CN117421972A - Identification method and terminal for comprehensive load model parameters - Google Patents
Identification method and terminal for comprehensive load model parameters Download PDFInfo
- Publication number
- CN117421972A CN117421972A CN202311211200.7A CN202311211200A CN117421972A CN 117421972 A CN117421972 A CN 117421972A CN 202311211200 A CN202311211200 A CN 202311211200A CN 117421972 A CN117421972 A CN 117421972A
- Authority
- CN
- China
- Prior art keywords
- model
- parameter
- comprehensive load
- parameters
- curve
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 32
- 230000004044 response Effects 0.000 claims abstract description 107
- 238000005259 measurement Methods 0.000 claims abstract description 81
- 238000012937 correction Methods 0.000 claims abstract description 54
- 238000013135 deep learning Methods 0.000 claims abstract description 11
- 238000004088 simulation Methods 0.000 claims description 21
- 238000005070 sampling Methods 0.000 claims description 15
- 238000004590 computer program Methods 0.000 claims description 8
- 230000003068 static effect Effects 0.000 claims description 6
- 230000006978 adaptation Effects 0.000 claims description 5
- 238000013459 approach Methods 0.000 claims description 5
- 230000008569 process Effects 0.000 abstract description 8
- 238000004422 calculation algorithm Methods 0.000 description 14
- 238000005457 optimization Methods 0.000 description 11
- 230000009471 action Effects 0.000 description 9
- 230000002787 reinforcement Effects 0.000 description 9
- 238000012549 training Methods 0.000 description 9
- 239000003795 chemical substances by application Substances 0.000 description 8
- 230000003993 interaction Effects 0.000 description 4
- 230000009286 beneficial effect Effects 0.000 description 3
- 238000006243 chemical reaction Methods 0.000 description 3
- 210000002569 neuron Anatomy 0.000 description 3
- 238000004364 calculation method Methods 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 230000006870 function Effects 0.000 description 2
- 238000000342 Monte Carlo simulation Methods 0.000 description 1
- 230000003044 adaptive effect Effects 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000003111 delayed effect Effects 0.000 description 1
- 238000001914 filtration Methods 0.000 description 1
- 230000004927 fusion Effects 0.000 description 1
- 230000006698 induction Effects 0.000 description 1
- 238000002347 injection Methods 0.000 description 1
- 239000007924 injection Substances 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000001052 transient effect Effects 0.000 description 1
Landscapes
- Supply And Distribution Of Alternating Current (AREA)
Abstract
The invention discloses a method and a terminal for identifying parameters of a comprehensive load model, wherein a model response curve and a parameter mode library of the comprehensive load model are established, and initial parameters are determined according to a matching result of real-time measurement information and the mode library during online identification, so that the efficiency of subsequent optimizing iteration can be improved; and then, constructing a parameter correction model of the comprehensive load model based on deep learning, and carrying out iterative correction on initial parameters according to real-time measurement information so as to determine a final model parameter identification result. Therefore, the parameter identification is carried out by combining matching and correction, the efficiency of the parameter identification can be improved, and in the process of correcting the model, the global optimizing capability can be enlarged, the problem that the optimizing iteration is stopped when the model falls into the local optimum is avoided, so that a more accurate comprehensive load model parameter identification result is obtained.
Description
Technical Field
The invention relates to the technical field of parameter identification, in particular to an identification method and terminal for comprehensive load model parameters.
Background
The research on the stability of the power system is mainly based on a simulation analysis means, wherein the accuracy of a load model has an important influence on the simulation result of the system. However, as a large number of power electronic devices such as distributed power supplies are connected, the power system load presents characteristics of variability and complexity, and accurate modeling is difficult to perform. Therefore, it is necessary to perform adaptive equivalence modeling according to the external load characteristics, and the parameter identification of the load model is an important task, so as to determine the relevant parameters of the model as accurately as possible.
The power system element model is generally composed of model inputs, model structures, model parameters, model outputs and the like, and when the structures and inputs are the same, the model parameters and the outputs generally have a one-to-one mapping relationship. Because the structure, input and output of the model are known or can be obtained by measurement, the unknown model parameters can be reversely solved according to the model output. For relatively complex load models, it is common to give them different parameters, the same input and generate multiple sets of outputs, and continuously adjust the parameters until a set of parameters is screened that minimizes the difference between the model output and the actual measured output, which is used as the final result of the parameter identification. It follows that parameter identification is essentially an iterative optimization process.
Therefore, some filtering algorithms or artificial intelligence algorithms are usually adopted for parameter identification, but for a comprehensive load model containing a distributed power supply, a high-dimensional nonlinear relation exists between model parameters and response output, identification difficulty and complexity are high, and multiple groups of parameters possibly exist to enable the model output to be close to actual output. In the prior art, when the problems of high nonlinearity and large parameter space dimension are solved, the algorithm efficiency is low, the requirement of an online time scale is difficult to adapt, and the optimization iteration is stopped easily when local optimization occurs in the parameter identification process, so that the identification result is not accurate enough.
Disclosure of Invention
The technical problems to be solved by the invention are as follows: the method and the terminal for identifying the comprehensive load model parameters can improve the accuracy of parameter identification and improve the efficiency of online identification.
In order to solve the technical problems, the invention adopts the following technical scheme:
a method for identifying comprehensive load model parameters comprises the following steps:
establishing a comprehensive load model according to the structure of the load to be identified, and obtaining a response curve of the comprehensive load model and a model parameter model library;
matching the measurement information acquired in real time with the mode library, and determining initial parameters of the comprehensive load model according to a matching result;
and constructing a parameter correction model of the comprehensive load model based on deep learning, inputting measurement information acquired in real time into the parameter correction model, and carrying out iterative correction on the initial parameters to obtain a parameter identification result.
In order to solve the technical problems, the invention adopts another technical scheme that:
an identification terminal for comprehensive load model parameters, comprising a memory, a processor and a computer program stored on the memory and capable of running on the processor, wherein the processor realizes the following steps when executing the computer program:
establishing a comprehensive load model according to the structure of the load to be identified, and obtaining a response curve of the comprehensive load model and a model parameter model library;
matching the measurement information acquired in real time with the mode library, and determining initial parameters of the comprehensive load model according to a matching result;
and constructing a parameter correction model of the comprehensive load model based on deep learning, inputting measurement information acquired in real time into the parameter correction model, and carrying out iterative correction on the initial parameters to obtain a parameter identification result.
The invention has the beneficial effects that: establishing a model response curve and a parameter mode library of the comprehensive load model, and determining initial parameters according to the matching result of the real-time measurement information and the mode library during online identification, so that the efficiency of subsequent optimizing iteration can be improved; and then, constructing a parameter correction model of the comprehensive load model based on deep learning, and carrying out iterative correction on initial parameters according to real-time measurement information so as to determine a final model parameter identification result. Therefore, the parameter identification is carried out by combining matching and correction, the efficiency of the parameter identification can be improved, and in the process of correcting the model, the global optimizing capability can be enlarged, the problem that the optimizing iteration is stopped when the model falls into the local optimum is avoided, so that a more accurate comprehensive load model parameter identification result is obtained.
Drawings
FIG. 1 is a flowchart of a method for identifying parameters of a comprehensive load model according to an embodiment of the present invention;
FIG. 2 is a schematic diagram of an identification terminal for integrated load model parameters according to an embodiment of the present invention;
FIG. 3 is a flowchart showing the specific steps of a method for identifying parameters of a comprehensive load model according to an embodiment of the present invention;
fig. 4 is a framework diagram of a load model parameter identification method based on deep reinforcement learning according to an embodiment of the invention.
Description of the reference numerals:
1. an identification terminal for comprehensive load model parameters; 2. a memory; 3. a processor.
Detailed Description
In order to describe the technical contents, the achieved objects and effects of the present invention in detail, the following description will be made with reference to the embodiments in conjunction with the accompanying drawings.
Referring to fig. 1, an embodiment of the present invention provides a method for identifying parameters of a comprehensive load model, including the steps of:
establishing a comprehensive load model according to the structure of the load to be identified, and obtaining a response curve of the comprehensive load model and a model parameter model library;
matching the measurement information acquired in real time with the mode library, and determining initial parameters of the comprehensive load model according to a matching result;
and constructing a parameter correction model of the comprehensive load model based on deep learning, inputting measurement information acquired in real time into the parameter correction model, and carrying out iterative correction on the initial parameters to obtain a parameter identification result.
From the above description, the beneficial effects of the invention are as follows: establishing a model response curve and a parameter mode library of the comprehensive load model, and determining initial parameters according to the matching result of the real-time measurement information and the mode library during online identification, so that the efficiency of subsequent optimizing iteration can be improved; and then, constructing a parameter correction model of the comprehensive load model based on deep learning, and carrying out iterative correction on initial parameters according to real-time measurement information so as to determine a final model parameter identification result. Therefore, the parameter identification is carried out by combining matching and correction, the efficiency of the parameter identification can be improved, and in the process of correcting the model, the global optimizing capability can be enlarged, the problem that the optimizing iteration is stopped when the model falls into the local optimum is avoided, so that a more accurate comprehensive load model parameter identification result is obtained.
Further, the building the comprehensive load model according to the structure of the load to be identified includes:
establishing a comprehensive load model which is the same as a load structure to be identified, wherein the comprehensive load model comprises a motor load, a static load and a distributed photovoltaic, and setting initial parameters in the value range of each parameter of the comprehensive load model: θ i ∈[θ imin ，θ imax ]；
In θ i Represents the initial value of the ith parameter, θ imin And theta imax The lower limit and the upper limit of the value range of the ith parameter are respectively indicated.
From the above description, it can be seen that the comprehensive load model is constructed according to the structure of the load to be identified, so that the parameter on-line identification can be conveniently performed for the comprehensive load model.
Further, obtaining the response curve of the comprehensive load model and the model parameter model library comprises:
inputting different working condition data of a preset group into the comprehensive load model to obtain a power response output curve, a voltage response output curve and a parameter group of the comprehensive load model under different working conditions of the preset group;
classifying parameter sets of a preset group of comprehensive load models, and taking a response track of each sample in each classification as a mode response value of each classification;
and taking the average model parameters of all the classifications as model parameters of all the classifications, and obtaining a model parameter model library of the comprehensive load model according to the model parameters of all the classifications and the model response values of all the classifications.
From the above description, it can be seen that different parameter sets of the model are obtained according to different working conditions, the parameter sets are classified, and a typical model library is obtained according to model parameters and model response values in each classification, so that initial iteration parameters of the model can be obtained according to the model library.
Further, inputting measurement information obtained in real time into the comprehensive load model to obtain a measurement response curve, wherein the measurement response curve comprises a measurement voltage curve and a measurement power curve;
matching the measurement response curve with a power response output curve and a voltage response output curve, and selecting the power response output curve and the voltage response output curve with the minimum vector distance with the measurement response curve as matching results:
wherein D represents a vector distance, V i And V j Data values respectively representing a measured voltage curve and a voltage response output curve at time t, P i And P j The data values of the measured power curve and the power response output curve at time T are respectively shown, and T represents the total data values at time T.
From the above description, it can be known that by constructing a typical pattern library and determining initial parameters for iterative optimization according to the matching result with the typical pattern library, the accuracy and efficiency of parameter identification can be improved.
Further, inputting the measurement information obtained in real time into the parameter correction model, and performing iterative correction on the initial parameters to obtain a parameter identification result, where the parameter identification result includes:
inputting the initial parameters obtained by matching and the measurement information obtained in real time into a parameter correction model, and carrying out iterative correction on the initial parameters to obtain a group of parameters with the minimum adaptation degree of model output and the measurement information obtained in real time as a model parameter identification result:
wherein F is d An adaptability index for indicating the approach degree of the model output and the measurement information acquired in real time, n represents the number of sampling points and P N And V N Rated value, P, of active power and voltage respectively j And V j The simulation values of the active power and the current at the sampling point j are respectively, P j0 And V j0 The active power and the actual voltage at the sampling point j are shown, respectively.
From the description, the model initial parameters are corrected by adopting the deep reinforcement learning TD3 algorithm, so that the time scale requirement of on-line identification of the model parameters can be met, the global optimizing capability can be expanded, the local optimization is avoided, and a more accurate comprehensive load model parameter identification result is obtained.
Referring to fig. 2, another embodiment of the present invention provides an identification terminal for comprehensive load model parameters, including a memory, a processor, and a computer program stored in the memory and capable of running on the processor, wherein the processor implements the following steps when executing the computer program:
establishing a comprehensive load model according to the structure of the load to be identified, and obtaining a response curve of the comprehensive load model and a model parameter model library;
matching the measurement information acquired in real time with the mode library, and determining initial parameters of the comprehensive load model according to a matching result;
and constructing a parameter correction model of the comprehensive load model based on deep learning, inputting measurement information acquired in real time into the parameter correction model, and carrying out iterative correction on the initial parameters to obtain a parameter identification result.
From the above description, the beneficial effects of the invention are as follows: establishing a model response curve and a parameter mode library of the comprehensive load model, and determining initial parameters according to the matching result of the real-time measurement information and the mode library during online identification, so that the efficiency of subsequent optimizing iteration can be improved; and then, constructing a parameter correction model of the comprehensive load model based on deep learning, and carrying out iterative correction on initial parameters according to real-time measurement information so as to determine a final model parameter identification result. Therefore, the parameter identification is carried out by combining matching and correction, the efficiency of the parameter identification can be improved, and in the process of correcting the model, the global optimizing capability can be enlarged, the problem that the optimizing iteration is stopped when the model falls into the local optimum is avoided, so that a more accurate comprehensive load model parameter identification result is obtained.
Further, the building the comprehensive load model according to the structure of the load to be identified includes:
establishing a comprehensive load model which is the same as a load structure to be identified, wherein the comprehensive load model comprises a motor load, a static load and a distributed photovoltaic, and setting initial parameters in the value range of each parameter of the comprehensive load model: θ i ∈[θ imin ，θ imax ]；
In θ i Represents the initial value of the ith parameter, θ imin And theta imax The lower limit and the upper limit of the value range of the ith parameter are respectively indicated.
From the above description, it can be seen that the comprehensive load model is constructed according to the structure of the load to be identified, so that the parameter on-line identification can be conveniently performed for the comprehensive load model.
Further, obtaining the response curve of the comprehensive load model and the model parameter model library comprises:
inputting different working condition data of a preset group into the comprehensive load model to obtain a power response output curve, a voltage response output curve and a parameter group of the comprehensive load model under different working conditions of the preset group;
classifying parameter sets of a preset group of comprehensive load models, and taking a response track of each sample in each classification as a mode response value of each classification;
and taking the average model parameters of all the classifications as model parameters of all the classifications, and obtaining a model parameter model library of the comprehensive load model according to the model parameters of all the classifications and the model response values of all the classifications.
From the above description, it can be seen that different parameter sets of the model are obtained according to different working conditions, the parameter sets are classified, and a typical model library is obtained according to model parameters and model response values in each classification, so that initial iteration parameters of the model can be obtained according to the model library.
Further, inputting measurement information obtained in real time into the comprehensive load model to obtain a measurement response curve, wherein the measurement response curve comprises a measurement voltage curve and a measurement power curve;
matching the measurement response curve with a power response output curve and a voltage response output curve, and selecting the power response output curve and the voltage response output curve with the minimum vector distance with the measurement response curve as matching results:
wherein D represents a vector distance, V i And V j Data values respectively representing a measured voltage curve and a voltage response output curve at time t, P i And P j The data values of the measured power curve and the power response output curve at time T are respectively shown, and T represents the total data values at time T.
From the above description, it can be known that by constructing a typical pattern library and determining initial parameters for iterative optimization according to the matching result with the typical pattern library, the accuracy and efficiency of parameter identification can be improved.
Further, inputting the measurement information obtained in real time into the parameter correction model, and performing iterative correction on the initial parameters to obtain a parameter identification result, where the parameter identification result includes:
inputting the initial parameters obtained by matching and the measurement information obtained in real time into a parameter correction model, and carrying out iterative correction on the initial parameters to obtain a group of parameters with the minimum adaptation degree of model output and the measurement information obtained in real time as a model parameter identification result:
wherein F is d An adaptability index for indicating the approach degree of the model output and the measurement information acquired in real time, n represents the number of sampling points and P N And V N Rated value, P, of active power and voltage respectively j And V j The simulation values of the active power and the current at the sampling point j are respectively, P j0 And V j0 The active power and the actual voltage at the sampling point j are shown, respectively.
From the above description, the initial parameters of the model are corrected by deep reinforcement learning, so that the time scale requirement of on-line identification of the model parameters can be met, the global optimizing capability can be expanded, the local optimization is avoided, and a more accurate comprehensive load model parameter identification result is obtained.
The method and the terminal for identifying the parameters of the comprehensive load model are suitable for improving the accuracy of parameter identification and improving the efficiency of online identification when the problem of parameter identification of a complex comprehensive load model is solved. The following description is made by way of specific embodiments:
example 1
Referring to fig. 1, a method for identifying parameters of a comprehensive load model includes the steps of:
s1, building a comprehensive load model according to a structure of a load to be identified, and obtaining a response curve of the comprehensive load model and a model parameter model library.
S11, establishing a comprehensive load model with the same load structure as the load to be identified, wherein the comprehensive load model comprises a motor load, a static load and distributed photovoltaic, and setting initial parameters in the value range of each parameter of the comprehensive load model:
θ i ∈[θ imin ，θ imax ]；
in θ i Represents the initial value of the ith parameter, θ imin And theta imax The lower limit and the upper limit of the value range of the ith parameter are respectively indicated.
S12, inputting different working condition data of a preset group into the comprehensive load model to obtain a power response output curve, a voltage response output curve and a parameter group of the comprehensive load model under different working conditions of the preset group.
Specifically, the Monte Carlo method is adopted to randomly set the conditions of the load level, the node injection power, the fault position and the like of the power system, simulation is carried out aiming at the different working conditions, and the power and voltage response output curves of the load models under N groups of different working conditions and the corresponding load model parameter groups are obtained.
S13, classifying parameter sets of the comprehensive load model of the preset group, and taking a response track of each sample in each classification as a mode response value of each classification.
Specifically, N groups of responses are classified, and a classification set M is set 1 ＝{S i I=1, 2, …, N }, classification setBase set->Threshold h, sequentially selecting S i ∈M 1 Put into the base point set M as the initial base point 3 And from M 1 Is removed.
Wherein V is i And V j Data values respectively representing a measured voltage curve and a voltage response output curve at time t, P i And P j Data values of the measured power curve and the power response output curve at time t are respectively represented.
According to the above calculation class set M 1 The correlation coefficient r of all responses in (a) and the current base point ij If the correlation coefficient is greater than the threshold h, the correlation coefficient is selected from the classification set M 1 Move to the classification set M 2 Taking a classification set M 1 The element with the smallest correlation coefficient is taken as a new base point and is moved into the base point set M 3 。
Collect the base points M 3 Each base point sample is taken as a class, and each classification set M 2 And (3) respectively comparing the correlation coefficients of the samples with the base points, and classifying the samples into the class corresponding to the base point with the maximum correlation coefficient, so that N groups of responses are classified into M classes.
Taking the response track of each sample in M classes as S i Typical pattern responses of the class.
S14, taking the average model parameters of all the classifications as model parameters of all the classifications, and obtaining a model parameter model library of the comprehensive load model according to the model parameters of all the classifications and the model response values of all the classifications.
Specifically, average model parameters of various scene samples are used as typical model parameters to which the types belong, and various typical model responses and the typical model parameters form a typical model library of load model responses and parameters thereof.
And S2, matching the measurement information acquired in real time with the mode library, and determining initial parameters of the comprehensive load model according to a matching result.
Specifically, inputting measurement information obtained in real time into the comprehensive load model to obtain a measurement response curve, wherein the measurement response curve comprises a measurement voltage curve and a measurement power curve;
matching the measurement response curve with a power response output curve and a voltage response output curve, and selecting the power response output curve and the voltage response output curve with the minimum vector distance with the measurement response curve as matching results:
wherein D represents a vector distance, V i And V j Data values respectively representing a measured voltage curve and a typical voltage response output curve at time t, P i And P j The data values of the measured power curve and the typical power response output curve at time T are respectively represented, and T represents the data values at a total of T times.
S3, constructing a parameter correction model of the comprehensive load model based on deep learning, inputting measurement information acquired in real time into the parameter correction model, and carrying out iterative correction on the initial parameters to obtain a parameter identification result.
S31, constructing an online identification framework of the comprehensive load model based on deep reinforcement learning, realizing model parameter identification through a deep reinforcement learning algorithm, and applying the intelligent body with the learned parameter identification experience to an online environment to realize online rapid identification of parameters. The specific structure is shown in fig. 4, and is mainly divided into 5 modules.
(1) A simulation system. The simulation system module simulates different parameters to be identified which are input by the interface module under the condition that other parameters of the simulation system are kept unchanged, and feeds back observed quantity obtained by simulation to the interface module. Meanwhile, in the absence of historical scenes, the simulation system module can provide supplementary scenes through offline simulation to enrich the scene library.
(2) An interface. The interface model realizes the conversion among observed quantity, parameters to be identified, historical (measurement) observed quantity, interaction state of original parameters and intelligent agents, rewards and actions of the simulation module through three conversion submodules. The interface module is a key for realizing the fusion of the deep reinforcement learning algorithm and the parameter identification problem, and is an important component of an offline training model and an online application module, but the interface module in the online application omits the conversion of rewards because the interface module does not need training in the online application.
(3) And (5) offline training. The offline training module interacts with the simulation model through the interface model to train the intelligent agent to recognize model parameters in different scenes in the scene library, learn model parameter recognition experience, and update the intelligent agent in the online application module regularly.
(4) And (5) online application. The online application module interacts with the simulation model through an interface, and carries out online quick identification on model parameters according to online acquired observables based on trained agents, and gives out the identified parameters.
(5) A scene library. The scene library module stores training samples for different scenes of the agent. The simulation system parameters including the model parameters and the corresponding historical observables in the scene samples can be provided and updated by the historical scenes of the actual system, and meanwhile, in order to cope with the problem of insufficient number of the historical scenes, the simulation system parameters and the corresponding historical observables can be supplemented through simulation.
The depth deterministic strategy gradient algorithm is a depth reinforcement learning algorithm based on an Actor-Critic mode, wherein the TD3 algorithm is realized by changing an Actor and Critic in the Actor-Critic method into a neural network, wherein the Actor network is used for generating an action strategy, and the Critic network is used for estimating a cost function of the current strategy. The specific interface scheme for carrying out load model parameter identification by adopting the TD3 algorithm is as follows:
(1) action interface: the load model parameter set is a column vector C, and the intelligent agent needs to correct C according to feedback of the environment and the current state selection step length, so that the interaction relation between the action and the environment is as follows: c=c+a, where a is a column vector of the same scale as C, the upper and lower limits of which are determined according to the circumstances.
(2) Status interface: under specific working conditions, different model parameters C correspond to different dynamic response curves o (C), so that the identified model parameters are close to the real parameter sets C corresponding to the actual response curves 0 State s needs to contain C at the same time 0 And information of the existing identification value C: s= (o (C), C).
(3) A reward interface: the influence of the reward function on the training effect is maximum, in order to enable the estimated value to be continuously close to the true value in training, the corresponding dynamic response curve error is required to be used as feedback, the reward with small error is large, and meanwhile, the exploration step length of the agent is required to be punished, so that the exploration density of the agent is increased, and therefore, the reward is set as follows:
wherein: o (C) target )、o(C 0 ) And o (C) respectively represent a target curve, an output curve corresponding to the real parameter and an output curve corresponding to the current parameter.
After each interaction is completed, corresponding current state and action data are recorded in the experience pool, and in the subsequent iteration process, a plurality of records are randomly extracted from the experience pool, and the Actor and Critic network are trained according to the above formula.
S32, inputting the initial parameters obtained by matching and the measurement information obtained in real time into a parameter correction model, and performing iterative correction on the initial parameters to obtain a group of parameters with the minimum adaptation degree between model output and the measurement information obtained in real time as a model parameter identification result:
wherein F is d An adaptability index for indicating the approach degree of the model output and the measurement information acquired in real time, n represents the number of sampling points and P N And V N Representing active power and voltage, respectivelyRated value of P j And V j The simulation values of the active power and the current at the sampling point j are respectively, P j0 And V j0 The active power and the actual voltage at the sampling point j are shown, respectively.
And selecting the internal parameters of the comprehensive load model under the node 2 as identification objects in the WSCC-9 node calculation example, wherein the disturbance parameters of each interaction are randomly changed to simulate the disturbance level of an actual power grid, and the comprehensive load comprises static load, induction motor load and distributed photovoltaic.
The state interface collects voltage, active and reactive curves output by transient stability simulation of the system under small interference, records 75 cycle data and current parameter generation states, and all state space dimensions 249; the action range of the action interface is [ -0.1,0.1]. Using a TD3 algorithm, for a Critic network, connecting the state and the action through the full-connection layers with the number of 2 layers of neurons of 256 and 64 respectively, and outputting through the full-connection layer with the number of 2 layers of neurons of 512; for the Actor network, the state is output through the output layer after passing through 4 full-connection layers, and the number of neurons of the full-connection layers is 256, 512 and 512 respectively. 10 steps are set for correction per round during training, and Gaussian white noise with 1% standard deviation is introduced to output actions per step. After 1000 training rounds, the average prize stage by stage converged to around 20% of the initial average error. Comparing the comprehensive load model parameter results obtained by the method of the embodiment with those obtained by other existing methods, as shown in table 1, the comparison results show that the model parameter identification result obtained by the method of the embodiment has smaller error.
Table 1 comprehensive load parameter identification results using different methods
/>
Therefore, in this embodiment, by constructing the typical library, determining the initial parameters of iterative optimization according to the matching result with the typical pattern library, the accuracy and efficiency of parameter identification can be improved; the model initial parameters are corrected by adopting a depth reinforcement learning TD3 algorithm, so that the time scale requirement of on-line identification of the model parameters can be met, the global optimizing capability can be enlarged, the local optimization is avoided, and a more accurate comprehensive load model parameter identification result is obtained.
Example two
Referring to fig. 2, an identification terminal 1 for integrated load model parameters includes a memory 2, a processor 3, and a computer program stored in the memory 2 and executable on the processor 3, wherein the processor 3 implements the steps of an identification method for integrated load model parameters according to the first embodiment when executing the computer program.
In summary, according to the method and the terminal for identifying the parameters of the comprehensive load model, when the parameter identification problem of the complex comprehensive load model is processed in the prior art, the algorithm efficiency is low and is difficult to adapt to the requirement of an online time scale due to the high-dimensional nonlinear characteristic of the model parameters, and the optimization iteration is stopped easily when local optimization occurs in the parameter identification process, so that the identification result is inaccurate. In order to solve the problems, the invention firstly establishes an offline mode library of a model response curve and corresponding parameters thereof, determines initial iteration parameters according to a matching result with a typical mode library during online identification, and improves optimizing iteration efficiency; then, a comprehensive load model parameter correction framework model is constructed based on a double-delay depth deterministic strategy gradient algorithm (twin delayed deep deterministic policy gradient, TD 3) of the depth reinforcement learning, and initial parameters are subjected to iterative correction according to online real-time measurement information, so that a final model parameter identification result is determined.
The foregoing description is only illustrative of the present invention and is not intended to limit the scope of the invention, and all equivalent changes made by the specification and drawings of the present invention, or direct or indirect application in the relevant art, are included in the scope of the present invention.
Claims (10)
1. The method for identifying the comprehensive load model parameters is characterized by comprising the following steps:
establishing a comprehensive load model according to the structure of the load to be identified, and obtaining a response curve of the comprehensive load model and a model parameter model library;
matching the measurement information acquired in real time with the mode library, and determining initial parameters of the comprehensive load model according to a matching result;
and constructing a parameter correction model of the comprehensive load model based on deep learning, inputting measurement information acquired in real time into the parameter correction model, and carrying out iterative correction on the initial parameters to obtain a parameter identification result.
2. The method for identifying parameters of a comprehensive load model according to claim 1, wherein the building the comprehensive load model according to the structure of the load to be identified comprises:
establishing a comprehensive load model which is the same as a load structure to be identified, wherein the comprehensive load model comprises a motor load, a static load and a distributed photovoltaic, and setting initial parameters in the value range of each parameter of the comprehensive load model: θ i ∈[θ imin ，θ imax ]；
In θ i Represents the initial value of the ith parameter, θ imin And theta imax The lower limit and the upper limit of the value range of the ith parameter are respectively indicated.
3. The method for identifying parameters of a comprehensive load model according to claim 1, wherein obtaining a response curve of the comprehensive load model and a model parameter pattern library comprises:
inputting different working condition data of a preset group into the comprehensive load model to obtain a power response output curve, a voltage response output curve and a parameter group of the comprehensive load model under different working conditions of the preset group;
classifying parameter sets of a preset group of comprehensive load models, and taking a response track of each sample in each classification as a mode response value of each classification;
and taking the average model parameters of all the classifications as model parameters of all the classifications, and obtaining a model parameter model library of the comprehensive load model according to the model parameters of all the classifications and the model response values of all the classifications.
4. The method for identifying parameters of a comprehensive load model according to claim 3, wherein the step of matching the measurement information obtained in real time with the pattern library, and determining initial parameters of the comprehensive load model according to the matching result comprises:
inputting measurement information obtained in real time into the comprehensive load model to obtain a measurement response curve, wherein the measurement response curve comprises a measurement voltage curve and a measurement power curve;
matching the measurement response curve with a power response output curve and a voltage response output curve, and selecting the power response output curve and the voltage response output curve with the minimum vector distance with the measurement response curve as matching results:
wherein D represents a vector distance, V i And V j Data values respectively representing a measured voltage curve and a voltage response output curve at time t, P i And P j The data values of the measured power curve and the power response output curve at time T are respectively shown, and T represents the total data values at time T.
5. The method for identifying parameters of a comprehensive load model according to claim 4, wherein inputting measurement information obtained in real time into the parameter correction model, performing iterative correction on the initial parameters, and obtaining a parameter identification result comprises:
inputting the initial parameters obtained by matching and the measurement information obtained in real time into a parameter correction model, and carrying out iterative correction on the initial parameters to obtain a group of parameters with the minimum adaptation degree of model output and the measurement information obtained in real time as a model parameter identification result:
wherein F is d An adaptability index for indicating the approach degree of the model output and the measurement information acquired in real time, n represents the number of sampling points and P N And V N Rated value, P, of active power and voltage respectively j And V j The simulation values of the active power and the current at the sampling point j are respectively, P j0 And V j0 The active power and the actual voltage at the sampling point j are shown, respectively.
6. An identification terminal for comprehensive load model parameters, comprising a memory, a processor and a computer program stored on the memory and capable of running on the processor, characterized in that the processor implements the following steps when executing the computer program:
establishing a comprehensive load model according to the structure of the load to be identified, and obtaining a response curve of the comprehensive load model and a model parameter model library;
matching the measurement information acquired in real time with the mode library, and determining initial parameters of the comprehensive load model according to a matching result;
and constructing a parameter correction model of the comprehensive load model based on deep learning, inputting measurement information acquired in real time into the parameter correction model, and carrying out iterative correction on the initial parameters to obtain a parameter identification result.
7. The terminal for identifying parameters of a comprehensive load model according to claim 6, wherein said establishing the comprehensive load model according to the structure of the load to be identified comprises:
establishing a comprehensive load model which is the same as a load structure to be identified, wherein the comprehensive load model comprises a motor load, a static load and a distributed photovoltaic, and setting initial parameters in the value range of each parameter of the comprehensive load model: θ i ∈[θ imin ，θ imax ]；
In θ i Represents the initial value of the ith parameter, θ imin And theta imax The lower limit and the upper limit of the value range of the ith parameter are respectively indicated.
8. The terminal for identifying parameters of a comprehensive load model according to claim 6, wherein obtaining a response curve of the comprehensive load model and a model parameter pattern library comprises:
inputting different working condition data of a preset group into the comprehensive load model to obtain a power response output curve, a voltage response output curve and a parameter group of the comprehensive load model under different working conditions of the preset group;
classifying parameter sets of a preset group of comprehensive load models, and taking a response track of each sample in each classification as a mode response value of each classification;
and taking the average model parameters of all the classifications as model parameters of all the classifications, and obtaining a model parameter model library of the comprehensive load model according to the model parameters of all the classifications and the model response values of all the classifications.
9. The terminal for identifying parameters of a comprehensive load model according to claim 8, wherein the matching the measurement information obtained in real time with the pattern library, and determining initial parameters of the comprehensive load model according to the matching result comprises:
inputting measurement information obtained in real time into the comprehensive load model to obtain a measurement response curve, wherein the measurement response curve comprises a measurement voltage curve and a measurement power curve;
matching the measurement response curve with a power response output curve and a voltage response output curve, and selecting the power response output curve and the voltage response output curve with the minimum vector distance with the measurement response curve as matching results:
wherein D represents a vector distance, V i And V j Data values respectively representing a measured voltage curve and a voltage response output curve at time t, P i And P j The data values of the measured power curve and the power response output curve at time T are respectively shown, and T represents the total data values at time T.
10. The terminal for identifying parameters of a comprehensive load model according to claim 9, wherein inputting measurement information obtained in real time into the parameter correction model, performing iterative correction on the initial parameters, and obtaining a parameter identification result comprises:
inputting the initial parameters obtained by matching and the measurement information obtained in real time into a parameter correction model, and carrying out iterative correction on the initial parameters to obtain a group of parameters with the minimum adaptation degree of model output and the measurement information obtained in real time as a model parameter identification result:
wherein F is d An adaptability index for indicating the approach degree of the model output and the measurement information acquired in real time, n represents the number of sampling points and P N And V N Rated value, P, of active power and voltage respectively j And V j The simulation values of the active power and the current at the sampling point j are respectively, P j0 And V j0 The active power and the actual voltage at the sampling point j are shown, respectively.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311211200.7A CN117421972A (en) | 2023-09-19 | 2023-09-19 | Identification method and terminal for comprehensive load model parameters |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311211200.7A CN117421972A (en) | 2023-09-19 | 2023-09-19 | Identification method and terminal for comprehensive load model parameters |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN117421972A true CN117421972A (en) | 2024-01-19 |
Family
ID=89531535
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311211200.7A Pending CN117421972A (en) | 2023-09-19 | 2023-09-19 | Identification method and terminal for comprehensive load model parameters |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117421972A (en) |
-
2023
- 2023-09-19 CN CN202311211200.7A patent/CN117421972A/en active Pending
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109145516B (en) | Analog circuit fault identification method based on improved extreme learning machine | |
| CN109840595B (en) | Knowledge tracking method based on group learning behavior characteristics | |
| CN113240105B (en) | Power grid steady state discrimination method based on graph neural network pooling | |
| CN111523728B (en) | Four-stage hybrid short-term wind direction prediction method | |
| CN113449919B (en) | Power consumption prediction method and system based on feature and trend perception | |
| CN108647772B (en) | Method for removing gross errors of slope monitoring data | |
| CN111967271A (en) | Analysis result generation method, device, equipment and readable storage medium | |
| CN111192158A (en) | Transformer substation daily load curve similarity matching method based on deep learning | |
| CN111476495A (en) | Evaluation and optimization method and system for improving learning efficiency | |
| CN109540089B (en) | Bridge deck elevation fitting method based on Bayes-Kriging model | |
| CN113919221A (en) | Fan load prediction and analysis method and device based on BP neural network and storage medium | |
| CN117421972A (en) | Identification method and terminal for comprehensive load model parameters | |
| CN115879412A (en) | Layout level circuit diagram size parameter optimization method based on transfer learning | |
| CN115859521A (en) | Neural network-based milling error reconstruction method and system | |
| CN113783186B (en) | Voltage prediction method considering topological structure change of power distribution network | |
| CN114372418A (en) | Wind power space-time situation description model establishing method | |
| CN114723998A (en) | Small sample image classification method and device based on large-boundary Bayes prototype learning | |
| CN114254199A (en) | Course recommendation method based on bipartite graph projection and node2vec | |
| CN113642784A (en) | Wind power ultra-short term prediction method considering fan state | |
| CN113642029A (en) | Method and system for measuring correlation between data sample and model decision boundary | |
| CN113449631A (en) | Image classification method and system | |
| Lim et al. | Analyzing deep neural networks with noisy labels | |
| He et al. | Application of ALO-ELM in Load Forecasting Based on Big Data | |
| Angel et al. | Metaheuristic Tuning and Practical Implementation of a PID Controller Employing Genetic Algorithms | |
| Yuan | English Translation Ability Assessment Based on Neural Network Algorithm of Particle Swarm Optimization |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |