Energy storage black box model parameter identification method based on double sensitivity

Abstract

The invention provides an energy storage black box model parameter identification method based on double sensitivity, and relates to the field of new energy power generation control. Comprising the following steps: S1, building a main circuit model by utilizing an energy storage electromechanical transient model, setting a parameter range, linearly scanning each parameter, obtaining simulation data under different parameter values, and forming a test data set; s2, acquiring semi-physical simulation data of the stored energy under a low-voltage ride-through working condition, and calculating an average relative error between the test data set and the reference data set; s3, performing first sensitivity analysis based on the test data set, and screening out to-be-identified parameters; s4, performing second sensitivity analysis based on the average relative error, and determining the sensitivity of the parameter to the final identification error according to the screened parameter to be identified; s5, parameter identification is carried out by adopting a particle swarm optimization algorithm, global search and local optimization are carried out, and identified energy storage black box model parameters are obtained; and the identification precision and the calculation efficiency are improved.

Description

Technical Field

[0001] This invention relates to the field of new energy power generation control, and in particular to a parameter identification method for a dual-sensitivity energy storage black box model. Background Technology

[0002] Currently, new energy storage systems, represented by energy storage systems, are being widely used in power systems. Energy storage systems play a crucial role in improving the absorption of new energy sources and ensuring the safe and stable operation of the power grid. However, energy storage systems typically involve the coupling of multiple physical processes, including electrochemistry, power electronics, and complex control strategies. Furthermore, manufacturers often do not disclose their core control logic, resulting in a highly "black box" characteristic, making it difficult to obtain precise mechanistic models.

[0003] In current electromechanical transient simulation analysis of power systems, simplified energy storage equivalent models are typically used. The accuracy of such models is highly dependent on the values ​​of their key control parameters. If these parameters are set improperly, the model will fail to accurately reflect the dynamic response behavior of energy storage under critical operating conditions such as fault ride-through, thereby affecting the accuracy of the assessment of the safety and stability of power systems containing energy storage.

[0004] Currently, the main approach to obtaining energy storage model parameters is parameter identification based on measured operational data. However, energy storage systems exhibit nonlinear and multi-timescale dynamic characteristics. Traditional parameter identification methods generally suffer from the following limitations when dealing with such "black box" models: 1. They struggle to effectively handle strongly coupled and nonlinear relationships within the model, resulting in weak generalization ability for different operating conditions (especially transient processes such as low-voltage ride-through). 2. Due to the high dimensionality of the parameter space and the presence of numerous redundant or weakly sensitive parameters, traditional methods are prone to getting trapped in local optima, making it difficult to obtain globally accurate parameter solutions. 3. They require extensive blind or semi-blind simulation calculations to traverse or search the vast parameter space, leading to a time-consuming and lengthy identification process. Summary of the Invention

[0005] The main objective of this invention is to provide a parameter identification method for a black box model of energy storage based on dual sensitivity, which solves the problems of insufficient adaptability, low computational efficiency, and low identification accuracy in the prior art.

[0006] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is: a parameter identification method for an energy storage black box model based on dual sensitivity, comprising the following steps: S1: Build the main circuit model using the existing energy storage electromechanical transient model in PSASP, and set the parameter range according to the initial values ​​given by the system in the energy storage electromechanical model; based on the parameter range, perform a linear scan on each parameter, changing only one parameter at a time while keeping other parameters unchanged, to obtain simulation data under different parameter values, and form a test dataset; S2: Obtain hardware-in-the-loop simulation data of energy storage under low voltage ride-through conditions, process the data and use it as a reference dataset, and calculate the average relative error between the test dataset and the reference dataset. S3: Based on the test dataset, perform a first sensitivity analysis, calculate the sensitivity of each scanned parameter change to the simulation output data, classify the parameters according to the sensitivity results, and select the parameters to be identified. S4: Based on the average relative error, a second sensitivity analysis is performed. For the selected parameter to be identified, the trend of the minimum average relative error that can be obtained by randomly changing other parameters when the parameter is fixed at different values ​​is analyzed to determine the sensitivity of the parameter to the final identification error. S5: The particle swarm optimization algorithm is used for parameter identification. The weighted sum of the average relative errors is defined as the objective function, and a penalty function term based on the parameter variation range is introduced into the objective function. Based on the results of the second sensitivity analysis, the particle swarm algorithm is guided to perform global search and local optimization in the preferred subspace of key parameters until the convergence condition is met, and the identified energy storage black box model parameters are obtained.

[0007] In the preferred embodiment, step S1 specifically includes the following steps: Based on the default initial parameters, ignoring the dynamics of the DC voltage side and considering only the dynamics of the output side, the expression for the preset parameter variation range symmetrically around zero is: ; In the formula, -5a is the lower limit of the parameter range, and 5a is the upper limit of the parameter range; A single parameter changes from the lower limit of its range to the upper limit by a preset magnitude, while the other parameters remain unchanged; After the current parameter is scanned, the initial value is reset and the next parameter is scanned. Physical quantity data under different scan values ​​of each parameter are obtained as test datasets.

[0008] In a preferred embodiment, step S2 includes the following steps: The low-voltage ride-through fault condition of energy storage is set up, and the step size of the hardware-in-the-loop simulation is consistent with that of the power system simulation software. A reference dataset is obtained after data processing. The average relative error is calculated for both steady-state and dynamic sampling points, using the relative error between steady-state test data and steady-state reference data, and between dynamic test data and dynamic reference data, respectively. The formula is as follows: ; In the formula, W The average relative error, N The number of discrete sampling points. O s ( t )andO v ( t () represent steady-state and dynamic test data, respectively. O ms ( t )and O mv ( t () represent steady-state and dynamic reference data, respectively. S This represents the number of steady-state sampling points. V Indicates the number of dynamic sampling points; where: .

[0009] In the preferred embodiment, step S3 includes: The first sensitivity analysis uses the average relative sensitivity, and the formula is: ; In the formula, The parameter is Test data at that time The parameter is Test data at that time, and Similar to the scan value step size, it is the difference between the upper and lower limits of the parameter divided by the number of scan points.

[0010] In the preferred embodiment, based on the results of the first sensitivity analysis, the parameters are classified as follows: Irrelevant parameters that have no effect on the output data throughout the entire scan range; The limiting threshold parameter of the control logic only affects the value when it is within a specific range; it is a steady-state parameter that mainly affects the steady-state output data. And the dynamic parameters that mainly affect the output data of the transient process; The amplitude limiting threshold parameter is set to a fixed value according to the actual logic of the controller, and the steady-state parameter and dynamic parameter are used as parameters to be identified.

[0011] In the preferred embodiment, the second sensitivity analysis in step S4 includes: For the set of parameters to be identified θ ={ θ 1, ..., θ n In}, any parameter Within its range of variation Take multiple fixed values; For each fixed value ， In the remaining parameters Perform multiple random sampling combinations within the initial value range, calculate the average relative error for each combination, and take the minimum value. ; analyze The curves showing the variation of the minimum error value for different fixed values, and the minimum point on the curves corresponding to these curves. The value is considered its theoretical identification value, and the parameters are calculated based on this curve. Sensitivity The formula is: ; in, Indicates parameters Sensitivity to model output Is with The parameter value corresponding to the lowest point on the [the graph]. and The closest On Two values.

[0012] In the preferred embodiment, in step S5, the objective function is defined as: ; in, and These represent the average relative errors of voltage, current, active power, and reactive power, respectively. and These are the weighting coefficients for each physical quantity. For the first i The penalty weight coefficients for each parameter, Let it be the penalty function; The expression is: ; in, , These are the upper and lower limits of the parameter range, respectively. It is the midpoint of the parameter range. This represents the range of parameter variations.

[0013] In the preferred embodiment, the parameter identification using the particle swarm optimization algorithm includes: Initialize the particle swarm, in 3D target space generation A population of particles, with particle initialization positions determined by the upper and lower boundaries of the target space. Sure; Iterative updates of particle position and velocity, using the following formula: ; ; In the formula, It is the first i The speed of movement of each particle It is the firsti The current position of each particle As a learning factor, For the first i The optimal solution for each particle. It is the optimal solution for the entire population; Inertial weight; for Random numbers within an interval.

[0014] In the preferred embodiment, the inertial weight Using a linear decreasing strategy, the calculation formula is as follows: ; In the formula, It is the initial inertia weight. It is the inertia weight when it is iterated to the maximum.

[0015] In the preferred embodiment, S6 is further included: substituting the identified parameters into the energy storage electromechanical transient model for simulation to obtain the simulation response curve of the key electrical quantities, and comparing it with the response curve generated by the reference dataset to verify the accuracy of the identification results.

[0016] This invention provides a dual-sensitivity-based parameter identification method for energy storage black-box models. The method comprises the following steps: S1: Constructing a main circuit model using an energy storage electromechanical transient model, setting parameter ranges, and performing linear scanning on each parameter to obtain simulation data under different parameter values, forming a test dataset; S2: Acquiring semi-physical simulation data of energy storage under low-voltage ride-through conditions and calculating the average relative error between the test dataset and a reference dataset; S3: Performing a first sensitivity analysis based on the test dataset to filter out parameters to be identified; S4: Performing a second sensitivity analysis based on the average relative error to determine the sensitivity of the selected parameters to the final identification error; S5: Using a particle swarm optimization algorithm for parameter identification, performing global search and local optimization to obtain the identified parameters of the energy storage black-box model. This method achieves systematic and automated identification of parameters for the "black box" model of energy storage systems, without relying on internal model mechanism knowledge. It achieves high-precision parameter fitting solely through external input-output data, improving identification efficiency and adaptability, enhancing identification accuracy and efficiency, effectively avoiding local optima, and improving identification robustness. Attached Figure Description

[0017] The present invention will be further described below with reference to the accompanying drawings and embodiments: Figure 1 This is a flowchart of the identification method of the present invention; Figure 2 This is a flowchart of the sensitivity and particle swarm algorithm parameter identification steps of the present invention; Figure 3This is a sensitivity result diagram of energy storage parameters and voltage in an embodiment of the present invention; Figure 4 This is a graph showing the sensitivity results of energy storage parameters and current in an embodiment of the present invention. Figure 5 This is a sensitivity result diagram of energy storage parameters and active power in an embodiment of the present invention; Figure 6 This is a sensitivity result diagram of energy storage parameters and reactive power in an embodiment of the present invention; Figure 7 This is a sensitivity result diagram of the key energy storage parameters and average relative error in the embodiments of the present invention; Figure 8 This is a comparison chart of simulation results and test results in the embodiments of the present invention. Detailed Implementation

[0018] Example 1 like Figure 1-8 As shown, a parameter identification method for an energy storage black box model based on dual sensitivity includes the following steps: S1: Build the main circuit model using the existing energy storage electromechanical transient model in PSASP, and set the parameter range according to the initial values ​​given by the system in the energy storage electromechanical model; based on the parameter range, perform a linear scan on each parameter, changing only one parameter at a time while keeping other parameters unchanged, to obtain simulation data under different parameter values, and form a test dataset.

[0019] S2: Obtain hardware-in-the-loop simulation data of energy storage under low voltage ride-through conditions, process the data and use it as a reference dataset, and calculate the average relative error between the test dataset and the reference dataset.

[0020] S3: Based on the test dataset, perform the first sensitivity analysis, calculate the sensitivity of each scanned parameter to the simulation output data, classify the parameters according to the sensitivity results, and select the parameters to be identified. S4: Based on the average relative error, a second sensitivity analysis is performed. For the selected parameter to be identified, the trend of the minimum average relative error that can be obtained by randomly changing other parameters when the parameter is fixed at different values ​​is analyzed to determine the sensitivity of the parameter to the final identification error.

[0021] S5: The particle swarm optimization algorithm is used for parameter identification. The weighted sum of the average relative error is defined as the objective function, and a penalty function term based on the parameter variation range is introduced into the objective function. Based on the results of the second sensitivity analysis, the particle swarm algorithm is guided to perform global search and local optimization in the optimal subspace of key parameters until the convergence condition is met, and the identified parameters of the energy storage black box model are obtained.

[0022] S6: Substitute the identified parameters into the energy storage electromechanical transient model for simulation to obtain the simulation response curves of key electrical quantities. Compare these curves with the response curves generated by the reference dataset to verify the accuracy of the identification results.

[0023] In this embodiment, the main circuit model is built using the existing energy storage electromechanical transient model in PSASP (Power System Analysis Software Package). Simulation data and semi-physical simulation data are used to form test and reference datasets. Sensitivity analysis is performed on both parameter-simulation data and parameter-error parameters. Particle swarm optimization is used for parameter identification. Combined with the objective function, the identified parameters of the energy storage black-box model are obtained. Finally, simulation comparison is performed to verify the accuracy of the identification results. This achieves systematic and automated identification of parameters for the energy storage system black-box model. It does not rely on the internal mechanism knowledge of the model and can complete high-precision parameter fitting only through external input-output data, improving identification efficiency and adaptability, enhancing identification accuracy and efficiency, effectively avoiding local optima, and improving identification robustness.

[0024] In the preferred embodiment, S1 specifically includes the following steps: S1.1: Based on the existing electromechanical transient simulation model in PSASP, the default initial parameters in PSASP are obtained, as shown in Table 1. The dynamic response process of the DC voltage side is ignored, and only the dynamic characteristics of the output side are considered.

[0025] S1.2: Based on the default initial parameters, and with zero as the center, a preset parameter variation range is established. In this embodiment, a parameter variation range of ±500% of the initial value is constructed, expressed as: ; In the formula, -5a is the lower limit of the parameter range, and 5a is the upper limit of the parameter range. The other parameters are also set according to the current parameter range.

[0026] S1.3: A single parameter changes from the lower limit of its range to the upper limit by 5%, while the other parameters remain unchanged. After scanning the current parameter, the initial value is reset, and the scanning continues for the next parameter. Physical quantity data under different scan values ​​for each parameter are obtained as the test dataset.

[0027] Table 1 Initial Parameters and Parameter Ranges

[0028] In the preferred embodiment, step S2 includes the following steps: S2.1: Set the low-voltage ride-through fault condition for energy storage. The step size of the hardware-in-the-loop simulation is consistent with that of the power system simulation software. The reference dataset is obtained after data processing. S2.2: Mean relative error. For steady-state sampling points and dynamic sampling points, the relative error between steady-state test data and steady-state reference data, and the relative error between dynamic test data and dynamic reference data are calculated respectively. The formula is: ; In the formula, W The average relative error, N The number of discrete sampling points. O s ( t )and O v ( t () represent steady-state and dynamic test data, respectively. O ms ( t )and O mv ( t () represent steady-state and dynamic reference data, respectively. S This represents the number of steady-state sampling points. V Indicates the number of dynamic sampling points; where: .

[0029] In the preferred embodiment, step S3 includes: The first sensitivity analysis is a parameter-data sensitivity analysis, performed based on the test dataset and the scanned parameters, using the average relative sensitivity method. The formula is: ; In the formula, The parameter is Test data at that time The parameter is Test data at that time, and Similar to the scan value step size, it is the difference between the upper and lower limits of the parameter divided by the number of scan points.

[0030] In the preferred scheme, the parameters are classified according to the results of the first sensitivity analysis and the sensitivity curve results: ① Irrelevant parameters: Parameters that have no effect on the test data throughout the entire scan range.

[0031] ② Limiting threshold parameter: The parameter that triggers limiting control only when the parameter takes a value within a certain range.

[0032] ③ Steady-state parameters: Parameters that affect the output data in steady state.

[0033] ④ Dynamic parameters: Parameters that affect the data when the system undergoes transient behavior.

[0034] Finally, the four influential parameters are the proportional-integral coefficients Kp_deltaP and Ki_deltaP of the active power loop, and Kp_deltaQ and Ki_deltaQ of the reactive power loop, which are represented as parameters 1, 2, 3, and 4 in the figure. Figure 3 The figure shows the sensitivity results of energy storage parameters and voltage, illustrating the sensitivity changes of different energy storage parameters to the terminal voltage within a time range of 0-20 seconds. This visually demonstrates the degree of influence of each parameter on the voltage characteristics. The figure shows that parameters 1 and 2 affect the steady-state voltage after fault recovery and the steady-state voltage during the fault, respectively, with sensitivity coefficients of -0.005 and -0.014. Parameters 3 and 4 only affect the voltage during fault occurrence and recovery; in the transient state, parameter 4 has a greater impact than parameter 3.

[0035] The parameters to be identified, namely parameters 1, 2, 3 and 4, are selected. The average sensitivity coefficients of the four parameters for voltage, current, active power and reactive power at all times are shown in Table 2 below. Table 2. Percentage of Average Sensitivity Coefficients of Parameters Selected for Identification

[0036] like Figure 4 The figure shows the sensitivity results of energy storage parameters and current, illustrating the changes in the sensitivity of relevant energy storage parameters to grid-connected current within the corresponding time interval, including the fluctuation trend of current sensitivity at different time periods. When a fault occurs, the four parameters have relatively small impacts. Parameter 4 has the greatest impact on the transient current during fault recovery, with a sensitivity peak reaching 2.8. The maximum sensitivity values ​​of the other parameters are less than 0.3. Only parameter 1 has an impact on the steady-state state after fault recovery, with a sensitivity coefficient of approximately 0.03.

[0037] like Figure 5 The figure shows the sensitivity results of energy storage parameters and active power, reflecting the sensitivity response of each energy storage parameter to the system's active power within 0-20 seconds, thus clarifying the correlation between parameters and active power characteristics. In the figure, when a fault occurs, parameter 3 has the greatest impact, with a sensitivity peak of -8. Other parameters have relatively small effects; for example, parameter 2 has a sensitivity peak of only 0.2. Parameters 2 and 4 have the greatest impact on the fault recovery transient state, with sensitivity peaks of 6.5 and 6 respectively. Among the parameters' influence on the fault recovery steady state, parameter 1 has an impact on the steady state and exhibits a fluctuating pattern.

[0038] like Figure 6The figure shows the sensitivity results of energy storage parameters and reactive power, illustrating the sensitivity changes of energy storage parameters to reactive power within a specified time range, reflecting the influence of parameters on the dynamic and steady-state characteristics of reactive power. In the figure, all four parameters have relatively small impacts on reactive power transients during fault occurrence, while parameters 2 and 4 have significant impacts during fault recovery, and their influence on steady-state conditions also fluctuates.

[0039] In this embodiment, by setting a large-range parameter scan centered at zero and combining it with a quantitative evaluation based on average relative error and average relative sensitivity, the parameter sensitivity is fully exposed, the optimal solution is not missed due to an inappropriate search range, and the reliability of the data is improved.

[0040] According to the energy storage controller settings, some parameters are fixed, such as the limiting threshold parameter, so that the limiting control is within the starting range. The main circuit parameters are consistent with the settings in the actual controller. The remaining steady-state and static parameters that affect the data are identification parameters.

[0041] In this embodiment, the second sensitivity analysis is a parameter-error sensitivity analysis.

[0042] In the preferred embodiment, in S4, the second sensitivity analysis includes: S4.1: After selecting the identification parameters based on parameter-data sensitivity, for the set of parameters to be identified... θ ={ θ 1, ..., θ n}middle, It is a set θ The key parameters to be analyzed are within the sensitivity range obtained by subtracting the sensitivity of the parameters from the data. For any parameter... Within its range of variation Take multiple fixed values; For each fixed value ， In the remaining parameters Perform multiple random sampling combinations within the initial value range: θ −j ={ θ 1, ..., θ j-1 , θ j+1 , ..., θ n} Calculate the average relative error for each combination and take the minimum value. The total solution space of the model parameters is defined as follows: .

[0043] Will Set to a fixed value Other parameters It is randomly selected from within its initial value range. Then, it is possible to obtain the current value. The set of numerical values ​​of error when changing: ; in, express The k Group of random values, m yes The total number of random combinations. When m When large enough, It can represent In plane θ j = θ j0 The projection on the surface. For parameters Optimized subspace.

[0044] In actual parameter identification, when Fixed as At that time, the optimization algorithm is most likely to identify The value will be the optimized subspace The global minimum value in is denoted as . In order to comprehensively analyze the parameters Changes and objective function The relationship, when When changes occur, more optimization subspaces need to be computed. Therefore, the range of initial values ​​is further expanded. Dividing by the equal intervals yields the result. λ +1 interval Endpoint value: ; Combining the two formulas above, we can obtain other optimization subspaces. Then, by selecting the minimum value from all optimization subspaces, the average relative error can be expressed as: ; Where min(·) represents the minimum value in the optimal subspace. The point on represents when The objective function of the model when fixed for different values J ( θ The minimum value of ). When When there are enough random samples, it is possible to simulate the simultaneous identification of multiple parameters. The optimization process. The value of the average relative error is usually used to quantitatively represent the accuracy of parameter identification, therefore The minimum value on can indicate The theoretical identification value.

[0045] S4.2: Unlike the parameter-data sensitivity method above, which classifies parameters and selects identification parameters, the parameter-error sensitivity method is used to evaluate the impact of model parameters on the observed output and is related to the reference data.

[0046] analyze The curves showing the variation of the minimum error value for different fixed values, according to S4.1 The theoretical identification value is obtained, and parameters are calculated based on the curve. Sensitivity The formula is: ; in, Indicates parameters Sensitivity to model output Is with The parameter value corresponding to the lowest point on the [the graph]. and The closest On Two values. Since the parameters change from small to large, we choose one here to improve recognition efficiency. .

[0047] like Figure 7 As shown, the sensitivity results of key energy storage parameters and average relative error are plotted, i.e., parameter-error sensitivity curves. This curve shows the sensitivity fluctuation of the model's average relative error as the key parameter values ​​change, which can be used to locate the theoretical value of the key parameter that minimizes the error.

[0048] In this embodiment, the optimization objective is to minimize the error.

[0049] In the preferred embodiment, in step S5, the objective function is defined as follows: ; in, and These represent the average relative errors of voltage, current, active power, and reactive power, respectively. and These are the weighting coefficients for each physical quantity. The weights can be set according to the needs of observing the physical quantity under different disturbances. For example, when observing the voltage drop during a low-voltage breakdown, the voltage weight can be set relatively large. To ensure that the parameters are identified within their respective subspaces, As a penalty item, For the first i The penalty weight coefficients for each parameter, Let be the penalty function.

[0050] The expression is: ; in, , These are the upper and lower limits of the parameter range, respectively. It is the midpoint of the parameter range. This represents the range of parameter variations.

[0051] In this embodiment, an objective function combining weighted error of multiple physical quantities and parameter out-of-bounds penalty is constructed. This allows for flexible adjustment of weights based on the identification focus, such as emphasizing voltage recovery characteristics during low-voltage breakdown, thereby enhancing the adaptability of the algorithm. Furthermore, the search process is constrained by the penalty function mechanism, ensuring that the identification results are within a physically reasonable parameter range, thus improving the robustness of the algorithm and the engineering practicality of the solution.

[0052] In this embodiment, particles move randomly using a specific algorithm, and their movement quality is evaluated using a fitness function. Therefore, particle swarm optimization algorithm is used for parameter identification.

[0053] In the preferred solution, such as Figure 2 As shown, in step 5, particle swarm optimization algorithm is also used for parameter identification, including: S5.1: Initialize the particle swarm, in Randomly generated target space A population of particles, whose initial positions are determined by the target, should be represented as follows: A vector of dimension 1, the first dimension i Each particle is represented as: .

[0054] Assume the upper and lower boundaries of the target space are defined by... Indicate, then The initialization location is expressed as: .

[0055] S5.2: The particle swarm generates random solutions during the initialization phase, and then finds the optimal solution through iteration. In each iteration, particles update themselves by tracking their own experience and the collective experience, iteratively updating the particle's position and velocity using the following formula: ; ; ; In the formula, It is the first i The speed of movement of each particle It is the first i The current position of each particle The learning factor is usually set to... . For the firsti The optimal solution for each particle. It is the optimal solution for the entire population; Inertial weight; for Random numbers within an interval; inertia weights A linear decreasing strategy is adopted. It is a non-negative number that decreases linearly. It is the initial inertia weight. This is the inertia weight when it is iterated to the maximum. In this embodiment, it is set as follows: =0.9, =0.4.

[0056] In the updated particle formula, the first part This indicates that the particle is affected by its previous velocity and inertia; Part Two. This indicates that particles are influenced by their own experience, Part Three. Reflecting the shared experience of the particle swarm, these three parts determine the particle's next motion.

[0057] S5.3: Repeat steps S5.1 and S5.2 based on the parameter-error sensitivity results until the predetermined number of iterations is reached or the objective function is less than the given convergence error, i.e., find the value corresponding to the minimum error. And its parameter subspace. Based on the parameter-data sensitivity results, repeat steps S5.1 and S5.2 to obtain the static parameters with the smallest static error. Finally, based on the parameter-data sensitivity results, repeat steps S5.1 and S5.2 to obtain the dynamic parameters with the smallest error.

[0058] The identification results obtained in this embodiment are shown in Table 3.

[0059] Table 3. Final identification results obtained using the particle swarm optimization algorithm.

[0060] like Figure 2 As shown, the specific steps of combining dual sensitivity analysis and particle swarm optimization are broken down in detail, covering steps such as fixing key parameters, obtaining parameter subspace, calculating the minimum error of subspace, establishing parameter-error sensitivity mapping, initializing particle swarm parameters, locating the global optimal parameters and subspace, and identifying static and dynamic parameters step by step.

[0061] In this embodiment, a particle swarm optimization algorithm is adopted and a linearly decreasing inertial weight is introduced. Combined with the optimal subspace provided by the aforementioned second sensitivity analysis, an optimization strategy combining global coarse search and local fine-tuning is formed. This effectively balances global exploration and local development capabilities, reduces the risk of getting trapped in local optima in complex high-dimensional parameter spaces, and thus improves the global optimality and accuracy of the final identification results.

[0062] Step S6: Substitute the identified energy storage black box model parameters into the electromechanical transient model for simulation to obtain the simulation response curves of terminal voltage, reactive current, active power and reactive power. Compare these response curves with the response curves generated by the reference dataset obtained from the hardware-in-the-loop simulation to verify the accuracy of the identification results.

[0063] like Figure 8 The figure shows a comparison between simulation and test results. It compares the simulation results of the electromechanical transient model with the identified parameters substituted into the model under a set low-voltage ride-through condition, and the test results of the semi-physical simulation reference dataset. Specifically, it includes the response curves of four types of electrical quantities: terminal voltage U, current I, active power P, and reactive power Q, verifying the identification method. Step S6 performs simulation to obtain the simulated response curves of key electrical quantities, and compares them with the response curves generated by the reference dataset, as shown below. Figure 8 As shown in Table 4, the average relative error between the simulation data and the semi-physical test data obtained from the identification results is shown in Table 4.

[0064] Table 4. Average Relative Errors Between Simulation Data and Semi-Physical Test Data

[0065] This embodiment implements a complete closed-loop verification process: the identification parameters are substituted back into the model for simulation and compared with the reference data. The effectiveness and accuracy of the identification method are intuitively verified in the form of objective curve comparison, which enhances the persuasiveness and reliability of the technical solution.

[0066] The above embodiments are merely preferred technical solutions of the present invention and should not be considered as limitations on the present invention. The scope of protection of the present invention should be limited to the technical solutions described in the claims, including equivalent substitutions of the technical features described in the claims. That is, equivalent substitutions and improvements within this scope are also within the scope of protection of the present invention.

Claims

1. A method for parameter identification of a dual-sensitivity energy storage black box model, characterized in that, Includes the following steps: S1: Build the main circuit model using the existing energy storage electromechanical transient model in PSASP, and set the parameter range according to the initial values ​​given by the system in the energy storage electromechanical model; based on the parameter range, perform a linear scan on each parameter, changing only one parameter at a time while keeping other parameters unchanged, to obtain simulation data under different parameter values, and form a test dataset; S2: Obtain hardware-in-the-loop simulation data of energy storage under low voltage ride-through conditions, process the data and use it as a reference dataset, and calculate the average relative error between the test dataset and the reference dataset. S3: Based on the test dataset, perform a first sensitivity analysis, calculate the sensitivity of each scanned parameter change to the simulation output data, classify the parameters according to the sensitivity results, and select the parameters to be identified. S4: Based on the average relative error, a second sensitivity analysis is performed. For the selected parameter to be identified, the trend of the minimum average relative error that can be obtained by randomly changing other parameters when the parameter is fixed at different values ​​is analyzed to determine the sensitivity of the parameter to the final identification error. S5: The particle swarm optimization algorithm is used for parameter identification. The weighted sum of the average relative errors is defined as the objective function, and a penalty function term based on the parameter variation range is introduced into the objective function. Based on the results of the second sensitivity analysis, the particle swarm optimization algorithm is guided to perform global search and local optimization in the preferred subspace of key parameters until the convergence condition is met, thus obtaining the identified parameters of the energy storage black box model.

2. The parameter identification method for energy storage black box model based on dual sensitivity as described in claim 1, characterized in that, S1 specifically includes the following steps: Based on the default initial parameters, ignoring the dynamics of the DC voltage side and considering only the dynamics of the output side, the expression for the preset parameter variation range symmetrically around zero is: ； In the formula, -5a is the lower limit of the parameter range, and 5a is the upper limit of the parameter range; A single parameter changes from the lower limit of its range to the upper limit by a preset magnitude, while the other parameters remain unchanged; After the current parameter is scanned, the initial value is reset and the next parameter is scanned. Physical quantity data under different scan values ​​of each parameter are obtained as test datasets.

3. The parameter identification method for energy storage black box model based on dual sensitivity as described in claim 1, characterized in that, S2 includes the following steps: The low-voltage ride-through fault condition of energy storage is set up, and the step size of the hardware-in-the-loop simulation is consistent with that of the power system simulation software. A reference dataset is obtained after data processing. The average relative error is calculated for both steady-state and dynamic sampling points, using the relative error between steady-state test data and steady-state reference data, and between dynamic test data and dynamic reference data, respectively. The formula is as follows: ； In the formula, W The average relative error, N The number of discrete sampling points. O s ( t )and O v ( t () represent steady-state and dynamic test data, respectively. O ms ( t )and O mv ( t () represent steady-state and dynamic reference data, respectively. S This represents the number of steady-state sampling points. V Indicates the number of dynamic sampling points; where: 。 4. The parameter identification method for a dual-sensitivity energy storage black box model according to claim 1, characterized in that, S3 includes: The first sensitivity analysis uses the average relative sensitivity, and the formula is: ； In the formula, The parameter is Test data at that time The parameter is Test data at that time, and Similar to the scan value step size, it is the difference between the upper and lower limits of the parameter divided by the number of scan points.

5. The parameter identification method for a dual-sensitivity energy storage black box model according to claim 4, characterized in that, Based on the results of the first sensitivity analysis, the parameters are classified as follows: Irrelevant parameters that have no effect on the output data throughout the entire scan range; The limiting threshold parameter of the control logic only affects the value when it is within a specific range; it is a steady-state parameter that mainly affects the steady-state output data. And the dynamic parameters that mainly affect the output data of the transient process; The amplitude limiting threshold parameter is set to a fixed value according to the actual logic of the controller, and the steady-state parameter and dynamic parameter are used as parameters to be identified.

6. The parameter identification method for a dual-sensitivity energy storage black box model according to claim 1, characterized in that, In S4, the second sensitivity analysis includes: For the set of parameters to be identified θ ={ θ 1, ..., θ n In}, any parameter Within its range of variation Take multiple fixed values; For each fixed value ， In the remaining parameters Perform multiple random sampling combinations within the initial value range, calculate the average relative error for each combination, and take the minimum value. ; analyze The curves showing the variation of the minimum error value for different fixed values, and the minimum point on the curves corresponding to these curves. The value is considered its theoretical identification value, and the parameters are calculated based on this curve. Sensitivity The formula is: ； in, Indicates parameters Sensitivity to model output Is with The parameter value corresponding to the lowest point on the [the graph]. and The closest On Two values.

7. The parameter identification method for a dual-sensitivity energy storage black box model according to claim 1, characterized in that, In step S5, the objective function is defined as: ； in, and These represent the average relative errors of voltage, current, active power, and reactive power, respectively. and These are the weighting coefficients for each physical quantity. For the first i The penalty weight coefficients for each parameter, Let it be the penalty function; The expression is: ； in, , These are the upper and lower limits of the parameter range, respectively. It is the midpoint of the parameter range. This represents the range of parameter variations.

8. The parameter identification method for a dual-sensitivity energy storage black box model according to claim 7, characterized in that, The parameter identification using the particle swarm optimization algorithm includes: Initialize the particle swarm, in 3D target space generation A population of particles, with particle initialization positions determined by the upper and lower boundaries of the target space. Sure; Iterative updates of particle position and velocity, using the following formula: ； ； In the formula, It is the first i The speed of movement of each particle It is the first i The current position of each particle As a learning factor, For the first i The optimal solution for each particle. It is the optimal solution for the entire population; Inertial weight; for Random numbers within an interval.

9. The parameter identification method for a dual-sensitivity energy storage black box model according to claim 8, characterized in that, The inertial weight Using a linear decreasing strategy, the calculation formula is as follows: ； In the formula, It is the initial inertia weight. It is the inertia weight when it is iterated to the maximum.

10. The method for parameter identification of a dual-sensitivity energy storage black box model according to any one of claims 1-9, characterized in that, It also includes S6: Substituting the identified parameters into the energy storage electromechanical transient model for simulation, obtaining the simulation response curve of the key electrical quantities, and comparing it with the response curve generated by the reference dataset to verify the accuracy of the identification results.

Images