CN114626957A - Voltage sag state evaluation method based on gate control cycle unit deep learning model - Google Patents
Voltage sag state evaluation method based on gate control cycle unit deep learning model Download PDFInfo
- Publication number
- CN114626957A CN114626957A CN202210106391.XA CN202210106391A CN114626957A CN 114626957 A CN114626957 A CN 114626957A CN 202210106391 A CN202210106391 A CN 202210106391A CN 114626957 A CN114626957 A CN 114626957A
- Authority
- CN
- China
- Prior art keywords
- data
- voltage
- voltage data
- deep learning
- node
- 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
- 238000013136 deep learning model Methods 0.000 title claims abstract description 42
- 238000011156 evaluation Methods 0.000 title claims abstract description 25
- 238000012544 monitoring process Methods 0.000 claims abstract description 40
- 238000012549 training Methods 0.000 claims abstract description 35
- 238000010606 normalization Methods 0.000 claims abstract description 16
- 238000012795 verification Methods 0.000 claims abstract description 13
- 238000004458 analytical method Methods 0.000 claims abstract description 6
- 238000000034 method Methods 0.000 claims description 19
- 230000008569 process Effects 0.000 claims description 15
- 230000006870 function Effects 0.000 claims description 11
- 210000002569 neuron Anatomy 0.000 claims description 9
- 239000011159 matrix material Substances 0.000 claims description 8
- 238000012360 testing method Methods 0.000 claims description 8
- 125000004122 cyclic group Chemical group 0.000 claims description 7
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 claims description 4
- 238000003062 neural network model Methods 0.000 claims description 4
- 239000002689 soil Substances 0.000 claims description 4
- 230000004913 activation Effects 0.000 claims description 3
- 238000005070 sampling Methods 0.000 claims description 3
- 238000007781 pre-processing Methods 0.000 claims description 2
- 238000012545 processing Methods 0.000 claims description 2
- 238000013528 artificial neural network Methods 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 230000001537 neural effect Effects 0.000 description 1
- 230000005855 radiation Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000004044 response Effects 0.000 description 1
- 239000004065 semiconductor Substances 0.000 description 1
- 238000004088 simulation Methods 0.000 description 1
- 238000007619 statistical method Methods 0.000 description 1
- 239000000126 substance Substances 0.000 description 1
- 230000001052 transient effect Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/06—Energy or water supply
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/044—Recurrent networks, e.g. Hopfield networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/047—Probabilistic or stochastic networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Health & Medical Sciences (AREA)
- General Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Computational Linguistics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Biomedical Technology (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Biophysics (AREA)
- Business, Economics & Management (AREA)
- Economics (AREA)
- Probability & Statistics with Applications (AREA)
- Public Health (AREA)
- Water Supply & Treatment (AREA)
- Human Resources & Organizations (AREA)
- Marketing (AREA)
- Primary Health Care (AREA)
- Strategic Management (AREA)
- Tourism & Hospitality (AREA)
- General Business, Economics & Management (AREA)
- Supply And Distribution Of Alternating Current (AREA)
Abstract
The invention discloses a voltage sag state evaluation method based on a gate control cycle unit deep learning model, which comprises the steps of collecting monitoring node voltage data A and node voltage data B to be evaluated under each operation condition; calculating the root mean square value of the three-phase voltages of A and B to form voltage sag sample data C; carrying out normalization pretreatment on the A and the B to form voltage data E and voltage data F; taking the minimum value in the F to divide the section, and numbering to form L; forming voltage sag sample data M by the E and the L; dividing M into sample sets; training to obtain a trained model; performing fitting verification to select an optimal model; and inputting actual voltage data acquired by the monitoring node into the optimal model, wherein the output data is the voltage sag state analysis result of the node to be evaluated. The invention reduces the economic cost to a certain extent and improves the reliability.
Description
Technical Field
The invention belongs to the technical field of power system analysis, and particularly relates to a voltage sag state evaluation method based on a gate control cycle unit deep learning model.
Background
The quality of electric energy is directly related to the safe and efficient operation of the power system. The problem of voltage sag in power quality has attracted much attention, and has become an urgent problem to be solved in academic and industrial fields. The voltage sag not only brings huge economic loss to the interested parties, but also may cause great social influence, especially greater loss to high-end manufacturing industries which take technologies such as semiconductors, digitalization and informatization as the core.
The method has the advantages that the accurate voltage sag state assessment, namely the voltage sag degree of each node in the power grid is accurately assessed, the important significance is realized on clearing the responsibility of events and improving the power quality of the power grid, and the important research significance and the practical significance are realized. However, the access of the large-scale distributed power supply fundamentally changes the structure of the single-power-supply radiation type power distribution network, the complexity and uncertainty of the power distribution network structure are increased, the transient response characteristic of the power system is greatly changed, and the electrical characteristic quantity after the voltage sag occurs is deeply changed. The existing voltage sag state evaluation methods have the following defects: the real-time monitoring statistical method is based on real-time statistical data, can intuitively and accurately evaluate the voltage sag frequency, but has the problems of high economic cost and long monitoring period; the random estimation method utilizes a random model for evaluation, but is easily influenced by subjective factors and has low result reliability. In fact, most of voltage sag is caused by faults, and the voltage sag state evaluation of each node in the whole network can be realized by effectively utilizing monitoring data of limited power grid nodes by means of electric energy quality monitoring equipment which is widely installed in a power grid.
Disclosure of Invention
The invention aims to provide a voltage sag state evaluation method based on a gate control cycle unit deep learning model, and solves the problems of high economic cost, easily influenced result by subjective factors and low reliability in the prior art.
The technical scheme adopted by the invention is that,
the voltage sag state evaluation method based on the gate control cycle unit deep learning model specifically comprises the following steps:
step 1: combining an electric power system network to be evaluated, and under different working conditions, respectively changing short circuit capacity, grounding resistance, fault type, fault duration, fault starting and stopping time, load capacity, line impedance and fault line in the electric power system network; the fault types comprise single-phase faults, two-phase faults and three-phase faults;
collecting monitoring node voltage data A and node voltage data B to be evaluated under each operating condition by using voltage measuring equipment installed in a power grid; each group of monitoring node voltage data A and node voltage data B to be evaluated correspond to the same working condition;
step 2: calculating the root mean square value of the three-phase voltage of the monitoring node voltage data A and the node voltage data B to be evaluated, and if the root mean square value of the voltage of the monitoring node voltage data A or the node voltage data B to be evaluated is reduced to 10% -90% of a rated value, forming voltage sag sample data C by the monitoring node voltage data A and the corresponding node voltage data B to be evaluated;
and step 3: carrying out normalization preprocessing on the voltage data A of the monitoring node and the voltage data B of the node to be evaluated in the step 2 to form voltage data E corresponding to the monitoring node A and voltage data F corresponding to the voltage data B of the node to be evaluated; the voltage data E and the voltage data F correspond to the same working condition;
and 4, step 4: taking the minimum value in the voltage data F, dividing the sections, and numbering to form L;
and 5: forming a plurality of groups of voltage sag sample data M by the voltage data E and the serial number L;
step 6: and (3) randomly dividing voltage sag sample data M according to the following steps of 7: 2: 1, dividing the test sample into a training sample set, a verification sample set and a test sample set;
and 7: constructing an integral structure of a gating cycle unit deep learning model;
and 8: training the gating cycle unit deep learning model in the step 7 by using the training sample set in the step 6 to obtain a trained model;
and step 9: performing overfitting verification on the gating cycle unit deep learning model in the step 7 by using the verification sample set in the step 6; if the accuracy rate is reduced by less than or equal to 3%, judging that the gate control cycle unit deep learning model does not generate an overfitting phenomenon at the moment; if the accuracy rate is reduced by more than 3%, judging that the gate control cycle unit deep learning model has an overfitting phenomenon at the moment; at this time, the gating cycle unit deep learning model needs to be trained again according to the step 8, and the number of hidden layers, the number of neurons, the learning rate, the number of generations, the number of iterations, the discarding rate and other super parameters are modified until the gating cycle unit deep learning model does not generate an overfitting phenomenon, and then the gating cycle unit deep learning model is used as an optimal model;
step 10: and (3) inputting actual voltage data acquired by the monitoring node into the optimal model obtained in the step (9), wherein output data are voltage sag state analysis results of the nodes to be evaluated.
The invention is also characterized in that;
the step 2 specifically comprises the following steps: and (3) carrying out root mean square value of the following formula (1) on the monitoring node voltage data A and the node voltage data B to be evaluated:
wherein N is the number of data points, x, collected in each period1，x2，x3，……，xNAnd sequentially obtaining the data values corresponding to the sampling data points in each period.
The step 3 specifically comprises the following steps: carrying out normalization processing of the following formula (2) on the voltage data A of the monitoring node and the voltage data B of the node to be evaluated to obtain preprocessed data;
wherein x is*Is normalized data output, x is original data, xmaxIs the maximum value, x, in the input sample dataminIs the minimum value in the input sample data.
In step 4, if the minimum value in the voltage data F is more than or equal to 0.9, the number is 1; if the minimum value in the voltage data F is more than or equal to 0.8 and less than 0.9, the number is 2; if the minimum value in the voltage data F is more than or equal to 0.7 and less than 0.8, the number is 3; if the minimum value in the voltage data F is more than or equal to 0.6 and less than 0.7, the number is 4; if the minimum value in the voltage data F is more than or equal to 0.5 and less than 0.6, the number is 5; if the minimum value in the voltage data F is <0.5, number 6.
The step 7 specifically comprises the following steps: the whole model structure of the gate control cycle unit deep learning model can be divided into three parts, namely an input layer, a hidden layer and an output layer; the first part is an input layer, the second part is a hidden layer, and the third part is an output layer; the input layer only comprises one input layer, provides a data input interface for the model and processes the voltage data E into matrix data which can be efficiently operated and processed in batch; the hidden layer part comprises a plurality of hidden layers, and each hidden layer comprises a gated cycle unit layer, a batch normalization layer, a discarding layer and a full connection layer; the output layer part is a Softmax layer.
The step 8 specifically comprises the following steps: updating parameters by using a back propagation algorithm, and training by using an Adam optimizer, wherein the loss function is a cross soil moisture loss function; the input data is voltage data E matrix format data, and the output data is the voltage sag state of the node to be evaluated; the number of hidden layers, the number of neurons, the type of an activation function, the learning rate, the generation number, the iteration number, the discarding rate and other model hyperparameters can be adjusted; each training cycle traverses each training data in the training set, each traversal is called a generation, and the neural network model is trained for multiple generations.
The voltage sag state evaluation method based on the gate control cycle unit deep learning model has the beneficial effects that: by adopting the method, the problem of voltage sag state evaluation is solved by establishing a deep nonlinear relation between the voltage data of the monitoring node and the voltage sag state of the node to be evaluated based on the independent cyclic neural model, and the problems of high economic cost, easily influenced result by subjective factors and low reliability in the prior art are solved.
Drawings
FIG. 1 is a schematic diagram of a model structure of a voltage sag state evaluation method based on a gated cycle unit deep learning model;
FIG. 2 is a training iterative update diagram of accuracy in a voltage sag state evaluation method based on a gated cycle unit deep learning model.
Detailed Description
The voltage sag state evaluation method based on the gated loop unit deep learning model according to the present invention is further described in detail with reference to the accompanying drawings and the detailed description.
As shown in fig. 1, the voltage sag state evaluation method based on the gated loop unit deep learning model of the present invention is specifically implemented according to the following steps:
step 1: combining an electric power system network to be evaluated, respectively changing short-circuit capacity, grounding resistance, fault types (including single-phase fault, two-phase fault and three-phase fault), fault duration, fault starting and stopping time, load capacity, line impedance and fault lines in the electric power system network under different working conditions, and collecting monitoring node voltage data A and node voltage data B to be evaluated under each operating condition by using voltage measuring equipment already installed in an electric network; the voltage data a and the voltage data B must be the same corresponding operating condition.
Step 2: calculating the root mean square value of the three-phase voltage of the monitoring node voltage data A and the node voltage data B to be evaluated, and reducing the root mean square value of the voltage of A or B to 10% -90% of a rated value, so that voltage sag sample data C are formed by the monitoring node voltage data A and the corresponding node voltage data B to be evaluated; the specific process is as follows:
the root mean square value shown in equation (1) is performed for the voltage data a and the voltage data B:
wherein N is the number of data points, x, collected in each period1，x2，x3，……，xNIn turn for each sampled data point in each cycleThe corresponding data value.
And step 3: carrying out normalization pretreatment on A, B in the step 2 to form voltage data E corresponding to the monitoring node A and voltage data F corresponding to the voltage data B of the node to be evaluated; the voltage data E and the voltage data F are required to be in the same corresponding working condition; the specific process is as follows:
normalizing the voltage data A and the voltage data B as shown in the formula (2) to obtain preprocessed data;
wherein x is*Is normalized data output, x is original data, xmaxIs the maximum value, x, in the input sample dataminIs the minimum value in the input sample data;
and 4, step 4: taking the minimum value in the voltage data F, dividing the sections, and numbering to form L; the specific process is as follows:
if the minimum value in the voltage data F is more than or equal to 0.9, the number is 1; if the minimum value in the voltage data F is more than or equal to 0.8 and less than 0.9, the number is 2; if the minimum value in the voltage data F is more than or equal to 0.7 and less than 0.8, the number is 3; if the minimum value in the voltage data F is more than or equal to 0.6 and less than 0.7, the number is 4; if the minimum value in the voltage data F is more than or equal to 0.5 and less than 0.6, the number is 5; if the minimum value in the voltage data F is <0.5, number 6.
And 5: forming a plurality of groups of voltage sag sample data M by the voltage data E and the serial number L;
step 6: and (3) randomly dividing the voltage sag sample data M according to the following steps of 7: 2: 1, dividing the test sample into a training sample set, a verification sample set and a test sample set;
and 7: constructing an integral structure of a gating cycle unit deep learning model; as shown in fig. 1, the specific process is as follows:
the whole model structure of the gate control cycle unit deep learning model can be divided into three parts, namely an input layer, a hidden layer and an output layer. The first part is an input layer, the second part is a hidden layer, and the third part is an output layer. The input layer only comprises one input layer, provides a data input interface for the model, and processes the voltage data E into matrix data which can be efficiently operated and processed in batch. The hidden layer part comprises a plurality of hidden layers, wherein the hidden layers comprise a gated cycle unit layer, a batch normalization layer, a discarding layer and a full connection layer. The output layer part is a Softmax layer. And the rest of the neural network layers except the first input layer are linked with the previous neural network layer through an activation function. For example: using 1 input layer and 10 hidden layers, wherein the hidden layers are a gated circulation unit layer, a batch normalization layer, a gated circulation unit layer, a batch normalization layer, a discard layer and a full connection layer in sequence; and the output layer part is a Softmax layer and is used for obtaining the voltage sag state of the node to be evaluated.
And 8: training the model in the step 7 by using the training sample set in the step 6 to obtain a trained model; the specific process is as follows:
and updating parameters by using a back propagation algorithm, training by using an Adam optimizer, wherein the loss function is a cross soil moisture loss function, and adjusting the model and observing the output accuracy of the model by using different hyper-parameters in the training. The input data is voltage data E matrix format data, and the output data is the voltage sag state of the node to be evaluated. Each training cycle traverses each training data in the training set, each traversal is called a generation, and the neural network model is trained for multiple generations.
The number of hidden layers, the number of neurons, the learning rate, the number of generations, the number of iterations and the discarding rate can be adjusted;
and step 9: performing overfitting verification on the model in the step 7 by using the verification sample set in the step 6; if the accuracy rate is reduced by less than or equal to 3 percent, the model is inferred not to have an overfitting phenomenon; if the accuracy rate is reduced by more than 3%, an overfitting phenomenon occurs in the model, at the moment, model training needs to be carried out again according to the step 8, model hyperparameters such as the number of hidden layers, the number of neurons, the learning rate, the number of generations, the number of iterations, the discarding rate and the like are modified, and the model is used as an optimal model until the overfitting phenomenon does not occur in the model;
step 10: and (3) inputting actual voltage data acquired by the monitoring node into the optimal model obtained in the step (9), wherein the output data is the voltage sag state analysis result of the node to be evaluated.
The voltage sag state evaluation method based on the gated loop unit deep learning model of the present invention is further described in detail by the following specific embodiments.
The invention relates to a voltage sag state evaluation method based on a gated cyclic unit deep learning model, which specifically comprises the following steps of:
step 1: the embodiment of the invention is based on an IEEE39 node network model, 3 monitoring nodes (bus3, bus24 and bus38) are selected, and the voltage sag state of the to-be-evaluated node bus5 is evaluated. The method comprises the steps that Matlab/simulink simulation software is used for modeling to generate simulation data, under different working conditions, short-circuit capacity, grounding resistance, fault types (including single-phase faults, two-phase faults and three-phase faults), fault duration, fault starting and stopping time, load capacity, line impedance and fault lines in a power system network are changed respectively, and voltage measurement equipment installed in a power grid is utilized to collect voltage data A of 3 monitoring nodes (bus3, bus24 and bus38) and voltage data B of nodes to be evaluated (bus 5) under each operating condition; the monitoring node voltage data A and the node voltage data B to be evaluated are required to be in the same corresponding working condition.
Step 2: calculating the three-phase voltage root mean square values of voltage data A of 3 monitoring nodes (bus3, bus24 and bus38) and voltage data B of a node to be evaluated (bus 5), and reducing the root mean square value of the voltage of A or B to 10-90% of a rated value, so that voltage data A of the monitoring nodes and the corresponding voltage data B of the node to be evaluated form voltage sag sample data C; the specific process is as follows:
the root mean square value shown in equation (1) is performed for the voltage data a and the voltage data B:
wherein the content of the first and second substances,n is the number of data points, x, collected per cycle1，x2，x3，……，xNAnd sequentially obtaining the data value corresponding to each sampling data point in each period.
And 3, step 3: carrying out normalization pretreatment on A, B in the step 2 to form voltage data E corresponding to the monitoring node A and voltage data F corresponding to the voltage data B of the node to be evaluated; the voltage data E and the voltage data F are required to be in the same corresponding working condition; the specific process is as follows:
normalizing the voltage data A and the voltage data B as shown in the formula (2) to obtain preprocessed data;
wherein x is*Is normalized data output, x is original data, xmaxIs the maximum value, x, in the input sample dataminIs the minimum value in the input sample data;
and 4, step 4: taking the minimum value in the voltage data F, dividing the sections, and numbering to form L; the specific process is as follows:
if the minimum value in the data F is more than or equal to 0.9, the number is 1; if the minimum value in the data F is more than or equal to 0.8 and less than 0.9, the number is 2; if the minimum value in the data F is more than or equal to 0.7 and less than 0.8, the number is 3; if the minimum value in the data F is more than or equal to 0.6 and less than 0.7, the number is 4; if the minimum value in the data F is more than or equal to 0.5 and less than 0.6, the number is 5; if the minimum value in the data F is <0.5, the number is 6.
And 5: forming a plurality of groups of voltage sag sample data M by the monitoring voltage data E and the serial number L;
step 6: and (3) randomly dividing voltage sag sample data M according to the following steps of 7: 2: 1, dividing the test sample into a training sample set, a verification sample set and a test sample set;
and 7: building an integral structure of a gating cycle unit deep learning model; as shown in fig. 2, the specific process is as follows:
the whole model structure of the gate control cycle unit deep learning model can be divided into three parts, namely an input layer, a hidden layer and an output layer. The first part is an input layer, the second part is a hidden layer, and the third part is an output layer. The input layer only comprises one input layer, provides a data input interface for the model and processes the voltage data E into matrix data which can be efficiently operated and processed in batch. The hidden layer part comprises a plurality of hidden layers, wherein the hidden layers comprise a gated cycle unit layer, a batch normalization layer, a discarding layer and a full connection layer. The output layer part is a Softmax layer. In this embodiment, 1 input layer (first layer) is used, 8 hidden layers are used, and the hidden layers are a gated cycle unit layer (second layer), a batch normalization layer (third layer), a gated cycle unit layer (fourth layer), a batch normalization layer (fifth layer), a gated cycle unit layer (sixth layer), a batch normalization layer (seventh layer), a discard layer (eighth layer), and a full connection layer (ninth layer) in this order; and the output layer part is a Softmax layer (tenth layer) and obtains the voltage sag state of the node to be evaluated.
As shown in fig. 2, fig. 2 shows the accuracy change trend in the model training process, and as the number of iterations increases, the model accuracy increases. Through 200 generations of iterations, the accuracy rate is over 99.5%. And 8: training the model in the step 7 by using the training sample set in the step 6 to obtain a trained model; the specific process is as follows:
and updating parameters by using a back propagation algorithm, training by using an Adam optimizer, wherein the loss function is a cross soil moisture loss function, and adjusting the model and observing the output accuracy of the model by using different hyper-parameters in the training. The input data is voltage data E matrix format data, and the output data is the voltage sag state of the node to be evaluated. Each training cycle traverses each training data in the training set, each traversal is called a generation, and the neural network model is trained for multiple generations.
The number of hidden layers, the number of neurons, the learning rate, the number of generations, the number of iterations, and the discarding rate can be adjusted. In this embodiment, the number of hidden layers is 8, the number of neurons is 256, the learning rate is 0.0001, the generation number is 200, the iteration number is 200, and the discarding rate is 0.3.
And step 9: performing overfitting verification on the model in the step 7 by using the verification sample set in the step 6; if the accuracy rate is reduced by less than or equal to 3 percent, the model has no overfitting phenomenon; if the accuracy rate is reduced by more than 3%, an overfitting phenomenon occurs in the model, at the moment, model training needs to be carried out again according to the step 8, model hyperparameters such as the number of hidden layers, the number of neurons, the learning rate, the number of generations, the number of iterations, the discarding rate and the like are modified, and the model is used as an optimal model until the overfitting phenomenon does not occur in the model;
step 10: actual voltage data acquired by 3 monitoring nodes (bus3, bus24 and bus38) are input into the optimal model obtained in the step 9, and output data are voltage sag state analysis results of the node to be evaluated bus 5.
The voltage sag state evaluation method based on the gate control cycle unit deep learning model solves the problem of voltage sag state evaluation by reasonably designing and establishing the model, reduces the economic cost to a certain extent, reduces the influence degree of the result which is easily influenced by subjective factors, improves the reliability of the result, and has certain practical significance.
Claims (6)
1. The voltage sag state evaluation method based on the gated loop unit deep learning model is characterized by comprising the following steps:
step 1: combining an electric power system network to be evaluated, and under different working conditions, respectively changing short-circuit capacity, grounding resistance, fault type, fault duration, fault starting and stopping time, load capacity, line impedance and fault line in the electric power system network; the fault types comprise single-phase faults, two-phase faults and three-phase faults;
collecting monitoring node voltage data A and node voltage data B to be evaluated under each operating condition by using voltage measuring equipment installed in a power grid; each group of monitoring node voltage data A and node voltage data B to be evaluated correspond to the same working condition;
step 2: calculating three-phase voltage root-mean-square values of the monitoring node voltage data A and the node voltage data B to be evaluated, and if the voltage root-mean-square value of the monitoring node voltage data A or the node voltage data B to be evaluated is reduced to 10% -90% of a rated value, forming voltage sag sample data C by the monitoring node voltage data A and the corresponding node voltage data B to be evaluated;
and step 3: carrying out normalization preprocessing on the voltage data A of the monitoring node and the voltage data B of the node to be evaluated in the step 2 to form voltage data E corresponding to the monitoring node A and voltage data F corresponding to the voltage data B of the node to be evaluated; the voltage data E and the voltage data F correspond to the same working condition;
and 4, step 4: taking the minimum value in the voltage data F, dividing the sections, and numbering to form L;
and 5: forming voltage sag sample data M by the voltage data E and the serial number L;
step 6: and (3) randomly dividing voltage sag sample data M according to the following steps of 7: 2: 1, dividing the test sample into a training sample set, a verification sample set and a test sample set;
and 7: constructing an integral structure of a gating cycle unit deep learning model;
and 8: training the gating cycle unit deep learning model in the step 7 by using the training sample set in the step 6 to obtain a trained model;
and step 9: performing overfitting verification on the gating cycle unit deep learning model in the step 7 by using the verification sample set in the step 6; if the accuracy rate is reduced by less than or equal to 3%, judging that the gate control cycle unit deep learning model does not generate an overfitting phenomenon at the moment; if the accuracy rate is reduced by more than 3%, judging that the gate control cycle unit deep learning model has an overfitting phenomenon at the moment; at this time, the gating cycle unit deep learning model needs to be trained again according to the step 8, and the number of hidden layers, the number of neurons, the learning rate, the number of generations, the number of iterations, the discarding rate and other super parameters are modified until the gating cycle unit deep learning model does not generate an overfitting phenomenon, and then the gating cycle unit deep learning model is used as an optimal model;
step 10: and (3) inputting actual voltage data acquired by the monitoring node into the optimal model obtained in the step (9), wherein the output data is the voltage sag state analysis result of the node to be evaluated.
2. The voltage sag state evaluation method based on the gated cyclic unit deep learning model according to claim 1, wherein the step 2 specifically comprises: and (3) carrying out root mean square value of the following formula (1) on the monitoring node voltage data A and the node voltage data B to be evaluated:
wherein N is the number of data points, x, collected in each period1，x2，x3，……，xNAnd sequentially obtaining the data values corresponding to the sampling data points in each period.
3. The voltage sag state evaluation method based on the gated cyclic unit deep learning model according to claim 1, wherein the step 3 specifically comprises: carrying out normalization processing of the following formula (2) on the voltage data A of the monitoring node and the voltage data B of the node to be evaluated to obtain preprocessed data;
wherein x is*Is normalized data output, x is original data, xmaxIs the maximum value, x, in the input sample dataminIs the minimum value in the input sample data.
4. The voltage sag state evaluation method based on the gated cyclic unit deep learning model according to claim 1, wherein in the step 4, if the minimum value in the voltage data F is greater than or equal to 0.9, the number is 1; if the minimum value in the voltage data F is more than or equal to 0.8 and less than 0.9, the number is 2; if the minimum value in the voltage data F is more than or equal to 0.7 and less than 0.8, the number is 3; if the minimum value in the voltage data F is more than or equal to 0.6 and less than 0.7, the number is 4; if the minimum value in the voltage data F is more than or equal to 0.5 and less than 0.6, the number is 5; if the minimum value in the voltage data F is <0.5, the number is 6.
5. The voltage sag state evaluation method based on the gated cyclic unit deep learning model according to claim 1, wherein the step 7 specifically comprises: the integral model structure of the gate control cycle unit deep learning model can be divided into three parts, namely an input layer, a hidden layer and an output layer; the first part is an input layer, the second part is a hidden layer, and the third part is an output layer; the input layer only comprises one input layer, provides a data input interface for the model and processes the voltage data E into matrix data which can be efficiently operated and processed in batch; the hidden layer part comprises a plurality of hidden layers, and each hidden layer comprises a gated cycle unit layer, a batch normalization layer, a discarding layer and a full connection layer; the output layer part is a Softmax layer.
6. The voltage sag state evaluation method based on the gated cyclic unit deep learning model according to claim 1, wherein the step 8 specifically comprises: updating parameters by using a back propagation algorithm, and training by using an Adam optimizer, wherein the loss function is a cross soil moisture loss function; the input data is voltage data E matrix format data, and the output data is a voltage sag state of a node to be evaluated; the number of hidden layers, the number of neurons, the type of an activation function, the learning rate, the generation number, the iteration number, the discarding rate and other model hyperparameters can be adjusted; each training cycle traverses each training data in the training set, each traversal is called a generation, and the neural network model is trained for multiple generations.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210106391.XA CN114626957A (en) | 2022-01-28 | 2022-01-28 | Voltage sag state evaluation method based on gate control cycle unit deep learning model |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210106391.XA CN114626957A (en) | 2022-01-28 | 2022-01-28 | Voltage sag state evaluation method based on gate control cycle unit deep learning model |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN114626957A true CN114626957A (en) | 2022-06-14 |
Family
ID=81898135
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210106391.XA Pending CN114626957A (en) | 2022-01-28 | 2022-01-28 | Voltage sag state evaluation method based on gate control cycle unit deep learning model |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114626957A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115775114A (en) * | 2022-12-29 | 2023-03-10 | 国网甘肃省电力公司经济技术研究院 | New energy station transient voltage stability evaluation method based on gated cycle unit |
-
2022
- 2022-01-28 CN CN202210106391.XA patent/CN114626957A/en active Pending
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115775114A (en) * | 2022-12-29 | 2023-03-10 | 国网甘肃省电力公司经济技术研究院 | New energy station transient voltage stability evaluation method based on gated cycle unit |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107104442B (en) | Method for calculating probability load flow of power system including wind power plant by considering parameter ambiguity | |
| CN113205207A (en) | XGboost algorithm-based short-term power consumption load fluctuation prediction method and system | |
| CN111475909B (en) | Wind turbine generator output correlation mapping modeling method based on long-term and short-term memory network | |
| CN110570122A (en) | Offshore wind power plant reliability assessment method considering wind speed seasonal characteristics and current collection system element faults | |
| CN111091141B (en) | Photovoltaic backboard fault diagnosis method based on layered Softmax | |
| CN106570790A (en) | Wind farm output power data restoration method considering segmental characteristics of wind speed data | |
| CN114626957A (en) | Voltage sag state evaluation method based on gate control cycle unit deep learning model | |
| CN113379116A (en) | Cluster and convolutional neural network-based line loss prediction method for transformer area | |
| CN113804997B (en) | Voltage sag source positioning method based on bidirectional WaveNet deep learning | |
| CN116796403A (en) | Building energy saving method based on comprehensive energy consumption prediction of commercial building | |
| CN105024645A (en) | Matrix evolution-based photovoltaic array fault location method | |
| CN117077546B (en) | Power system load modeling method and system based on data driving | |
| Xu et al. | Short-term wind speed prediction based on GRU | |
| CN115983507B (en) | Method and system for predicting broadband oscillation risk of section of power grid of transmitting end source | |
| CN116502149A (en) | Low-voltage power distribution network user-transformation relation identification method and system based on current characteristic conduction | |
| CN106992516A (en) | The method that probability air extract is obtained with Density Estimator is simulated based on quasi-Monte Carlo | |
| CN116317937A (en) | Distributed photovoltaic power station operation fault diagnosis method | |
| CN113949079B (en) | Power distribution station user three-phase unbalance prediction optimization method based on deep learning | |
| CN114545147A (en) | Voltage sag source positioning method based on deep learning in consideration of time-varying topology | |
| CN111027816B (en) | Photovoltaic power generation efficiency calculation method based on data envelope analysis | |
| Duan et al. | High performance computing (hpc) for advanced power system studies | |
| CN109494747B (en) | Power grid probability load flow calculation method based on alternating gradient algorithm | |
| CN112186750B (en) | Voltage sag state estimation method considering power grid reconstruction | |
| Guo et al. | An Optimized Back Propagation Neural Network Based on Genetic Algorithm for Line Loss Calculation in Low-voltage Distribution Grid | |
| Zhong et al. | Low-voltage distribution network topology identification method based on segmented current features and cnn-lstm deep learning |
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 |