time k-1; The exponential decay coefficient has dimensions of . ; For a fixed sampling period; It is a natural constant; The d-axis control increment at time k-1; The increment for q-axis control at time k-1.

[0034] Abbreviated as And satisfy ,in . To control the input vector (dq axis voltage command). .

[0035] The core function of switching memory states: As an increment of control over the past few steps The exponentially weighted memory is explicitly used in the MPC model for state prediction. It is used in conjunction with the switching inertia compensation matrix. The combined effect can apply smooth memory to future control sequences during mode switching, ensuring that control increments remain first-order continuous at switching boundaries, thereby reducing shocks.

[0036] Run mode label Defined as: 1: Grid-connected droop mode; 2: Off-grid virtual synchronous machine mode; 3: Multi-machine parallel master-slave mode; 4: Transient switching transition mode.

[0037] The mode label, as a scheduling parameter, is independent of the state vector and is determined by the mode scheduler (see step S2) based on external instructions and electrical quantity logic judgments. It does not participate in the matrix multiplication of the continuous state equations.

[0038] The operating mode label is independent of the composite state vector. It refers to the four types of operating mode labels: grid-connected droop, off-grid virtual synchronous machine, multi-machine parallel master-slave, and transient switching transition. These labels are discrete scheduling parameters and are not components of the 7-dimensional composite continuous state vector. Their values ​​are determined by the mode scheduler based on external instructions, local electrical quantity detection logic, and switching timing information. They do not participate in the matrix operations of the continuous state equation. This design can keep the mode label constant in the prediction time domain, so that the rolling optimization quadratic programming problem does not need to deal with discrete variables to maintain the convex optimization characteristics. The anti-jump control of mode switching is implemented by the mode scheduler's own logic threshold and hysteresis mechanism.

[0039] Establish a discrete prediction model (sampling period) Using an incremental approach, the optimization variable is set as ): ; in: , This is the system matrix corresponding to the current mode; This is an equivalent additive perturbation; To control the increment for the previous move; The tuning method for switching the inertial compensation matrix is ​​described in step S6.

[0040] The method for establishing the state-space matrix (taking grid-connected mode as an example): Based on a three-phase LCL filter circuit, the state equations for inductor current, capacitor voltage, and grid voltage are established in the dq rotating coordinate system. After small-signal linearization, a continuous-time state matrix is ​​obtained, and then the zero-order hold method is used with a sampling period... Discretization to obtain the electrical component sub-matrix , The power deviation and virtual inertia state are appended to the state vector, and corresponding rows are constructed according to the power calculation formula to form a complete state vector. , .

[0041] In particular, The update pattern is already implicit in and middle: Chinese correspondence The diagonal piece is , Chinese correspondence The block is Predicting time domain Since it is a variable to be optimized, The recursion can be performed directly. It is a 2-order identity matrix.

[0042] In summary, the state-space model of the coupled grid converter and the organic liquid flow battery in the water system is constructed as follows: Based on the three-phase LCL filter circuit of the grid converter, a continuous state equation for electrical quantities is established in the dq rotating coordinate system. After small-signal linearization, the equation is discretized using the zero-order hold method with a fixed sampling period to obtain the electrical component sub-matrix. The power deviation, virtual inertia state, and dq axis switching memory continuous state are extended into composite state variables. Corresponding matrix elements are constructed according to the power calculation relationship, the differential relationship of the virtual synchronous machine rotor motion, and the first-order exponential recursion rule. These elements are then assembled to form a 7-dimensional complete system matrix and control input matrix corresponding to the operating mode label. Finally, an equivalent additive disturbance term and a switching inertia compensation term characterizing the slow time-varying characteristics of the liquid flow battery are introduced. An incremental discrete prediction model is established with the control increment as the optimization variable, thus completing the construction of the coupled state-space model.

[0043] Step S2, mode scheduling and prediction mode assumptions in the time domain.

[0044] Step S2 determines the current operating mode label based on external instructions, local electrical quantity detection results, and switching timing information; within the model prediction time domain of the discrete prediction model, the operating mode label remains constant.

[0045] In step S2, the logic for determining the current operating mode label based on local electrical quantity detection results and switching timing information is as follows: The effective voltage value and grid frequency at the grid-connected point of the grid-connected converter are continuously detected. If either parameter exceeds ±5% of the corresponding rated value, it is determined as an over-limit event. If over-limit events occur for three consecutive power frequency cycles, it is determined as an off-grid state, and the corresponding current operating mode label is off-grid virtual synchronous machine mode; otherwise, it is determined as a grid-connected state, and the corresponding current operating mode label is grid-connected droop mode. If the newly determined grid-connected or off-grid state is inconsistent with the currently effective operating mode label, a mode switch is triggered, and the current operating mode label is switched to transient switching transition mode. A fixed transition time is timed according to the switching timing information in step S2. After the timer expires, it automatically switches to the target stable operating mode label to prevent frequent mode jumps. The specific process is as follows: The mode scheduler determines the mode at each sampling time based on the following information. : External commands, grid connection switch status, host computer mode commands; External commands specifically include grid-connected switch status signals and mode control commands issued by the host computer. These serve as one of the input bases for the mode scheduler and are used to issue planned mode switching commands to the converter. At each sampling time, the mode scheduler combines the external commands, local electrical quantity over-limit detection results, and switching timing information to jointly determine and identify the currently effective operating mode label.

[0046] Local electrical quantity judgment: Continuously monitor the effective value of voltage and frequency. If either voltage or frequency exceeds the rated value by ±5%, it is recorded as an over-limit event. If over-limit events occur within three consecutive power frequency cycles (60 ms), the system is determined to be in an off-grid state; otherwise, it remains connected to the grid. This hysteresis logic can prevent false switching caused by short-term peak disturbances.

[0047] After switching to the timer and entering transition mode, a fixed time will elapse. Automatically switch to the target stable mode.

[0048] External commands refer to the grid-connected switch status signals and mode control commands issued by the host computer. These are acquired by the converter through the grid-connected switch status acquisition channel and the host computer communication link, and are used to issue planned mode switching commands. Local electrical quantity detection results refer to the real-time acquisition data of the grid-connected point voltage RMS value, grid frequency, and over-limit judgment results. These are obtained by acquiring grid-connected point electrical signals through voltage and current sensors, and by power calculation and over-limit hysteresis logic judgment for three consecutive power frequency cycles. This is used to identify the current grid-connected or off-grid operation status of the system. Switching timing information refers to the timing duration data of the transient transition process after triggering mode switching. This is obtained by the timing module built into the mode scheduler when entering the transient switching transition mode. It is used to control the fixed transition duration. After the timing ends, the system switches to the target stable operation mode to prevent frequent mode jumps.

[0049] Within the prediction time domain ( (Step), this invention assumes that the pattern label remains unchanged at its current value, that is This assumption is based on the fact that mode switching is usually triggered by slow scheduling commands (on the order of seconds) or large disturbances on the grid side, and the probability of switching again within the prediction time domain of tens of milliseconds is extremely low. For known switching commands, such as planned disconnection from the grid, a feedforward approach can be used: the host computer issues the switching command before... The target mode information is sent out at 50 ms, and the controller pre-sets the prediction time domain starting from this time point. The target mode is used to achieve feedforward transition.

[0050] The cost of mode switching is not borne by the penalty term within the QP, but rather by the logical threshold and switching hysteresis of the mode scheduler itself: when a switching request is detected, the scheduler forces the system into a transition mode. It latches the target mode and switches to it only after the transition time has ended, thus avoiding mode jumps. This design eliminates the need to handle discrete variables when solving QP, preserving the convexity of the optimization problem.

[0051] Step S3: Rolling optimization of the objective function, in QP form.

[0052] Step S3 constructs a rolling optimization objective function in the form of a quadratic programming based on the 7-dimensional composite state vector and the current operating mode label.

[0053] In step S3, the rolling optimization objective function includes three independent cost terms: the first is a weighted squared term of output deviation composed of voltage, frequency, active power, and reactive power deviations; the second is a weighted squared term of control increment; and the third is a weighted squared term of virtual inertia state change. The weighted squared term of virtual inertia state change is used to constrain the gradual change of the virtual inertia trajectory and suppress secondary voltage and frequency oscillations caused by mode switching and sudden load changes. The specific process is as follows: because The objective function does not require a mode transition penalty term. Define the output vector. The output matrix That is, the output only includes four electrical quantities: voltage deviation, frequency deviation, active power deviation, and reactive power deviation. Cost function:

[0054] in, Predict the mode label in the time domain, which is assumed to be equal to the current mode label according to step S2. ; Output vector, which includes four electrical quantities: voltage deviation, frequency deviation, active power deviation, and reactive power deviation; The output matrix maps the state vector to the output vector. ; : 4th order identity matrix; A zero matrix with 3 rows and 3 columns; Predict the time domain (number of steps); Control the time domain (number of steps); Based on the first Information at time 1 The predicted value output at any given time; The smoothed sequence after self-adjustment is used as the target output value, see step S4; Based on the first Information at time 1 The predicted value of the increment is controlled at all times; Based on the first Information at time 1 Predicted value of virtual inertia state at any given time; : No. The value of the rolling optimization cost function at each time step; Output error weight matrix, This is used to adjust the weights of each output error term; : Control the incremental weight matrix, This is used to adjust the weights of each control increment term; For reference to the smoothed sequence after self-shaping, see step S4; This is the penalty coefficient for changes in virtual inertia, with a typical value. The physical meaning of this penalty term: the virtual inertia state during mode switching or sudden load changes. Drastic changes can cause secondary oscillations in the voltage frequency. By constraining the prediction domain... The trajectory changes smoothly, which can effectively suppress transient oscillations and enhance the dynamic stability of grid converters.

[0055] Step S4, refer to online estimation of self-tuning and perturbation.

[0056] Step S4: Generate a smooth reference sequence using an exponential transition function; based on the output residual method and a state observer, estimate the equivalent additive perturbation in the discrete prediction model online, and input the perturbation value as a feedforward quantity into the discrete prediction model. The specific process is as follows: A smooth reference sequence is constructed using an exponential transition function for voltage and frequency deviation targets:

[0057] in, This is the smoothed reference value for the i-th step in the time domain at the current sampling time k. The weighting coefficients of the exponential transition function vary with the number of prediction steps. Exponential decay; The target reference value vector before the switch; This is the target reference value vector after the switch; The attenuation coefficient; For predicting the time-domain incremental step index; For a fixed sampling period.

[0058] in As the attenuation coefficient, its dimensions are its reciprocal This is the transition time constant. In this invention... Pick (correspond This range depends on the sampling period. and prediction step size This ensures the transition process completes within several prediction domains, avoiding both reference abrupt changes and excessively long adjustment lags. For shock-sensitive applications, [further details are needed]. For larger capacity energy storage systems, it is advisable to take Up to 50 ms to smooth load fluctuations.

[0059] Online perturbation estimation uses the output residual method, which requires knowledge of the state estimate from the previous time step. The state observer is designed as follows: Measurable: , , , Obtained through voltage / current sensors and power calculation modules; Unmeasurable: Update via observer with correction: ; in To measure electromagnetic power, , The rated electromagnetic power is obtained by taking the average value within 1 to 2 seconds of grid-connected steady-state operation. The active power deviation component in the state estimate. Observer gain (typical value) Virtual mechanical power The virtual inertia is set as a constant by the host computer based on the rated active power output. Typical values ​​range from 0.1 to 1 s, damping coefficient The typical value range is 0.1 to 2, and the specific value can be adjusted according to the grid strength and the desired dynamic response. This correction term can suppress integral drift.

[0060] Calculate directly based on the first-order recurrence relation in step S1 (based on storage). history).

[0061] After each mode switch, if (Typical threshold value, adjustable according to controller resolution) will then Reset to 0; otherwise, retain the current estimate and continue to converge naturally through the observer to avoid state jumps. Virtual mechanical power is set as a constant by the host computer based on the rated active power output; Measured electromagnetic power; Virtual inertia time constant, typically ranging from 0.1 to 1 s; Virtual damping coefficient, typically ranging from 0.1 to 2; Observer gain, typically 0.1; The model calculates the electromagnetic power, which is equal to the rated electromagnetic power. With active power deviation estimation sum; Rated electromagnetic power is obtained by taking the average value within 1-2 seconds of grid-connected steady-state operation; : Active power deviation component in the state estimate; : Internal state of virtual inertia The estimated value.

[0062] The above numerical integration yields and the calculated Substituting the state vector, we obtain The disturbance estimation formula is: ; This perturbation term is used for feedforward compensation in the prediction model. Wherein, : No. The state estimation vector at time t; : No. Estimated equivalent additive perturbation value at time t.

[0063] Step S5: QP solution and computation timeliness assurance.

[0064] Step S5 takes the discrete prediction model, rolling optimization objective function, smoothed reference sequence and disturbance value as input, completes state recursion and quadratic programming solution in the model prediction time domain, outputs the optimal control increment sequence, and takes the first increment as the basic control quantity.

[0065] In step S5, during the quadratic programming solution, the rolling optimization objective function in the quadratic programming form constructed in step S3 is constrained by multiple types of linear inequalities, including electrical quantity amplitude, control increment, voltage frequency, and virtual inertia. The interior-point method is used as the solution algorithm, and a fixed maximum number of iterations is set to ensure that the overall solution time is less than the fixed sampling period in step S1, thus meeting the real-time operation requirements of the grid converter controller. The specific process is as follows: Control cycle Predicting the time domain Control time domain The state dimension is 7, and the input dimension is 2. There are 7 linear inequalities: 2 magnitude constraints, 2 increment constraints, 2 voltage-frequency constraints, and 1 virtual inertia constraint. The interior-point method is used to solve the QP, supplemented by a fixed maximum number of iterations (30) to ensure bounded worst-case solution time. On a prototype based on the TI TMS320F28379D controller, using single-precision floating-point, optimization level O2, and a maximum of 30 iterations for the interior-point solver, the typical average solution time in 1000 random tests was approximately 78 μs, with a maximum of approximately 96 μs, both less than the 100 μs control cycle. Explicit MPC is considered as an alternative, and the number of critical regions can be obtained offline using the MPT toolbox (Matlab).

[0066] Step S6: Switch the tuning method of the inertial compensation matrix G.

[0067] Step S6 calculates the switching inertia compensation matrix based on the operating modes of the current and previous times; the compensation matrix is ​​then used to perform feedforward correction on the control increment of the previous time step, and the correction result is superimposed on the basic control quantity. The specific process is as follows: like Figure 3 As shown, Figure 3 This is a schematic diagram of the switching inertia compensation logic of the present invention. Figure 3 The logic flow of inertia compensation during switching is explained. When a mode switching request is detected, the compensation matrix is ​​calculated based on the difference between the system matrices of the source mode and the target mode, and the control increment of the previous cycle is corrected to ensure that the control input remains first-order continuous at the switching boundary; if there is no switching, the compensation matrix is ​​zero, and no additional correction is generated.

[0068] For source mode To target mode Switching ( ), compensation matrix The purpose is to ensure that the control input remains first-order continuous at the switching moment. The operating mode at the current moment and the previous moment essentially refers to the tag number corresponding to the operating mode that is effective at each of the two moments, thereby determining the source mode and target mode of the switching operation and then calculating the switching inertia compensation matrix. This invention proposes the following tuning formula (using standard Tikhonov regularization): ; This formula can be viewed as a solution in the least squares sense. This minimizes the sum of squared state prediction errors from the source mode to the target mode at the instant of switching, i.e., minimizes... Regularization term To avoid Numerical ill-conditioned when approaching singularity. Parameters This is an adjustable coefficient used to adjust the compensation intensity. When (When running in the same mode) .

[0069] in, The source mode number for the switching operation; Number the target mode for the switching operation; The adjustable coefficient for switching the inertial compensation intensity; The system matrix for the target mode; The system matrix for the source mode; The control input matrix for the source mode; I is the regularization parameter; I is the AND matrix. An identity matrix of the same dimension.

[0070] In engineering implementation, to reduce the complexity of online matrix operations, the above analytical solution can be approximated as a compensation term in the same direction as the input matrix, i.e. This is equivalent to applying proportional compensation in the original input channel direction. Recommendations for common switching paths. The values ​​are shown in the table below.

[0071] Table 1

[0072] Step S7, multi-timescale coordinated processing (dual-rate predictive control).

[0073] Step S7: Combining the model's prediction of state recursion in the time domain, the complete system matrix and the frozen matrix are called to perform dual-rate predictive control, and the state is updated alternately according to a preset period and the results are sent back to step S5 to adapt to the slow time-varying conditions of the aqueous organic flow battery.

[0074] In step S7, the specific execution rules for dual-rate predictive control are as follows: using the fixed sampling period of step S1 as the timing reference, the dual-rate period multiple is set to 10; every 10 fixed sampling periods, the complete system matrix is ​​called to complete the update of all state variables; in other regular sampling periods, the frozen matrix is ​​called to perform state prediction calculations; the frozen matrix is ​​obtained by modifying the complete system matrix generated in step S1, specifically by replacing the matrix rows corresponding to the virtual inertia state and the dq-axis switching memory continuous state in the complete system matrix with standard basis vectors, so that the two types of slow dynamic state variables, the virtual inertia state and the dq-axis switching memory continuous state, remain numerically constant during the state prediction process, adapting to the inherent slow time-varying operating conditions of the aqueous organic flow battery. The specific process is as follows: To handle electrical quantities (fast dynamics, time constant) ) and the slowly varying characteristics and virtual inertia (slow dynamics, time constant) of flow batteries To address the time scale differences between the two rates, this invention employs dual-rate predictive control. Two prediction matrices are defined: Complete matrix : Includes all state coupling; Frozen Matrix :exist Based on this, rows 5, 6, and 7 (corresponding to) , , Set as a unit row vector, so that the corresponding slow variable follows the same pattern in the prediction domain. The form retains the values ​​from the previous time step and is no longer updated on a fast time scale with other states and inputs, which is equivalent to freezing these three states within the prediction. Specifically: ; in: : The prediction matrix used to freeze slow variables in dual-rate predictive control, with rows 5 to 7 being standard basis vectors; Complete matrix The first 4 rows (corresponding to electrical quantity status); , , It is a 7-dimensional standard basis vector (the 5th, 6th, and 7th elements are 1, and the rest are 0). Slow variables retain the constraint form of their previous value within the prediction domain, and are used to freeze the variable. Modulo operation, returns Divide by The remainder is used to determine whether the current time is... Integer multiples of; : The period multiple for slow variable updates in dual-rate control, with a value of 10.

[0075] Control process: Each fast cycle ( All of them perform a complete QP solution.

[0076] State prediction uses For 9 consecutive steps (with the slow variable remaining constant), use it once every 10th step (i.e., every 1ms). Update slow variables.

[0077] The logic for determining the update time: If If the matrix is ​​complete, use the full matrix; otherwise, use the frozen matrix.

[0078] For the state update according to a preset period, it means that the fixed sampling period in step S1 is used as the time reference, and 10 times the fixed sampling period is used as the preset slow variable update period. The frozen matrix and the complete system matrix are alternately called to perform state prediction recursion. The state refers to the 7-dimensional composite state vector defined in step S1, which includes four types of fast dynamic electrical quantity states: voltage deviation, frequency deviation, active power deviation, and reactive power deviation, as well as two types of slow dynamic states: virtual inertia state and dq axis switching memory continuous state. During the normal sampling period, the frozen matrix is ​​called to keep the slow dynamic state constant and only the fast dynamic state participates in the recursive update. Every 10 sampling periods, the complete system matrix is ​​called to complete the full update of all state quantities. The two are executed alternately to adapt to the inherent slow time-varying operating conditions of aqueous organic flow batteries.

[0079] Example 2, as Figure 4 As shown, Figure 4 This is a flowchart of steps S1-S8 of a grid-connected converter coordinated control method for an aqueous organic liquid flow energy storage battery according to Embodiment 2 of the present invention. Based on the method of Embodiment 1, the method of Embodiment 2 further includes the following steps: Step S8: Communication architecture and degradation strategy.

[0080] In step S8, when multiple grid-connected converters are running in parallel, each device uses peer-to-peer distributed communication to exchange active power and frequency reference data. The average value of all device data is used as a common reference frequency, which is then used for the processing flow in steps S2 and S4. When a communication anomaly is detected, the operating mode is downgraded, and each parallel grid-connected converter automatically switches to an independent single-unit operating mode and enables a fixed reference frequency. In single-unit mode, all control processes from steps S1 to S7 are continuously executed to achieve overall coordinated operation of multiple units.

[0081] In step S8, the peer-to-peer distributed communication is set to a fixed communication period of 10ms. Each parallel-connected converter in the network periodically broadcasts its own active power and frequency reference data, and the average value of all parallel unit data is calculated as the common reference frequency. When the communication interruption lasts for more than 30ms, all parallel-connected converters in the network synchronously switch to independent single-machine operation mode and enable the fixed reference frequency. After the communication link is restored, each converter re-establishes peer-to-peer distributed communication, re-synchronizes the calculation of the common reference frequency, and switches back to multi-machine parallel coordinated operation mode. The specific process is as follows: When multiple converters are connected in parallel, to complement the rolling optimization framework of this invention, peer-to-peer distributed communication is preferably adopted. Each converter broadcasts its own active power and current frequency reference value via industrial Ethernet at 10 ms intervals. All units receive information from other units and calculate the average value as a common reference frequency. This architecture does not require a central controller. If the communication interruption exceeds 30 ms, each converter automatically switches to independent single-unit mode. =1 or 2 (determined by local voltage and frequency), and restores the fixed reference frequency, then resynchronizes after communication is restored. This communication and degradation strategy is only an optional implementation method to support multi-machine coordinated operation.

[0082] Example 3: Standalone grid-connected / off-grid switching. This example is based on the content of Examples 1 and 2.

[0083] Example of state space matrix construction (Mode 1, grid-connected droop).

[0084] Taking a system with a rated power of 100kW, DC side voltage of 650V, and AC side voltage of 380V / 50Hz as an example, the LCL filter parameters are: inverter side inductance. Grid-side inductor Filter capacitor Transformer equivalent resistance A continuous-time state equation is established in the dq coordinate system, and the electrical submatrix is ​​obtained after linearization. The zero-order hold method is adopted. Discretization yields After adding the power deviation and virtual inertia state, the discrete state matrix in Mode 1 (only the first 4 rows and 5 columns are shown; the complete 7×7 matrix is ​​implemented in the actual code) is as follows (values ​​are rounded):

[0085] Among them, the 5th line The corresponding update (close to 1, indicating slow change), lines 6 and 7. , Corresponding updates (including exponential decay coefficient) (and diagonal input).

[0086] Standalone network-to-grid / offline switching: System parameters are the same as above. Controller: TI TMS320F28379D. , , , , , Mode scheduling: Grid connection to When the off-network instruction is received, the scheduler is set Latch target mode 2, transition time 200ms. Reference self-tuning: Since the output is a deviation, the target deviation is 0 in grid-connected steady state, and the off-grid target is a 5% voltage increase and a frequency maintained at 50Hz (deviation 0). Therefore, it is set... , (Voltage deviation target +0.05 pu, frequency deviation 0). . .use (Right now Multiply by the input matrix of the current pattern.

[0087] in, The proportional compensation coefficient under the simplified engineering approximation is determined by looking up a table according to the switching path; : The switching inertia compensation matrix used when switching from mode 1 to mode 4; : Input matrix under mode 1 (grid-connected droop mode).

[0088] Experimental results: During the switching process, the maximum effective voltage fluctuation was 4.1%, the maximum frequency shift was 0.18Hz, and the peak current was 1.48 times the rated value; the traditional method (without G matrix, without reference shaping) resulted in a voltage fluctuation of 11.2% and a peak current of 2.3 times. The QP solution took an average of 78μs on the prototype, with a worst-case time of 96μs (convergence within 30 iterations, tested 1000 times). This verifies the ability of the online perturbation estimation mechanism of this invention to compensate for the natural drift of the flow battery; under changing electrolyte parameters, this mechanism can further compensate for the equivalent output impedance shift caused by gradually changing operating conditions.

[0089] Example 4: Multi-machine parallel circulating current suppression. This example is based on the content of Examples 1 and 2.

[0090] The example of establishing the state space matrix used in this embodiment (Mode 1, grid-connected droop) is shown in Embodiment 3.

[0091] Two 100kW converters are connected in parallel, with a line impedance difference. The load is 150kW. Peer-to-peer distributed communication (10ms period) is used, with each machine operating... When the load suddenly increases by 30%, the peak circulating current drops from the traditional drooping 8.2% to 2.1%, and the steady-state circulating current is <1%. Communication interruption test: Within 30ms after the interruption, each machine automatically switches to... The reference frequency returned to 50Hz, the load power was shared by the remaining units, and no overcurrent protection action occurred.

[0092] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

[0093] The above embodiments are merely illustrative examples and are not intended to limit the implementation. Those skilled in the art will recognize that other variations or modifications can be made based on the above description. It is neither necessary nor possible to exhaustively list all possible implementations. However, obvious variations or modifications derived therefrom are still within the scope of protection of this invention.

Claims

1. A grid-connected converter coordinated control method for an aqueous organic liquid flow energy storage battery, characterized in that, Includes the following: Step S1: Construct a state-space model of the coupled grid converter and the water-based organic liquid flow energy storage battery, and construct a 7-dimensional composite state vector; divide the operation mode into four categories of labels, with the operation mode labels being independent of the composite state vector; combine the state-space model with a fixed sampling period to establish a discrete prediction model, and generate a complete system matrix and a frozen matrix. Step S2: Determine the current operating mode label based on external instructions, local electrical quantity detection results, and switching timing information; Within the model prediction time domain of the discrete prediction model, the operating mode label remains constant; Step S3: Based on the 7-dimensional composite state vector and the current operating mode label, construct a rolling optimization objective function in the form of quadratic programming; Step S4: Generate a smooth reference sequence using an exponential transition function; Based on the output residual method and state observer, the equivalent additive perturbation in the discrete prediction model is estimated online, and the perturbation value is used as a feedforward quantity into the discrete prediction model. Step S5: Using the discrete prediction model, rolling optimization objective function, smoothed reference sequence and disturbance value as input, complete the state recursion and quadratic programming solution in the model prediction time domain, output the optimal control increment sequence, and take the first increment as the basic control quantity. Step S6: Calculate the switching inertia compensation matrix based on the operating modes of the current time and the previous time; use the compensation matrix to perform feedforward correction on the control increment of the previous time step, and superimpose the correction result into the basic control quantity; Step S7: Combining the model's prediction of state recursion in the time domain, the complete system matrix and the frozen matrix are called to perform dual-rate predictive control, and the state is updated alternately according to a preset period and the results are sent back to step S5 to adapt to the slow time-varying conditions of the aqueous organic flow battery.

2. The grid-connected converter coordinated control method for an aqueous organic liquid flow energy storage battery according to claim 1, characterized in that, The method further includes the following steps: In step S8, when multiple grid-connected converters are running in parallel, each device uses peer-to-peer distributed communication to exchange active power and frequency reference data. The average value of all device data is used as a common reference frequency, which is then used for the processing flow in steps S2 and S4. When a communication anomaly is detected, the operating mode is downgraded, and each parallel grid-connected converter automatically switches to an independent single-unit operating mode and enables a fixed reference frequency. In single-unit mode, all control processes from steps S1 to S7 are continuously executed to achieve overall coordinated operation of multiple units.

3. The grid-connected converter coordinated control method for an aqueous organic liquid flow energy storage battery according to claim 1, characterized in that, In step S1, the 7-dimensional composite state vector includes voltage deviation, frequency deviation, active power deviation, reactive power deviation, virtual inertia state, and dq-axis switching memory continuous state. The calculation formula for the dq-axis switching memory continuous state is as follows: ； in, To switch the memory continuous state for the d-axis, at time k; The q-axis is switched to a continuous state of memory at time k; To switch the memory continuous state for the d-axis, at time k-1; The memory continuous state is switched for the q-axis at time k-1; The exponential decay coefficient; For a fixed sampling period; It is a natural constant; The d-axis control increment at time k-1; The increment for q-axis control at time k-1.

4. The grid-connected converter coordinated control method for an aqueous organic liquid flow energy storage battery according to claim 1, characterized in that, In step S2, the logic for determining the current operating mode label based on the local electrical quantity detection results and switching timing information is as follows: The effective voltage value and grid frequency at the grid connection point of the grid-connected converter are continuously monitored. If any parameter exceeds ±5% of the corresponding rated value, it is judged as an over-limit event. If an over-limit event occurs for three consecutive power frequency cycles, the system is determined to be in an off-grid state, and the corresponding current operating mode label is off-grid virtual synchronous machine mode; otherwise, the system is determined to be in a grid-connected state, and the corresponding current operating mode label is grid-connected droop mode. If the newly determined grid-connected or off-grid status is inconsistent with the currently effective operating mode label, a mode switch is triggered, and the current operating mode label is switched to a transient switching transition mode. A fixed transition duration is timed according to the switching timing information in step S2. After the timing ends, the system automatically switches to the target stable operating mode label to prevent frequent mode jumps.

5. The grid-connected converter coordinated control method for an aqueous organic liquid flow energy storage battery according to claim 1, characterized in that, In step S3, the rolling optimization objective function includes three types of independent cost terms: the first type is the output deviation weighted square term composed of voltage, frequency, active power, and reactive power deviations; the second type is the control increment weighted square term; and the third type is the virtual inertia state change weighted square term. The virtual inertia state change weighted square term is used to constrain the smooth change of the virtual inertia trajectory and suppress the voltage and frequency secondary oscillations caused by mode switching and load sudden changes.

6. The grid-connected converter coordinated control method for an aqueous organic liquid flow energy storage battery according to claim 1, characterized in that, In step S4, an exponential transition function is used to perform self-shaping on the target reference value to generate a smooth reference sequence. The calculation formula for the function used in the self-shaping process is as follows: in, This is the smoothed reference value for the i-th step in the time domain at the current sampling time k. The weighting coefficients of the exponential transition function vary with the number of prediction steps. Exponential decay; The target reference value vector before the switch; This is the target reference value vector after the switch; The attenuation coefficient; For predicting the time-domain incremental step index; For a fixed sampling period.

7. The grid-connected converter coordinated control method for an aqueous organic liquid flow energy storage battery according to claim 1, characterized in that, In step S5, when solving the quadratic programming problem, multiple linear inequality constraints are configured for the rolling optimization objective function in the form of quadratic programming constructed in step S3, including electrical quantity amplitude, control increment, voltage frequency, and virtual inertia. The interior point method is used as the solution algorithm, and a fixed maximum number of iterations is set to ensure that the overall solution time is less than the fixed sampling period in step S1, thus meeting the real-time operation requirements of the controller of the grid converter.

8. The grid-connected converter coordinated control method for an aqueous organic liquid flow energy storage battery according to claim 1, characterized in that, In step S6, the formula for obtaining the switching inertia compensation matrix is: ； in, The source mode number for the switching operation; Number the target mode for the switching operation; The adjustable coefficient for switching the inertial compensation intensity; The system matrix for the target mode; The system matrix for the source mode; The control input matrix for the source mode; I is the regularization parameter; I is the AND matrix. An identity matrix of the same dimension.

9. The grid-connected converter coordinated control method for an aqueous organic liquid flow energy storage battery according to claim 1, characterized in that, In step S7, the specific execution rules of dual-rate predictive control are as follows: taking the fixed sampling period of step S1 as the timing reference, the dual-rate period multiple is set to 10; every 10 fixed sampling periods, the complete system matrix is ​​called to complete the update of all state variables; in other regular sampling periods, the frozen matrix is ​​called to carry out state prediction calculation; the frozen matrix is ​​obtained by modifying the complete system matrix generated in step S1. The specific modification method is: replacing the matrix rows corresponding to the virtual inertia state and the dq axis switching memory continuous state in the complete system matrix with standard basis vectors, so that the two types of slow dynamic state variables, the virtual inertia state and the dq axis switching memory continuous state, remain numerically constant during the state prediction process, adapting to the inherent slow time-varying operating conditions of the aqueous organic flow battery.

10. A grid-connected converter coordinated control method for an aqueous organic liquid flow energy storage battery according to claim 2, characterized in that, In step S8, the peer-to-peer distributed communication is set to a fixed communication period of 10ms; each parallel network converter in the network periodically broadcasts its own active power and frequency reference data, and the average value of all parallel unit data is taken as the common reference frequency; when the communication interruption time exceeds 30ms, all parallel network converters in the network synchronously switch to independent single-machine operation mode and enable the fixed reference frequency. Once the communication link is restored to normal, each converter will re-establish peer-to-peer distributed communication, re-synchronize and calculate the common reference frequency, and switch back to the multi-machine parallel coordinated operation mode.

Images