Energy cell-friendly interaction probabilistic evaluation method

Abstract

The invention relates to an energy cell-friendly interaction probabilistic evaluation method. The method comprises the steps of obtaining load characteristic data of modeling nodes under the energy cells, selecting a comprehensive load model as a load model structure based on the load characteristic data, and optimizing the load model structure to obtain model parameters so as to obtain a data driving model; establishing a physical mechanism model based on equipment of different load types under the energy cells; performing hybrid modeling on the data driving model and the physical mechanism model to obtain a hybrid model of the energy cells; and based on the hybrid model, simulating interaction scenes among different energy cells through a generative adversarial network, and evaluating the performance of the energy cells in different interaction scenes. According to the invention, efficient interaction between energy cells and optimization of the whole system can be realized.

Description

Technical Field

[0001] The present invention relates to the technical field of power systems, and in particular to a probabilistic evaluation method for friendly interaction of energy cells. Background Art

[0002] In the context of new power systems, the penetration rate of power electronic equipment such as distributed photovoltaics and energy storage systems in the power grid continues to increase, and the operating environment of the power system is becoming increasingly complex. In particular, distributed resources themselves are highly uncertain and volatile. How to effectively evaluate and optimize the coordinated aggregation of these energy units has become a key issue. Most existing methods rely only on single data-driven modeling or mechanism modeling. This isolated approach makes it difficult to fully capture the complex source-load relationship for accurate modeling. In addition, there is also a lack of effective evaluation methods for how to achieve efficient and stable interaction between multiple energy units to enhance the overall performance of the system. Summary of the invention

[0003] To this end, the present invention provides a probabilistic evaluation method for energy cell friendly interaction. By combining data-driven and physical mechanism modeling and using the Red-billed Blue Magpie Optimization Algorithm (RMBO), each minimized unit (i.e., "energy cell") in the system is modeled and optimized, thereby achieving efficient interaction between energy cells and optimization of the overall system.

[0004] In order to solve the above technical problems, the present invention provides a probabilistic evaluation method for friendly interaction of energy cells, wherein each of the energy cells is regarded as an independent energy management unit, including photovoltaic power generation, energy storage devices and load equipment, including:

[0005] Obtaining load characteristic data of the modeling node under the energy cell, selecting a comprehensive load model as a load model structure based on the load characteristic data, optimizing the load model structure, obtaining model parameters, and thus obtaining a data-driven model;

[0006] Based on the equipment of different load types under the energy cell, a physical mechanism model is established;

[0007] Performing hybrid modeling on the data-driven model and the physical mechanism model to obtain a hybrid model of energy cells;

[0008] Based on the hybrid model, the interaction scenarios between different energy cells are simulated through a generative adversarial network to evaluate the performance of energy cells in different interaction scenarios.

[0009] In one embodiment of the present invention, the load characteristic data of the modeling node under the energy cell is obtained, and based on the load characteristic data, a comprehensive load model is selected as the load model structure, including:

[0010] The load characteristic acquisition device based on the modeling node records the transient recording data on the bus of the modeling node, including sampling the bus voltage and bus current;

[0011] Preprocess the collected bus voltage and bus current data to obtain the power data of the modeling node load;

[0012] The basic form of the polynomial model of the load model structure is:

[0013]

[0014] Where P represents active power, Q represents reactive power, U represents voltage, f represents frequency, and subscript 0 represents the initial operating point, i.e., the rated value, and satisfies:

[0015] p z +p i +p p =1

[0016] q z +q i +q p =1

[0017] In the formula, U0 and f0 represent the rated voltage and rated frequency respectively, P0 and Q0 represent the rated active and reactive power at the rated voltage and frequency respectively; p z 、p i 、p p They represent the proportion of active power q that is related to the square of voltage, related to the first power of voltage, and independent of voltage. z ,q i ,q p They represent the proportion of reactive power related to the square of voltage, related to the first power of voltage, and independent of voltage. z ,q z 、p i ,q i 、p p ,q p are respectively regarded as the proportion of constant impedance, constant current and constant power load in the total power. Therefore, these parameters are called ZIP parameters, and the static load model in polynomial form is also called ZIP model; Δf=f-f0, L DP , L DQ They are the active frequency coefficient and the reactive frequency coefficient, respectively reflecting the response of active power and reactive power to frequency changes.

[0018] In one embodiment of the present invention, the load model structure is optimized to obtain model parameters, thereby obtaining a data-driven model, including:

[0019] The unknown parameters in the load model structure are optimized using the least squares method, and the optimal parameters obtained are the model parameters, including:

[0020] Assume there is a variable Y, and an n-dimensional variable X = (x1, x2, ... x n ) is linearly related, that is:

[0021] Y=Xθ

[0022] In the formula, Y represents the dependent variable column vector, X represents the independent variable matrix, and θ represents the parameter column vector, which can be expressed as:

[0023]

[0024] The optimal value of the parameter is directly obtained by solving the following formula by the least square method, which is the identification result of the load model parameters under data driving;

[0025] Use the least squares method to obtain unknown data and minimize the sum of squares of the errors between the obtained data and the actual data:

[0026]

[0027] In the formula: M is the number of unknowns, N is the number of equations, β is the coefficient to be solved, Z and H are known data samples. In order to find a suitable β to make the equation as valid as possible, the residual sum of squares function is introduced:

[0028] F(β)=||Zβ-H||

[0029] The above formula finds the maximum value of the β differential and after rearrangement, we can get:

[0030] β=(Z T Z) -1 Z T H

[0031] That is, the data-driven modeling part is completed through the system identification method.

[0032] In one embodiment of the present invention, a physical mechanism model is established based on the equipment of different load types under the energy cell, including:

[0033] Establish a static load model for a single power electronic load:

[0034]

[0035] Among them, P, Q, and U are the active power, reactive power, and voltage of the power electronic load, respectively; Po, Qo, and Uo are the steady-state active power, steady-state reactive power, and rated voltage of the power electronic load, respectively; U T Indicates the cut-off voltage. To ensure the safety of power electronic loads, UT Greater than the voltage critical value; q u Represents the reactive voltage coefficient of the power electronic load. Since the reactive power of the power electronic load has a negative characteristic with the voltage, q u Less than 0;

[0036] The static equivalent load model of the power electronic load group is established, and its expression is:

[0037] P E =h(U)P E0

[0038]

[0039] Where P E Indicates that P E0 Indicates that Q E Indicates that Q E0 Indicates; h(U) represents the normalized function related to voltage, using the sigmoid activation function h s (x), normal distribution probability distribution function h F (x), where:

[0040]

[0041] Based on the static characteristics of the power electronic load group, the dynamic simplified model of a single power electronic load is extended to the dynamic equivalent load model of the power electronic load group:

[0042] P=[h(U)P0](1+G p1 U′+G p2 U′ 2 )

[0043]

[0044] Where U′ represents the voltage derivative;

[0045]

[0046] In the formula, G P1 , G P2 , G Q1 , G Q2 , K1, K2 are 6 transfer function parameters to be determined, s is the complex variable in Laplace transform, which is used to represent the operation in the frequency domain; K2 is related to the oscillation frequency of the dynamic process. The larger K2 is, the faster the dynamic response rate is. K1 is related to both the oscillation frequency and the overshoot. GP1, GP2, GQ1, GQ2 are proportional coefficients, which determine the gain of each transfer function and together determine the power change of the dynamic process.

[0047] In one embodiment of the present invention, a physical mechanism model is established based on the equipment of different load types under the energy cell, and further includes:

[0048] The ampere-hour measurement method is used to estimate the state of charge SOC and establish a mechanism model:

[0049]

[0050] U dc =U imax ·SOC+U imin (1-SOC)-R i I dc

[0051] Among them, SOC t and SOC t+Δt They represent the battery state of charge at time t and the battery state of charge after Δt respectively; η represents the correction coefficient, which is related to the battery temperature, charge and discharge efficiency, and battery usage time; i(t) represents the battery current, with discharge as the positive direction; S Ah Indicates battery capacity in Ah; U imax Indicates the electromotive force when the battery is fully charged, i.e. SOC = 1; U imin Indicates the electromotive force when the battery is fully discharged, i.e. SOC = 0; U dc Indicates the battery discharge voltage; R i Indicates the internal resistance of the battery; I dc Indicates the battery discharge current.

[0052] In one embodiment of the present invention, the data-driven model and the physical mechanism model are hybrid modeled to obtain a hybrid model of energy cells, including:

[0053] Assume that the system state variable is y, the input variable is x, the data-driven model and the physical mechanism model are represented by F date =(x,y)=0 and F phys =(x,y)=0;

[0054] In order to build a unified hybrid model, the weight parameter α∈[0,1] is introduced to balance the influence of the two models. The implicit function of the hybrid model is expressed as:

[0055] αF date (x,y)+(1-α)F phys (x,y)=0

[0056] in

[0057] F date (x,y)=yG(x,θ)

[0058] Fphys (x,y)=A(x)·y+B(x)=0

[0059] G(x,θ) is the model function obtained by the data-driven method, θ is the identified parameter, and A(x) and B(x) are the matrix and vector constructed according to the physical model.

[0060] In one embodiment of the present invention, simulating the interaction scenarios between different energy cells by generating adversarial networks includes: simulating the following interaction scenarios by generating adversarial networks:

[0061] Load fluctuation scenarios, simulating load peaks and valleys at different times, and evaluating the energy cell's ability to respond to load changes;

[0062] Resource availability change scenarios, simulating fluctuations in renewable energy and evaluating the system’s ability to adapt under conditions of resource uncertainty;

[0063] Fault and abnormal scenarios, simulating unexpected events such as equipment failure and voltage anomalies, and evaluating the robustness and self-recovery capabilities of the system;

[0064] Seasonal and climate impact scenarios generate load and resource conditions under different seasons and climate conditions to test the performance of energy cells under different environmental conditions.

[0065] In one embodiment of the present invention, evaluating the performance of energy cells in different interaction scenarios includes:

[0066] Performing Monte Carlo simulation on the interaction scenarios, and evaluating the performance of energy cells in different interaction scenarios according to the key performance indicators of each energy cell; wherein the key performance indicators include power output stability, voltage stability and robustness, the power output stability is used to measure the fluctuation of power output of energy cells in different scenarios, and the voltage stability is used to evaluate whether the voltage level meets the standard to avoid voltage exceeding the limit;

[0067] A fitness function is constructed based on power output stability, voltage stability and robustness, and the fitness function is optimized by a red-billed blue magpie optimization algorithm.

[0068] In one embodiment of the present invention, the power output stability expression is:

[0069]

[0070] Where: P ref is the reference power, P i is the output power of the ith simulation, N is the number of simulations;

[0071] The voltage stability expression is:

[0072]

[0073] Where: V i is the ith analog voltage, V max ,V min They are the maximum and minimum values ​​allowed for voltage respectively;

[0074] The robustness expression is:

[0075]

[0076] Where: S i Give stability score to the ith simulation;

[0077] The fitness function is:

[0078] F=w1f1+w2f2+w3f3+C

[0079] Among them, w1, w2, w3 are weight coefficients, and C is the empirical correction coefficient.

[0080] In one embodiment of the present invention, after evaluating the performance of energy cells in different interaction scenarios, the method further includes:

[0081] Setting different deviation thresholds for each of the key performance indicators and performing feedback adjustments on the hybrid model based on the evaluation results; including:

[0082] When the deviations of all the key performance indicators are within the corresponding set deviation thresholds, the data-driven model is modified;

[0083] When the deviation of one or more of the key performance indicators exceeds a set deviation threshold, modifying the physical mechanism model;

[0084] The key performance indicator evaluation step is performed again based on the modified hybrid model. If the key performance indicators do not meet the preset range, the modification step is repeated until the key performance indicators all meet the preset range.

[0085] The above technical solution of the present invention has the following advantages compared with the prior art:

[0086] The probabilistic evaluation method of energy cell friendly interaction described in the present invention improves the coordination between independent units in a distributed energy system and enhances the stability and efficiency of the system. The method achieves internal harmony and stable interaction between energy cells through the fusion modeling of data-driven and physical mechanisms and the optimization control of each unit of the system through the red-billed blue magpie optimization algorithm.

[0087] In the modeling process, the present invention no longer relies solely on a single data-driven or single mechanism modeling method, but combines the two. In traditional methods, pure data-driven models are easily limited by the integrity and representativeness of sample data, while pure mechanism models may cause deviations due to incomplete understanding of the internal structure and parameters of the system. The present invention introduces weight parameters between the data-driven model and the physical mechanism model to achieve a dynamic balance between the two, thereby obtaining a more accurate hybrid model. This fusion ensures that when faced with new energy cell systems with multiple types, multiple levels, and multiple time-varying characteristics, the model still has high generalization ability and accuracy.

[0088] The present invention establishes a physical mechanism model (including power electronic load, battery SOC state estimation, etc.) based on various types of loads and energy storage devices. Through the ZIP model, static and dynamic equivalent models, and energy storage modeling methods under the ampere-hour measurement method, the present invention can take into account the complex characteristics and dynamic behaviors of diversified equipment. In the context of a large number of distributed power electronic devices and high penetration rates of energy storage units, this fine characterization method enables the model to have higher reproducibility and accuracy for the internal operating mechanism and state response of the system.

[0089] This invention introduces a generative adversarial network (GAN) to simulate the interaction scenarios between energy cells. Through adversarial training of the generator and the discriminator, GAN can realistically generate multiple types of typical or extreme operating conditions, including load fluctuations, changes in resource availability, failures and abnormal events, seasonal and climate impacts, etc. The introduction of such diverse scenarios will expand the evaluation from a single static operating condition to a rich set of dynamic uncertainty scenarios, thereby comprehensively examining the response characteristics and performance of energy cells in various uncertain environments, and improving the credibility and robustness of the evaluation results.

[0090] In the present invention, the performance evaluation of energy cells in different interaction scenarios is based on the Monte Carlo simulation method, and the uncertainty of the results is quantified by repeated simulation in a statistical sense. On this basis, the key performance indicators (such as power output stability, voltage stability and robustness) are analyzed by probabilistic methods, which helps to identify potential risk points and the degree of influence of uncertain factors. This makes the final evaluation results closer to the actual operating conditions, ensuring that the judgment of system safety and reliability in practical applications is more accurate.

[0091] After the evaluation, the present invention uses the Red-billed Blue Magpie Optimization Algorithm (RMBO) to optimize the fitness function, and corrects the hybrid model based on the feedback of the performance index deviation. When the system deviation is small, but there is still room for optimization, the data-driven model is fine-tuned to improve the sensitivity and accuracy of the model; when significant deviations occur, the mechanism model parameters are readjusted to ensure physical consistency and the model's adaptability to a wide range of state changes. The model parameters are continuously adjusted and optimized in a cyclic iterative manner to achieve continuous improvement in system performance, so that the entire energy cell system can always maintain a relatively optimal operating state under different conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0092] In order to make the contents of the present invention more clearly understood, the present invention is further described in detail below based on specific embodiments of the present invention in conjunction with the accompanying drawings.

[0093] Figure 1 It is a flow chart of the probabilistic evaluation method of energy cell friendly interaction of the present invention.

[0094] Figure 2 It is a probabilistic evaluation interaction flow chart of the present invention.

[0095] Figure 3 It is a flow chart of the red-billed blue magpie algorithm of the present invention. DETAILED DESCRIPTION

[0096] The present invention is further described below in conjunction with the accompanying drawings and specific embodiments so that those skilled in the art can better understand the present invention and implement it, but the embodiments are not intended to limit the present invention.

[0097] In the present invention, "several" means one or more, "multiple" means more than two, "greater than", "less than", "exceed" and the like are understood to exclude the number itself; "above", "below", "within" and the like are understood to include the number itself. In the description of the present invention, if there is a description of "first" or "second", it is only used for the purpose of distinguishing technical features, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features or implicitly indicating the order of the indicated technical features.

[0098] Reference Figure 1 As shown, it is a flow chart of a probabilistic evaluation method of energy cell friendly interaction of the present invention. The definition of energy cell: each energy cell is regarded as an independent energy management unit, including photovoltaic power generation, energy storage device and load equipment. The energy cell is internally constructed by combining data-driven machine learning modeling with physical mechanism modeling to build a complete cell internal state model. The interaction between cells, that is, the interaction between substations, is dynamically evaluated and corrected by setting the frequency interaction function.

[0099] The probabilistic evaluation method of energy cell friendly interaction includes:

[0100] Step S1, obtaining load characteristic data of the modeling node under the energy cell, selecting a Classic Load Model (CLM) as the load model structure based on the load characteristic data, optimizing the load model structure, obtaining model parameters, and thus obtaining a data-driven model.

[0101] It should be noted that the basic idea of ​​data-driven is to regard the modeling nodes (i.e. all loads, distributed generation sources, and energy storage in the area) as a whole, and record the fault power and voltage curves through the load characteristic acquisition device in the power grid.

[0102] Through the method of system identification, the load model and parameters under the modeling node are obtained. The load model parameter identification problem is transformed into a parameter optimization problem. For a given set of measured load power and voltage, this method calculates the load power under a given voltage and a large number of different model parameters, and then takes the minimum deviation between the calculated power and the measured power as the optimization goal, and uses the optimization algorithm to search for the optimal parameters, which is the load model parameter identification result.

[0103] This method does not require in-depth investigation into the load. Compared with the traditional statistical synthesis method, it only needs to model the fault recording data of the node, without knowing the composition and proportion of various types of electrical equipment inside the node. Therefore, it has a wider range of applications.

[0104] Step S1 specifically includes:

[0105] The load characteristic acquisition device based on the modeling node records the transient recording data on the bus of the modeling node, including sampling the bus voltage and bus current;

[0106] Preprocess the collected bus voltage and bus current data (such as removing abnormal data, normalizing, and obtaining the power data of the modeling node load;

[0107] The unknown parameters in the load model structure are optimized using the least squares method, and the optimal parameters obtained are the model parameters.

[0108] The Classic Load Model (CLM) is a commonly used model in power system model analysis. It adopts the structure of a static load model in parallel with an induction motor model. This model retains the dynamic and static characteristics of the load as much as possible on the basis of simplifying the complexity of the model. The static load model in CLM usually adopts the ZIP model, and the induction motor model usually adopts the third-order electromechanical transient model.

[0109] Among them, the basic form of the polynomial model of the load model structure is:

[0110]

[0111] Where P represents active power, Q represents reactive power, U represents voltage, f represents frequency, and subscript 0 represents the initial operating point, i.e., the rated value, and satisfies:

[0112] p z +p i +p p =1

[0113] q z +q i +q p =1

[0114] In the formula, U0 and f0 represent the rated voltage and rated frequency respectively, P0 and Q0 represent the rated active and reactive power at the rated voltage and frequency respectively; p z 、p i 、p p They represent the proportion of active power q that is related to the square of voltage, related to the first power of voltage, and independent of voltage. z ,q i ,q p They represent the proportion of reactive power related to the square of voltage, related to the first power of voltage, and independent of voltage. z ,q z 、p i ,q i 、p p ,q p are respectively regarded as the proportion of constant impedance, constant current and constant power load in the total power. Therefore, these parameters are called ZIP parameters, and the static load model in polynomial form is also called ZIP model; Δf=f-f0, L DP , L DQ They are the active frequency coefficient and the reactive frequency coefficient, respectively reflecting the response of active power and reactive power to frequency changes.

[0115] Assume there is a variable Y, and an n-dimensional variable X = (x1, x2, ... x n ) is linearly related, that is:

[0116] Y=Xθ

[0117] In the formula, Y represents the dependent variable column vector, X represents the independent variable matrix, and θ represents the parameter column vector, which can be expressed as:

[0118]

[0119] The optimal value of the parameter is directly obtained by solving the following formula by the least square method, which is the identification result of the load model parameters under data driving;

[0120] The least squares method is a mathematical optimization method that finds the best function matching of data by minimizing the sum of squares of errors. The least squares method is used to obtain unknown data and minimize the sum of squares of errors between the obtained data and the actual data:

[0121]

[0122] Where: M is the number of unknowns, N is the number of equations, β is the coefficient to be solved, Z and H are known data samples. This system of equations generally has no solution. In order to find a suitable β to make the equation as valid as possible, the residual sum of squares function is introduced:

[0123] F(β)=||Zβ-H||

[0124] The above formula finds the maximum value of the β differential and after rearrangement, we can get:

[0125] β=(Z T Z) -1 Z T H

[0126] That is, the data-driven modeling part is completed through the system identification method.

[0127] Step S2: establishing a physical mechanism model based on the equipment of different load types under the energy cell.

[0128] Physical mechanism modeling is based on the characteristics of different types of loads, and corresponding mechanism models are established respectively to accurately describe the diverse loads in the system.

[0129] At present, power electronic loads in the substation, such as frequency converters, are usually connected to the AC power grid through a rectifier-DC-inverter energy conversion device. The main circuit of the system can be considered to be composed of a three-phase rectifier, a DC system, a three-phase inverter and an induction motor. The three-phase rectifier and the three-phase inverter are connected together through a DC capacitor Cdc to achieve energy conversion on both sides. Among them, the three-phase rectifier usually adopts a bridge uncontrolled rectifier circuit due to cost and other factors. The three-phase inverter usually adopts a dual-loop control strategy. The outer loop realizes the speed control of the motor or the power control on the inverter side, and the inner loop mainly controls the voltage and frequency on the inverter side.

[0130] The commonly used control strategy for three-phase inverters is constant voltage-frequency ratio control (V / f control), which keeps the ratio between the output voltage and frequency at a constant value, thereby maintaining a constant magnetic flux. The torque of an asynchronous motor is generated by the interaction between the magnetic flux of the motor and the current flowing through the rotor. If the frequency is only reduced while the voltage is constant, the magnetic flux will increase, the magnetic circuit will be saturated, and in severe cases, the motor will burn out. Therefore, the V / f control strategy proportionally changes the voltage and frequency output by the inverter, so that the magnetic flux of the motor remains relatively constant and the motor speed is controlled. In addition to the inverter connected to the grid in the form of AC-DC-AC, some power electronic loads are directly connected to the grid in the form of AC-DC (common in devices such as LED lights). From the perspective of the AC grid, the load power characteristics of the two are similar.

[0131] Then, a static equivalent load model of the power electronic load group and a dynamic equivalent load model of the power electronic load group based on the second-order transfer function are established.

[0132] Establish a static load model for a single power electronic load:

[0133]

[0134] Among them, P, Q, and U are the active power, reactive power, and voltage of the power electronic load, respectively; Po, Qo, and Uo are the steady-state active power, steady-state reactive power, and rated voltage of the power electronic load, respectively; U T Indicates the cut-off voltage. To ensure the safety of power electronic loads, U T Greater than the voltage critical value; q u Represents the reactive voltage coefficient of the power electronic load. Since the reactive power of the power electronic load has a negative characteristic with the voltage, q u Less than 0;

[0135] The static equivalent load model of the power electronic load group is established, and its expression is:

[0136] P E =h(U)P E0

[0137]

[0138] Where P E Indicates that P E0 Indicates that Q E Indicates that Q E0 Indicates; h(U) represents the normalized function related to voltage, using the sigmoid activation function h s (x), normal distribution probability distribution function h F (x), where:

[0139]

[0140] Based on the static characteristics of the power electronic load group, the dynamic simplified model of a single power electronic load is extended to the dynamic equivalent load model of the power electronic load group:

[0141] P=[h(U)P0](1+G p1 U′+G p2 U′ 2 )

[0142] Q=[h(U)Q0U qu ](1+G q1 U′+G q2 U′ 2 )

[0143] The first part is the part enclosed by square brackets, which indicates the static characteristics of the power electronic load power as the voltage changes. Since the dynamic process of the power electronic load group is usually very rapid, its power still shows static characteristics during the voltage sag. The second part is the part enclosed by round brackets, which indicates the response of the power electronic load group to the voltage change. It includes the voltage derivative U′ and its square U′ 2 In the simplified model, U′ and U′ are used. 2 As input, compared with ΔU and ΔU in the linearized model 2 , which can ensure that the input quantity is 0 during the voltage sag phase, that is, after the dynamic process is over, the power is only related to the static characteristics of the power electronic load.

[0144]

[0145] In the formula, G P1 , G P2 , G Q1 , G Q2 , K1, K2 are 6 transfer function parameters to be determined, s is the complex variable in Laplace transform, which is used to represent the operation in the frequency domain; K2 is related to the oscillation frequency of the dynamic process. The larger K2 is, the faster the dynamic response rate is. K1 is related to both the oscillation frequency and the overshoot. GP1, GP2, GQ1, GQ2 are proportional coefficients, which determine the gain of each transfer function and together determine the power change of the dynamic process.

[0146] The basic physical and chemical principles of batteries are usually complex, and the battery's own charge and discharge characteristics have strong nonlinear characteristics, and the electrical parameters are also highly time-varying, making modeling difficult. In order to accurately describe the external characteristics of the battery, its model must reflect the relationship between the terminal voltage and the internal resistance. The internal resistance is related to the battery's usage time, SOC, and battery temperature. When SOC is equal to 1, it means that the battery capacity has reached its upper limit; when SOC is equal to 0, it means that the battery capacity is 0 and is fully discharged; when SOC is between 0 and 1, the battery can be charged or discharged as needed.

[0147] The widely used ampere-hour measurement method is used to estimate the state of charge SOC and establish a mechanism model:

[0148]

[0149] U dc =U imax ·SOC+U imin (1-SOC)-R i I dc

[0150] Among them, SOC t and SOC t+Δt They represent the battery state of charge at time t and the battery state of charge after Δt respectively; η represents the correction coefficient, which is related to the battery temperature, charge and discharge efficiency, and battery usage time; i(t) represents the battery current, with discharge as the positive direction; S Ah Indicates battery capacity in Ah; U imax Indicates the electromotive force when the battery is fully charged, i.e. SOC = 1; U imin Indicates the electromotive force when the battery is fully discharged, i.e. SOC = 0; U dc Indicates the battery discharge voltage; R i Indicates the internal resistance of the battery; I dc Indicates the battery discharge current.

[0151] It should be noted that the accurate characterization of energy storage devices requires the modeling of batteries. The basic physical and chemical principles of batteries are usually complex, and the charging and discharging characteristics of batteries themselves have strong nonlinear characteristics. The electrical parameters are also highly time-varying, making modeling difficult. In order to accurately describe the external characteristics of energy storage devices, the state of charge SOC is estimated by the ampere-hour measurement method. This model and the dynamic equivalent load model of the power electronic load group are both components of the "energy cell" and both belong to the mechanism modeling part.

[0152] S3. Perform hybrid modeling on the data-driven model and the physical mechanism model to obtain a hybrid model of energy cells.

[0153] Assume that the system state variable is y, the input variable is x, the data-driven model and the physical mechanism model are represented by F date =(x,y)=0 and F phys =(x,y)=0;

[0154] In order to build a unified hybrid model, the weight parameter α∈[0,1] is introduced to balance the influence of the two models. The implicit function of the hybrid model is expressed as:

[0155] αF date (x,y)+(1-α)F phys (x,y)=0

[0156] in

[0157] F date (x,y)=yG(x,θ)

[0158] F phys (x,y)=A(x)·y+B(x)=0

[0159] G(x,θ) is the model function obtained by the data-driven method, θ is the identified parameter, and A(x) and B(x) are the matrix and vector constructed according to the physical model.

[0160] Data-driven modeling and mechanism modeling are first modeled independently in parallel. The data-driven part uses the overall identification method to model, and obtains the behavior patterns of loads, energy storage and other units through historical data and system identification technology, without considering their internal physical structure. The mechanism modeling part describes each unit in the system in detail based on the principles of physics.

[0161] S4. Reference Figure 2 As shown, a probabilistic evaluation interaction is performed. Based on the hybrid model, the interaction scenarios between different energy cells are simulated by generating an adversarial network to evaluate the performance of energy cells in different interaction scenarios.

[0162] Specifically, after completing the modeling of a single energy cell (station), the interaction between stations is evaluated, and the overall system is optimized through a probabilistic evaluation model.

[0163] Based on historical operating data and possible future scenarios, different load and power generation scenarios are generated using a generative adversarial network (GAN) to ensure coverage of diverse uncertainty conditions. The generative adversarial network (GAN) consists of a generator and a discriminator, which generates realistic interaction scenarios through mutual adversarial training. Specifically, the generator is used to simulate different energy cell interaction scenarios, while the discriminator is used to discriminate the difference between the generated scenarios and real data, thereby continuously improving the generator's capabilities. During the training process, the input historical data includes load change curves, resource fluctuations, etc., to ensure that the generated scenarios are representative.

[0164] The interaction scenarios between different energy cells are simulated by generative adversarial networks (GAN), including: The following four types of interaction scenarios are simulated by generative adversarial networks:

[0165] Load fluctuation scenarios, simulating load peaks and valleys at different times, and evaluating the energy cell's ability to respond to load changes;

[0166] Resource availability change scenarios, simulating fluctuations in renewable energy (e.g., photovoltaic, wind power), and evaluating the system's ability to adapt under conditions of resource uncertainty;

[0167] Fault and abnormal scenarios, simulating unexpected events such as equipment failure and voltage anomalies, and evaluating the robustness and self-recovery capabilities of the system;

[0168] Seasonal and climate impact scenarios generate load and resource conditions under different seasons and climate conditions to test the performance of energy cells under different environmental conditions.

[0169] Specifically, the performance of energy cells in different interaction scenarios is evaluated, including:

[0170] Monte Carlo simulation is performed on the interaction scenario, and large-scale Monte Carlo simulation is performed on the generated interaction scenario. Repeated simulation runs are used to analyze the interaction performance between energy cells under different working conditions. Monte Carlo simulation can quantify the system response under different scenarios, including voltage fluctuations, self-sufficiency rate, energy loss and other indicators.

[0171] Evaluate the performance of energy cells in different interaction scenarios based on the key performance indicators of each energy cell; wherein the key performance indicators include power output stability, voltage stability and robustness. The power output stability is used to measure the fluctuation of power output of energy cells in different scenarios, and the voltage stability is used to evaluate whether the voltage level meets the standard and avoid voltage exceeding the limit;

[0172] A fitness function is constructed based on power output stability, voltage stability and robustness, and the fitness function is optimized by the Red-billed Blue Magpie (RMBO) optimization algorithm. The Red-billed Blue Magpie optimization algorithm is used to find the optimal solution for the energy cell interaction strategy. Its core lies in constructing a fitness function to evaluate the pros and cons of each strategy, and the fitness function is constructed based on key performance indicators.

[0173] Specifically, the power output stability expression is:

[0174]

[0175] Where: P ref is the reference power, P i is the output power of the ith simulation, N is the number of simulations;

[0176] The voltage stability expression is:

[0177]

[0178] Where: V i is the ith analog voltage, V max ,V min They are the maximum and minimum values ​​allowed for voltage respectively;

[0179] The robustness expression is:

[0180]

[0181] Where: S i Give stability score to the ith simulation;

[0182] The fitness function is:

[0183] F=w1f1+w2f2+w3f3+C

[0184] Among them, w1, w2, w3 are weight coefficients, and C is the empirical correction coefficient.

[0185] For each simulation scenario, the key performance indicators of each energy cell are calculated, and the overall performance of the energy cell in different interaction scenarios is evaluated through statistical analysis. By introducing a probabilistic risk assessment model, the interaction risks between different energy cells can be quantified, including energy shortage, overload risk, and the probability of voltage exceeding the limit. Based on these assessment results, potential risk areas can be identified. According to the results of the probabilistic risk assessment, the control strategy of the energy cell is adjusted for feedback. Through the Red-billed Blue Magpie Optimization Algorithm (RMBO), the system parameters are optimized to reduce risks and improve the overall stability and coordination of the system. The fitness function can perform a probabilistic evaluation of the "friendly interaction" of the "energy cell" in the simulated substation scenario through specific evaluation indicators.

[0186] It should be noted that the Red-billed Blue Magpie Optimization Algorithm (RMBO) is a nature-inspired intelligent optimization algorithm, which is inspired by the behavior of the red-billed blue magpie in finding, capturing and storing food. The algorithm simulates this natural process in the optimization problem and continuously iterates to find the global optimal solution. Figure 3 As shown, the following is a description of the process of the red-billed blue magpie optimization algorithm:

[0187] Initialize the population: Generate an initial population, each individual in the population represents a potential solution;

[0188] Set algorithm parameters: including population size, maximum number of iterations, mutation probability, fitness function, etc.;

[0189] Define the search space: determine the scope of the solution space of the optimization problem;

[0190] Evaluate the performance of each population individual according to the defined fitness function;

[0191] To update the position and search for prey, the blue magpie will adjust its position according to the distribution of prey and the location of food storage, simulating the improvement of individual solutions in the optimization process;

[0192] Combine the individual's current solution and the historical optimal solution to update the position to avoid falling into the local optimal solution;

[0193] Use random search strategies to introduce new possible solutions and enhance the global search capability of the algorithm;

[0194] Determine the position of the current optimal individual (i.e. the individual in the population closest to the optimal solution) and simulate the process of the red-billed blue magpie capturing prey;

[0195] All individuals move towards the optimal individual position, and random disturbances are introduced to enhance the diversity of the search;

[0196] The current optimal solution is recorded in the "food storage", indicating that the global optimal solution of the algorithm is updated. If the new solution is better than the current optimal solution, it is replaced by the new optimal solution.

[0197] During the search process, some individuals are randomly selected for mutation to escape from the local optimal solution;

[0198] The mutation operation further explores the search space by making small perturbations to the solution;

[0199] Check whether the termination condition is met:

[0200] The maximum number of iterations is reached; or the fitness value reaches the preset threshold; or the change of the population solution tends to be stable;

[0201] If the termination condition is met, output the optimal solution; otherwise, return to continue iterating.

[0202] Output the final optimal solution.

[0203] Specifically, after evaluating the performance of energy cells in different interaction scenarios, it also includes:

[0204] Different deviation thresholds are set for each of the key performance indicators, and feedback adjustments are made to the hybrid model based on the evaluation results; there are two types of correction methods:

[0205] Minor deviation correction (data-driven model): When the deviations of all the key performance indicators are within the corresponding set deviation thresholds, the data-driven model is corrected. Specifically, the weight coefficient correction method in the data-driven model is used to adjust the learning rate and weight coefficient in the data-driven model to more accurately adapt to the current system operation status, thereby improving the model's sensitivity and response speed to small-scale changes.

[0206] Significant deviation correction (mechanism model): When the deviation of one or more of the key performance indicators exceeds the set deviation threshold, the physical mechanism model is corrected. This includes re-identifying and updating the parameters in the physical model, such as the equivalent impedance of the load model, the inductance parameters of the motor, etc. This correction can ensure that the system can still maintain physical rationality and accuracy in the case of a large range of deviations.

[0207] Feedback loop and iterative optimization: Based on the modified hybrid model, the key performance indicator evaluation step is performed again. If it does not meet the preset range, the correction step is repeated until the key performance indicators meet the preset range (i.e., meet the system requirements). Through the continuous iteration of this feedback loop, the system can gradually approach the optimal state, thereby maintaining efficient and stable performance under complex and changing operating conditions.

[0208] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0209] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0210] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0211] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0212] Finally, it should be noted that the above specific implementation methods are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to examples, those skilled in the art should understand that the technical solution of the present invention can be modified or replaced by equivalents without departing from the spirit and scope of the technical solution of the present invention, which should be included in the scope of the claims of the present invention.

Claims

1. A probabilistic evaluation method for energy cell friendly interaction, wherein each energy cell is regarded as an independent energy management unit, including photovoltaic power generation, energy storage device and load equipment, characterized in that: include: Obtaining load characteristic data of the modeling node under the energy cell, selecting a comprehensive load model as a load model structure based on the load characteristic data, optimizing the load model structure, obtaining model parameters, and thus obtaining a data-driven model; Based on the equipment of different load types under the energy cell, a physical mechanism model is established; Performing hybrid modeling on the data-driven model and the physical mechanism model to obtain a hybrid model of energy cells; Based on the hybrid model, the interaction scenarios between different energy cells are simulated through a generative adversarial network to evaluate the performance of energy cells in different interaction scenarios.

2. A probabilistic evaluation method for energy cell friendly interaction according to claim 1, characterized in that: Obtaining load characteristic data of the modeling node under the energy cell, and selecting a comprehensive load model as a load model structure based on the load characteristic data, including: The load characteristic acquisition device based on the modeling node records the transient waveform data on the bus of the modeling node, including sampling the bus voltage and bus current; Preprocess the collected bus voltage and bus current data to obtain the power data of the modeling node load; The basic form of the polynomial model of the load model structure is: Where P represents active power, Q represents reactive power, U represents voltage, f represents frequency, and subscript 0 represents the initial operating point, i.e., the rated value, and satisfies: p z +p i +p p =1 q z +q i +q p =1 In the formula, U0 and f0 represent the rated voltage and rated frequency respectively, P0 and Q0 represent the rated active and reactive power at the rated voltage and frequency respectively; p z 、p i 、p p They represent the proportion of active power q that is related to the square of voltage, related to the first power of voltage, and independent of voltage. z ,q i ,q p They represent the proportion of reactive power related to the square of voltage, related to the first power of voltage, and independent of voltage. z ,q z 、p i ,q i 、p p ,q p are respectively regarded as the proportion of constant impedance, constant current and constant power load in the total power. Therefore, these parameters are called ZIP parameters, and the static load model in polynomial form is also called ZIP model; Δf=f-f0, L DP , L DQ They are the active frequency coefficient and the reactive frequency coefficient, respectively reflecting the response of active power and reactive power to frequency changes.

3. The method for evaluating the probabilistic interaction of energy cells according to claim 1, characterized in that: The load model structure is optimized to obtain model parameters, thereby obtaining a data-driven model, including: The unknown parameters in the load model structure are optimized using the least squares method, and the optimal parameters obtained are the model parameters, including: Assume there is a variable Y, and an n-dimensional variable X = (x1, x2, ... x n ) is linearly related, that is: Y=Xθ In the formula, Y represents the dependent variable column vector, X represents the independent variable matrix, and θ represents the parameter column vector, which can be expressed as: The optimal value of the parameter is directly obtained by solving the following formula by the least square method, which is the identification result of the load model parameters under data driving; Use the least squares method to obtain unknown data and minimize the sum of squares of the errors between the obtained data and the actual data: In the formula: M is the number of unknowns, N is the number of equations, β is the coefficient to be solved, Z and H are known data samples. In order to find a suitable β to make the equation as valid as possible, the residual sum of squares function is introduced: F(β)=||Zβ-H|| The above formula finds the maximum value of the β differential and after rearrangement, we can get: β=(Z T WITH) -1 WITH T H That is, the data-driven modeling part is completed through the system identification method.

4. The method for evaluating the probabilistic interaction of energy cells according to claim 1, characterized in that: Based on the equipment of different load types under the energy cell, a physical mechanism model is established, including: Establish a static load model for a single power electronic load: Among them, P, Q, and U are the active power, reactive power, and voltage of the power electronic load, respectively; Po, Qo, and Uo are the steady-state active power, steady-state reactive power, and rated voltage of the power electronic load, respectively; U T Indicates the cut-off voltage. To ensure the safety of power electronic loads, U T Greater than the voltage critical value; q u Represents the reactive voltage coefficient of the power electronic load. Since the reactive power of the power electronic load has a negative characteristic with the voltage, q u Less than 0; The static equivalent load model of the power electronic load group is established, and its expression is: P E =h(U)P E0 Where P E Indicates that P E0 Indicates that Q E Indicates that Q E0 Indicates; h(U) represents the normalized function related to voltage, using the sigmoid activation function h s (x), normal distribution probability distribution function h F (x), where: Based on the static characteristics of the power electronic load group, the dynamic simplified model of a single power electronic load is extended to the dynamic equivalent load model of the power electronic load group: P=[h(U)P0](1+G p1 U'+G p2 U' 2 ) Where U′ represents the voltage derivative; In the formula, G P1 , G P2 , G Q1 , G Q2 , K1, K2 are 6 transfer function parameters to be determined, s is the complex variable in Laplace transform, which is used to represent the operation in the frequency domain; K2 is related to the oscillation frequency of the dynamic process. The larger K2 is, the faster the dynamic response rate is. K1 is related to both the oscillation frequency and the overshoot. GP1, GP2, GQ1, GQ2 are proportional coefficients, which determine the gain of each transfer function and together determine the power change of the dynamic process.

5. The method for evaluating the probabilistic interaction of energy cells according to claim 4, characterized in that: Based on the equipment of different load types under the energy cell, a physical mechanism model is established, which also includes: The ampere-hour measurement method is used to estimate the state of charge SOC and establish a mechanism model: The dc =U imax ·SOC+U imin ·(1-SOC)-R i ·I dc Among them, SOC t and SOC t+Δt They represent the battery state of charge at time t and the battery state of charge after Δt respectively; η represents the correction coefficient, which is related to the battery temperature, charge and discharge efficiency, and battery usage time; i(t) represents the battery current, with discharge as the positive direction; S Ah Indicates battery capacity in Ah; U imax Indicates the electromotive force when the battery is fully charged, i.e. SOC = 1; U imin Indicates the electromotive force when the battery is fully discharged, i.e. SOC = 0; U dc Indicates the battery discharge voltage; R i Indicates the internal resistance of the battery; I dc Indicates the battery discharge current.

6. The method for evaluating the probabilistic interaction of energy cells according to claim 1, characterized in that: The data driven model and the physical mechanism model are hybrid modeled to obtain a hybrid model of energy cells, including: Assume that the system state variable is y, the input variable is x, the data-driven model and the physical mechanism model are represented by F date =(x,y)=0 and F phys =(x,y)=0; In order to build a unified hybrid model, the weight parameter α∈[0,1] is introduced to balance the influence of the two models. The implicit function of the hybrid model is expressed as: aF date (x,y)+(1-α)F phys (x,y)=0 in F date (x,y)=yG(x,θ) G(x,θ) is the model function obtained by the data-driven method, θ is the identified parameter, and A(x) and B(x) are the matrix and vector constructed according to the physical model.

7. The method for evaluating the probabilistic interaction of energy cells according to claim 1, characterized in that: Generate an adversarial network to simulate the interaction scenarios between different energy cells, including: Generate an adversarial network to simulate the following interaction scenarios: Load fluctuation scenarios, simulating load peaks and valleys at different times, and evaluating the energy cell's ability to respond to load changes; Resource availability change scenarios, simulating fluctuations in renewable energy and evaluating the system’s ability to adapt under conditions of resource uncertainty; Fault and abnormal scenarios, simulating unexpected events such as equipment failure and voltage anomalies, and evaluating the robustness and self-recovery capabilities of the system; Seasonal and climate impact scenarios generate load and resource conditions under different seasons and climate conditions to test the performance of energy cells under different environmental conditions.

8. The method for evaluating the probabilistic interaction of energy cells according to claim 1, characterized in that: Evaluate the performance of energy cells in different interaction scenarios, including: Performing Monte Carlo simulation on the interaction scenarios, and evaluating the performance of energy cells in different interaction scenarios according to the key performance indicators of each energy cell; wherein the key performance indicators include power output stability, voltage stability and robustness, the power output stability is used to measure the fluctuation of power output of energy cells in different scenarios, and the voltage stability is used to evaluate whether the voltage level meets the standard to avoid voltage exceeding the limit; A fitness function is constructed based on power output stability, voltage stability and robustness, and the fitness function is optimized by a red-billed blue magpie optimization algorithm.

9. The method for evaluating the probabilistic interaction of energy cells according to claim 8, characterized in that: The power output stability expression is: Where: P ref is the reference power, P i is the output power of the ith simulation, N is the number of simulations; The voltage stability expression is: Where: V i is the ith analog voltage, V max ,V min They are the maximum and minimum values ​​allowed for voltage respectively; The robustness expression is: Where: S i Give stability score to the ith simulation; The fitness function is: F=w1f1+w2f2+w3f3+C Among them, w1, w2, w3 are weight coefficients, and C is the empirical correction coefficient.

10. A probabilistic evaluation method for energy cell friendly interaction according to claim 9, characterized in that: After evaluating the performance of energy cells in different interaction scenarios, it also includes: Setting different deviation thresholds for each of the key performance indicators and performing feedback adjustments on the hybrid model based on the evaluation results; including: When the deviations of all the key performance indicators are within the corresponding set deviation thresholds, the data-driven model is modified; When the deviation of one or more of the key performance indicators exceeds a set deviation threshold, modifying the physical mechanism model; The key performance indicator evaluation step is performed again based on the modified hybrid model. If the key performance indicators do not meet the preset range, the modification step is repeated until the key performance indicators all meet the preset range.

Images