A method for online correction of parameters in a comprehensive load model considering distributed photovoltaic power.
By using a multi-objective particle swarm optimization algorithm to correct the parameters of the integrated load model of distributed photovoltaic power online, the time-varying and complex problems of load models in power systems are solved, accurate parameter updates are achieved under small disturbance conditions, and the accuracy of power grid simulation calculations is improved.
Patent Information
- Application Number
- CN202310474423.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2023-04-27
- Publication Date
- 2026-11-17
- Estimated Expiration
- 2043-04-27
AI Technical Summary
Existing power system load models are unable to accurately reflect the complex load changes after distributed photovoltaic (PV) integration, and fixed parameter models cannot adapt to the time-varying nature of the load, resulting in insufficient simulation accuracy.
A multi-objective particle swarm optimization algorithm is used to correct the parameters of the integrated load model of distributed photovoltaic power online. By acquiring power system data in real time, the algorithm distinguishes between large and small disturbance scenarios, corrects the dominant parameters and proportional parameters respectively, and constructs a multi-objective function to improve the model accuracy.
It expands the application scenarios of load model parameter identification, enables parameter updates under smaller voltage fluctuations, provides a more accurate load model, and offers higher precision for power grid simulation calculations.
Smart Images

Figure CN116388204B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to power system optimization methods, specifically to an online correction method for comprehensive load model parameters that takes into account distributed photovoltaic power. Background Technology
[0002] Power system load models have a significant impact on power system power flow calculations, transient stability, static stability, and voltage stability. Due to the strong randomness, complexity, and time-varying nature of loads, the accuracy of load models is difficult to guarantee. In recent years, the integration of distributed renewable energy generation has made load variation characteristics even more complex, posing new challenges to load modeling. However, current load models considered in power system simulation are relatively simple and cannot fully reflect the true dynamic characteristics of loads. Furthermore, due to the strong time-varying nature of loads, models with fixed parameters cannot completely characterize load characteristics. Therefore, it is necessary to establish a load model whose parameters can change in real time according to the load's operating state.
[0003] The application of phasor measurement units and wide-area measurement systems in power systems is becoming increasingly widespread, providing an important data foundation for the establishment of load models and the dynamic correction of load parameters. In selecting power system load models, considering the integration of distributed photovoltaic (PV) systems can more accurately characterize the load characteristics after a large number of PV systems are integrated. In the past, load identification objective functions were mostly single-objective functions, failing to fully consider the different amplitudes of active and reactive power changes under different power system operating scenarios. However, using a dual-objective function can fully extract the characteristics of active and reactive power under different operating scenarios, preventing the characteristics of one power type from being overlooked due to a large difference in their orders of magnitude. Summary of the Invention
[0004] The technical problem to be solved by this invention is: based on the fundamental theory of online correction of integrated load model parameters considering distributed photovoltaic power generation, and based on the multi-objective particle swarm optimization algorithm, this invention provides a method for online correction of integrated load model parameters considering distributed photovoltaic power generation. This method can correct model parameters under both large and small disturbance scenarios, expands the application scenarios of load model correction based on measurement data, and improves the accuracy of model parameter correction by using the multi-objective particle swarm optimization algorithm, thus providing a more accurate load model for power grid simulation calculation.
[0005] To achieve the above-mentioned objectives, the technical solution adopted by this invention is as follows:
[0006] A method for online correction of parameters in a comprehensive load model incorporating distributed photovoltaic (PV) power, wherein the comprehensive load model mainly includes a distribution network model, a dynamic induction motor model, a static load model, reactive power compensation, and distributed PV modules, comprising the following steps:
[0007] Step 1: Obtain real-time operating data of the active power P, reactive power Q, and node voltage U of the load node to be identified from the actual power system;
[0008] Step 2: By setting a data sample classifier, the real-time acquired data samples of active power P, reactive power Q, and node voltage U of the actual power system load nodes are divided into three categories: large disturbance, small disturbance, and no fluctuation.
[0009] Step 3: Perform parameter sensitivity analysis on the integrated load module of distributed photovoltaic power to obtain the dominant parameter {X}. S X D HM LF K M K PV} and load ratio parameter {K M K PV};
[0010] Step 4: Construct the first objective function and the second objective function, where:
[0011] 401. The actual active power data P measured at the load bus of the actual power system is P = [P t1 ,P t1 ,P t2 ,…,P tn As the actual value of load power, the simulated active power data P of the integrated load module of distributed photovoltaic power generation is used. f =[P ft1 ,P ft1 ,P ft2 ,…,P ftn As the calculated load power value, the first objective function is constructed according to the following formula:
[0012] f1 = |PP f |
[0013] 402. The reactive power data Q obtained from the actual measured load bus of the actual power system is Q = [Q t1 Q t1 Q t2 ,…,Q tn As the actual value of the load power, the simulated reactive power data Q of the integrated load module of the distributed photovoltaic system is used. f =[Q ft1 Q ft1 Q ft2 ,…,Q ftn As the calculated load power value, the second objective function is constructed according to the following formula:
[0014] f2 = |QQ f |
[0015] Step 5: By setting a time window, the real-time data of active power P, reactive power Q, and node voltage U of the power system load nodes are segmented to determine the occurrence of large and small disturbances.
[0016] Step 6: When a large disturbance occurs, based on the voltage and power data of the load nodes and the two objective functions in step 3, the dominant parameters of the integrated load module of the distributed photovoltaic system are corrected using a multi-objective particle swarm optimization algorithm, while the other parameters use typical values.
[0017] Step 7: When a small disturbance occurs, based on the voltage and power data of the load nodes and the two objective functions from Step 4, a multi-objective particle swarm optimization algorithm is used to optimize the two proportional parameters {K} of the integrated load model considering distributed photovoltaics. M K PV The values for the parameters were adjusted accordingly, while the remaining parameters used typical values.
[0018] Furthermore, step 4 involves determining the occurrence of large and small disturbances by segmenting the real-time acquired data on the active power P, reactive power Q, and node voltage U of the power system load nodes through a time window; this includes the following steps:
[0019] 501. Subtract the voltage data within each time window from the steady-state voltage value U0 and calculate the absolute value ΔU = [ΔU / U]. t1 ,ΔU t1 ,ΔU t2 ,…,ΔU tn If the absolute value of the difference contains more than 10% of the steady-state voltage data, i.e., ΔU>0.1U0, then a large disturbance is judged to have occurred.
[0020] 502. Subtract the voltage data within each time window from the steady-state voltage value and calculate the absolute value ΔU. If the absolute value of the difference contains data within the 0-10% fluctuation range of the steady-state voltage value, i.e., 0 < ΔU < 0.1U0, then subtract the active power data and reactive power data within each time window respectively, i.e.: ΔP = [ΔP] t1 ,ΔP t1 ,ΔP t2 ,…,ΔP tn ] and ΔQ = [ΔQ t1 ,ΔQ t1 ,ΔQ t2 ,…,ΔQ tn If there is a sudden change in active power and reactive power, that is, after removing noise, there are non-zero values in ΔP and ΔQ, then it is determined that a small disturbance caused by load power fluctuation has occurred.
[0021] 503. After determining whether a large or small disturbance occurs within the time window, extract the data within 1 second from the start of the disturbance within the corresponding time window as valid data to participate in subsequent parameter identification and correction.
[0022] Beneficial effects:
[0023] 1. This invention extends the load model parameter identification method based on measurement data. Compared with the traditional load model parameter identification method based on measurement data, which can only be performed in scenarios where the load bus fluctuation exceeds 10%, this invention can update the load proportional coefficient parameters when the voltage fluctuation caused by the power change of each component of the load is less than 10%. At the same time, it can make full use of power system measurement data and adaptively select the method to update the load parameters online according to different fluctuation scenarios.
[0024] 2. The present invention solves the problem of dynamic parameter updating of integrated load modules containing distributed photovoltaics in the current smart grid context. It can update the load model parameters online in real time, providing a more accurate load model for power system planning and operation. Attached image description:
[0025] Figure 1 This is a flowchart of an online correction method for integrated load model parameters that takes into account distributed photovoltaic power generation, according to the present invention.
[0026] Figure 2 This invention relates to a schematic diagram of an existing integrated load model structure that considers distributed photovoltaics. Detailed Implementation
[0027] The following is in conjunction with the appendix Figure 1-2 The present invention will be described in detail.
[0028] This invention provides a method for online correction of parameters in a comprehensive load model that takes into account distributed photovoltaic power generation, comprising the following steps:
[0029] Step 1: Determine the state of large and small disturbances based on power system load bus measurement data. Using power system measurement devices, read the voltage, active power, and reactive power data of the power system load nodes in real time. Set up a data sample classifier to classify data samples into three categories: large disturbance, small disturbance, and no fluctuation data. Set a time window for data acquisition, with each 10-second window representing the acquired voltage (U = [U...]). t1 U t1 U t2 ,…,U tn ]) and power (P=[P t1 ,P t1 ,P t2 ,…,P tn ],Q=[Q t1Q t1 Q t2 ,…,Q tn The data is segmented. Based on the different analysis results of the data within the time window, the occurrence of large and small disturbances is determined.
[0030] The voltage data within each time window is subtracted from the steady-state voltage value U0, and the absolute value ΔU = [ΔU] is calculated. t1 ,ΔU t1 ,ΔU t2 ,…,ΔU tn If the absolute value of the difference contains data exceeding 10% of the steady-state voltage value U0, i.e., ΔU > 0.1U0, then a large disturbance is determined. The voltage data within each time window is subtracted from the steady-state voltage value, and the absolute value ΔU is calculated. If the absolute value of the difference contains data within the 0-10% fluctuation range of the steady-state voltage value, i.e., 0 < ΔU < 0.1U0, the active power data and reactive power data within each time window are subtracted separately, i.e., ΔP = [ΔP] t1 ,ΔP t1 ,ΔP t2 ,…,ΔP tn ] and ΔQ = [ΔQ t1 ,ΔQ t1 ,ΔQ t2 ,…,ΔQ tn If there is a sudden change in active and reactive power, that is, after removing noise, there are non-zero values in ΔP and ΔQ, then it is determined that a small disturbance caused by load power fluctuation has occurred. Except for the large and small disturbances in the power system mentioned above, other cases are not considered as valid data.
[0031] Once a large or small disturbance is identified within the time window, data from the first second of the disturbance is extracted and used as valid data for subsequent parameter identification and correction.
[0032] Step 2: Analyze the parameters of the integrated load module considering distributed photovoltaic (PV) power, distinguishing between dominant and non-dominant parameters. Analyze the load parameters in the integrated load module considering PV power. Based on the trajectory sensitivity method, build an integrated load module containing PV power in the simulation system and construct transient fault scenarios for simulation analysis. Based on the typical values of each load model parameter, change the parameter value by 5% of the typical value. With other parameters unchanged, calculate the difference in transient power before and after the change of the same parameter and obtain the absolute value. Calculate the magnitude of the load power change when all parameters change by 5%, and classify the load model into dominant and non-dominant parameters based on the magnitude of the change. When the load model parameters change, the parameters with larger changes in load power output are dominant parameters, which are more easily identifiable; when the load model parameters change, the parameters with smaller changes in load power output are non-dominant parameters, which are less easily identifiable. The parameters of the integrated load module considering PV power, as well as the dominant and proportional parameters, are shown in Tables 1 and 2.
[0033] Table 1. Specifications of Integrated Load Modules Considering Distributed Photovoltaics
[0034]
[0035]
[0036] Table 2 Classification of dominant and proportional parameters of integrated load modules considering distributed photovoltaic power generation.
[0037] Parameter classification Dominant parameters proportional parameters Parameter symbol <![CDATA[X S X D H M LF K M K PV ]]> <![CDATA[K M K PV ]]>
[0038] Step 3: Construct the objective function. Using the actual active power data measured at the power system load bus, P = [P...] t1 ,P t1 ,P t2 ,…,P tn ] and reactive power data Q = [Q t1 Q t1 Q t2 ,…,Q tn The actual value of the load power is used as the load active power data P obtained from power system simulation under the same system operating conditions. f =[P ft1 ,P ft1 ,P ft2 ,…,P ftn ] and reactive power data Q f =[Q ft1 Q ft1 Q ft2 ,…,Q ftn The calculated value is the difference between the actual and calculated active power at the load node. The first objective function is: f1 = |PP|.f The second objective function is the difference between the actual and calculated reactive power values at the load nodes, i.e.: f2 = |QQ| f The smaller the values of the two objective functions, the closer the calculated and actual power values are. When the simulation conditions are the same as the actual power system operating conditions, smaller objective function values indicate that the load characteristics represented by the load parameters used in the simulation are closer to the load characteristics in the actual system, meaning the load parameters are more accurate.
[0039] Step 4: Online correction of model parameters for large disturbance conditions in the power system. When a large disturbance is detected, the system is simulated to simulate the large disturbance condition. The simulated fault parameters are exactly the same as the actual fault parameters. First, 30 sets of model dominant parameter combinations are randomly generated within the range of typical parameter changes of the integrated load module considering distributed photovoltaics. The 30 sets of dominant parameter combinations are then substituted into the load model for large disturbance fault simulation. The active and reactive power outputs of the load nodes within 1 second of the fault are substituted into the objective function in (3) to calculate the two objective function values of each dominant parameter combination under the current large disturbance fault. The multi-objective particle swarm algorithm is adopted. The objective function adopts the two objective functions f1 and f2 in (3). The multi-objective particle swarm algorithm calculates the 30 sets of dominant parameter combinations each time, calculates the optimal Pareto dominant parameter combination for each iteration and archives it. The current optimal Pareto dominant parameter combination is compared with the previously archived Pareto dominant parameter combination. The dominant is selected according to the dominance principle and the optimal Pareto dominant parameter combination is updated. The algorithm is set to calculate 100 iterations, and the optimal Pareto dominant parameter combination after 100 iterations is used as the dominant load parameter value under the current large disturbance.
[0040] Step 5: When a small disturbance is detected, the system is simulated using simulation. The simulated fault parameters are exactly the same as the actual fault parameters. First, 30 sets of model proportional parameter combinations are randomly generated within the range of typical parameter changes of the integrated load module of distributed photovoltaic. The 30 sets of dominant parameter combinations are then substituted into the load model for small disturbance fault simulation. The active and reactive power outputs of the load nodes within 1 second of the fault are substituted into the objective function in (3) to calculate the two objective function values of each dominant parameter combination under the current small disturbance fault. The multi-objective particle swarm algorithm is used, and the objective functions are the two objective functions f1 and f2 in (3). The multi-objective particle swarm algorithm calculates the 30 sets of dominant parameter combinations each time, calculates the optimal Pareto dominant parameter combination for each iteration and archives it. The current optimal Pareto dominant parameter combination is compared with the previously archived Pareto dominant parameter combination. The dominant is selected according to the dominance principle, and the optimal Pareto dominant parameter combination is updated. The algorithm is set to calculate 100 iterations, and the optimal Pareto dominant parameter combination after 100 iterations is used as the dominant load parameter value for the current small disturbance.
[0041] References:
[0042] [1]CACCoello,GTPulido and MSLechuga,"Handling multipleobjectives with particle swarm optimization,"in IEEE Transactions onEvolutionary Computation,vol.8,no.3,pp.256-279,June 2004
[0043] [2]Y,Ma WJ,Zhang JP,et al.The Parameters Selection of PSO Algorithminfluencing On performance of Fault Diagnosis[J].Matec Web of Conferences,2016,63.
[0044] [3]IEEE Guide for Load Modeling and Simulations for Power Systems," inIEEE Std 2781-2022, vol., no., pp.1-88, 30Sept.2022.
Claims
1. A method for online correction of parameters in a comprehensive load model considering distributed photovoltaic (PV) power generation, wherein the comprehensive load model mainly includes a distribution network model, a dynamic induction motor model, a static load model, reactive power compensation, and distributed PV power generation, characterized in that... Includes the following steps: Step 1: Obtain the active power of the load node to be identified from the actual power system. P reactive power Q and node voltage U Real-time operational data; Step 2: By setting up a data sample classifier, the real-time active power of the actual power system load nodes is obtained. P reactive power Q and node voltage U The data samples are divided into three categories: large disturbances, small disturbances, and no fluctuations. Step 3: Perform parameter sensitivity analysis on the integrated load model of distributed photovoltaic power to obtain the dominant parameter {X}. S X D HM LF K M K PV } and load ratio parameter {K M K PV }; Where: X S Indicates the stator reactance of the electric motor; X D H represents the distribution network reactance; H represents the inertial time constant of the induction motor; M represents the inertial time constant of the induction motor. LF K represents the load rate of the induction motor. M K represents the proportion of static load active power to total load active power; PV This represents the ratio of active power of distributed photovoltaic systems to active power of the load. Step 4: Construct the first objective function and the second objective function, where:
401. The active power data actually measured at the load bus of the actual power system. P =[ P t1 , P t1 , P t2 ,…, P tn As the actual value of load power, the simulated active power data of the integrated load module of distributed photovoltaic power generation will be used. P f =[ P ft1 , P ft1 , P ft2 ,…, P ftn As the calculated load power value, the first objective function is constructed according to the following formula: f 1 =| P - P f | 402. The reactive power data actually measured at the load bus of the actual power system. Q =[ Q t1 , Q t1 , Q t2 ,…, Q tn As the actual value of load power, the simulated reactive power data of the integrated load module of distributed photovoltaic power generation will be used. Q f =[ Q ft1 , Q ft1 , Q ft2 ,…, Q ftn As the calculated load power value, the second objective function is constructed according to the following formula: f 2 =| Q - Q f | Step 5: Set a time window to monitor the active power of the power system load nodes in real time. P reactive power Q and node voltage U The data is segmented to determine the occurrence of large and small disturbances; Step 6: When a large disturbance occurs, based on the voltage and power data of the load nodes and the two objective functions in Step 4, the dominant parameters of the integrated load model of distributed photovoltaics are corrected using the multi-objective particle swarm algorithm, and the other parameters are taken as typical values. Step 7: When a small disturbance occurs, based on the voltage and power data of the load nodes and the two objective functions in Step 4, the two proportional parameters of the integrated load model considering distributed photovoltaics are corrected using a multi-objective particle swarm algorithm. The remaining parameters are all taken as typical values.
2. The method for online correction of parameters of a comprehensive load model considering distributed photovoltaic power according to claim 1, characterized in that, In step 5, the active power of the power system load nodes acquired in real time is monitored by setting a time window. P reactive power Q and node voltage U The process of segmenting data to determine the occurrence of large and small disturbances includes the following steps:
501. Compare the voltage data within each time window with the steady-state voltage value. U 0 Find the difference and calculate the absolute value Δ U =[Δ U t1 ,Δ U t1 ,Δ U t2 ,…,Δ U tn If the absolute value of the difference contains more than 10% of the steady-state voltage data, that is: Δ U >0.1 U 0 If so, it is determined that a large disturbance has occurred; 502. Subtract the voltage data within each time window from the steady-state voltage value and calculate the absolute value Δ. U If the absolute value of the obtained difference contains data within the voltage steady-state value fluctuation range of 0-10%, i.e.: 0 < Δ U <0.1 U 0 Simultaneously, the active power data and reactive power data within each time window are subtracted, i.e.: Δ P =[Δ P t1 ,Δ P t1 ,Δ P t2 ,…,Δ P tn ] and Δ Q =[Δ Q t1 ,Δ Q t1 ,Δ Q t2 ,…,Δ Q tn If there is a sudden change in active and reactive power, i.e. after noise removal, Δ P and Δ Q If there is a non-zero value in the value, it is determined that a small disturbance caused by load power fluctuation has occurred; 503. After determining whether a large or small disturbance occurs within the time window, extract the data within 1 second from the start of the disturbance within the corresponding time window as valid data to participate in subsequent parameter identification and correction.
