A detection method for islanding faults of distributed power sources

Abstract

The present invention discloses an intelligent distributed power source islanding detection method based on variational mode decomposition, Teager energy operator and one-dimensional convolutional neural network. The method uses variational mode decomposition to characterize the features of three-phase PCC point voltage and converter output current. At the same time, in order to accurately track the instantaneous changes of the intrinsic mode function component IMF1 of the feature signal, the Teager energy operator is introduced to highlight the islanding fault features. The feature vector space composed of the Teager energy operator features, the intrinsic mode function IMF2 component features and the intrinsic mode function IMF3 component features is used as the input of the one-dimensional convolutional neural network. It can quickly and accurately classify and identify islanding events and non-islanding events without injecting perturbation signals into the system, without setting thresholds, and without parameter tuning.

Description

Technical Field

[0001] The present invention relates to the technical field of distributed power islanding fault detection, and particularly relates to a detection method for distributed power islanding faults based on adaptive VMD-Teager energy operator. Background Art

[0002] With the integration of distributed power systems and traditional power grids, energy efficiency and reliability have been improved, new transmission lines have been reduced, and problems such as line losses have been avoided. The power system has gradually developed towards low carbonization. However, the steady-state and dynamic characteristics of distributed power have brought technical challenges to control and protection, posing a potential threat to the reliability of the power system. Islanding fault is one of the main problems in this context. Islanding fault refers to the phenomenon that when the power grid stops working due to maintenance or accidental faults, the distributed generation system connected to the power grid cannot detect the fault in time and cannot disconnect from the power grid, resulting in the distributed power supplying power to the load alone. This kind of fault will cause problems such as the complication of power restoration and the decline of power quality, and even endanger the lives of workers. According to the provisions of the current standard GB / T33593-2017: "Distributed power should have the ability to quickly monitor islanding and immediately disconnect from the power grid, and the anti-islanding protection action time should not be greater than 2 s". Therefore, it is of great significance to quickly and accurately detect islanding faults in distributed power.

[0003] Currently, the detection methods for islanding faults can be divided into three categories: active method, passive method, and remote method. The active method judges the islanding state by introducing a disturbance signal to make the voltage or frequency deviate from the normal value after the islanding occurs, such as frequency shift method, power disturbance method, impedance measurement method, etc. However, this method will reduce the power quality of distributed power. The passive method judges islanding according to the measured abnormal voltage or frequency, such as over / under voltage detection method, over / under frequency detection method, and voltage harmonic detection method, etc. This method has a large detection blind area when the power is matched, that is, when all the power required by the load in the distributed power grid is provided by the distributed power, and it is difficult to select a suitable detection threshold. The remote method realizes detection through signal acquisition and communication technology. The disadvantage of this method is the high detection cost and it is difficult to popularize and apply. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to provide an intelligent distributed power islanding detection method based on variational mode decomposition (VMD), Teager energy operator, and one-dimensional convolutional neural network, which can quickly and accurately judge whether an islanding event occurs without injecting a disturbance signal into the system, without setting a threshold, and without parameter tuning. The simulation results show that the detection method has simple steps, accurate classification, and strong anti-noise ability.

[0005] The present invention is realized through the following technical solutions:

[0006] A method for detecting islanding faults of distributed power sources, as Figure 1 shown, includes:

[0007] S1: Under the conditions of non-islanding fault conditions and different islanding fault conditions of the distributed power source, extract the modal components of the point of common coupling (PCC) voltage signal and the modal components of the converter output current signal;

[0008] S2: Use the variational mode decomposition parameters to perform variational mode decomposition on the modal components of the voltage signal and the current signal collected in S1, perform Teager energy operator demodulation, and obtain the feature vector space of the islanding fault characteristic signal;

[0009] S3: Divide the signal feature vector space obtained in S2 into a training set and a test set, and use them as the input signals of the neural network model respectively. The training set is used for training the neural network recognition model of the islanding fault state, and the test set is used to verify the classification performance of the neural network recognition model of the islanding fault state.

[0010] S4: Use the trained neural network recognition model of the islanding fault state to detect the islanding faults of the distributed power source.

[0011] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0012] The variational mode decomposition (VMD) adopted by the present invention is a signal processing technology proposed in 2014. Its essence is Wiener filtering, Hilbert transform and frequency mixing. It decomposes the signal into K discrete intrinsic mode functions to reproduce the specific sparsity of the input signal. The theoretical analysis of the VMD algorithm is as follows: (1) Perform Hilbert transform on the signal to convert it into an analytic signal and obtain the single-sided spectrum of the signal. (2) Modulate the spectrum of the mode to the baseband through the estimated center frequency. (3) Use the Gaussian smoothness estimate to demodulate the signal bandwidth. The VMD decomposition process is based on a constrained variational optimization problem. To solve the constraint problem, VMD introduces a quadratic penalty parameter and a Lagrange multiplier, and finds the saddle point of the augmented Lagrangian function in the iterative sub-optimization sequence, and uses the alternating direction multiplier method to solve this optimization problem.

[0013] The variational mode decomposition (VMD) converts the signal decomposition from a recursive screening mode into a non-recursive, variational mode decomposition mode, which is suitable for the feature extraction of non-stationary power oscillation signals in the power system. In the iterative screening process, the signal is decomposed into a series of modal functions with physical meanings according to the law of decreasing frequency one by one. Each modal function can decompose the corresponding amplitude and frequency, and finally the original signal is obtained through the recombination of the modal functions. The Teager energy operator is a non-linear difference operator, which can quickly capture the instantaneous change of the signal, and the calculation process is simple and fast, so it is widely used in the demodulation analysis of signals.

[0014] The present invention uses variational mode decomposition to characterize the three-phase PPC point voltage and the converter output current characteristics. At the same time, in order to accurately track the instantaneous changes of the intrinsic mode function component (IMF1) of the characteristic signal, the Teager energy operator is introduced to highlight the islanding fault characteristics, and the feature vector space composed of the Teager energy operator characteristics, the IMF2 component characteristics, and the IMF3 component characteristics is used as the input of a one-dimensional convolutional neural network (1D-CNN). It can quickly and accurately classify and identify islanding events and non-islanding events without injecting perturbation signals into the system, without setting thresholds, and without parameter tuning. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] In order to more clearly illustrate the technical solutions of the exemplary embodiments of the present invention, the drawings required for the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as limiting the scope. For those of ordinary skill in the art, other related drawings can be obtained based on these drawings without creative efforts. In the drawings:

[0016] Figure 1 is the flow chart of a method for detecting islanding faults of distributed power sources provided by an embodiment of the present invention;

[0017] Figure 2 is the characteristic waveform of the PCC point circuit breaker tripping voltage and its VMD decomposition result when the quality factor is 1 provided by an embodiment of the present invention;

[0018] Figure 3 is the Teager energy operator of the IMF2 of the PCC point circuit breaker tripping voltage when the quality factor is 1 provided by an embodiment of the present invention;

[0019] Figure 4 is the characteristic waveform of the PCC point circuit breaker tripping voltage and its VMD decomposition result when the quality factor is 2.5 provided by an embodiment of the present invention;

[0020] Figure 5 is the Teager energy operator of the IMF2 of the PCC point circuit breaker tripping voltage when the quality factor is 2.5 provided by an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0021] To make the objectives, technical solutions, and advantages of the present invention clearer and more understandable, the present invention will be further described in detail below in conjunction with the embodiments and the drawings. The illustrative embodiments and descriptions of the present invention are only used to explain the present invention and are not intended to limit the present invention.

[0022] Embodiment

[0023] A detection method for islanding faults of distributed power sources of the present invention includes:

[0024] T1: Under the conditions of non-islanding fault conditions and different islanding fault conditions of the distributed power source, respectively extract the modal components of the point of common coupling (PCC) voltage signal and the modal components of the converter output current signal of the distributed power source, specifically including:

[0025] T101: Set the quality factors to 1 and 2.5 respectively, and collect the PCC voltage signal and the converter output current signal when three different islanding fault conditions of PCC point breaker tripping, three-phase short circuit, and local load mutation and non-islanding fault conditions occur.

[0026] T102: Respectively extract the modal components of the PCC output voltage signal and the modal components of the converter output current signal.

[0027] T2: Use the variational mode decomposition parameters to perform variational mode decomposition on the output voltage signal and the output current signal of the distributed power source, and use the Teager energy operator for demodulation to obtain the signal feature vector space characterizing the islanding fault characteristics, specifically including:

[0028] T201: To avoid over-decomposition and incomplete decomposition in variational mode decomposition, preset the decomposition layer number to 3 and the penalty factor to 1500, set the range and step size of the decomposition layer number K, and calculate the relative entropy of the modal components corresponding to different K values, where the K value corresponding to the minimum relative entropy is the optimal decomposition layer number.

[0029] T202: Use the optimal decomposition layer number K obtained in T201, set the range and step size of the penalty factor, and calculate the relative entropy of the modal components corresponding to different penalty factors, where the penalty factor corresponding to the minimum relative entropy is the optimal penalty factor.

[0030] T203: According to the optimal decomposition layer number K and the penalty factor determined in steps T201 and T202, perform variational mode decomposition on the a, b, and c three phases in the PCC point voltage signal and the converter output current signal respectively to form the IMF feature vector of the PCC point three-phase voltage signal and the IMF feature vector of the three-phase current signal output by the converter.

[0031] T204: Respectively perform Teager energy operator demodulation on each phase IMF feature vector, extract the instantaneous information of each phase signal, and form the signal feature vector space characterizing the islanding fault characteristics.

[0032] T3: Divide the signal feature vector space containing the islanding fault characteristics into a training set and a test set, and use them as the input signals of the 1D-CNN model respectively, specifically including:

[0033] T301: Construct a one-dimensional convolutional neural network model for islanding fault characteristics, and set the number of input layer nodes according to the number of signal feature vector spaces selected in step T204.

[0034] T302: Set the number of output layer nodes to 2 for classification and identification of islanding fault states and non-islanding fault states.

[0035] T303: Divide the signal feature vector space selected in step T205 into a training set and a test set. Use the training set to train the 1D-CNN model to obtain the optimal parameters and form an islanding fault state recognition model. The test set is used to verify the classification performance of the network.

[0036] T304: Input the test set into the 1D-CNN model formed in T303 for classification performance testing.

[0037] T4: Use the trained neural network recognition model for islanding fault detection of distributed power sources.

[0038] In step T102, since different types of faults may occur in the system, all three-phase signals should be considered when analyzing the islanding fault of the system. At the same time, in order to reduce the analysis time of each phase and the computer memory, the modal components of the PCC point voltage signal and the converter output current signal are calculated using Equation (1).

[0039]

[0040] In the formula, I m is the modal current, V m is the modal voltage, m1 is the modal coefficient of phase a voltage, V a is the phase a voltage, m2 is the modal coefficient of phase b voltage, V b is the phase b voltage, m3 is the modal coefficient of phase c voltage, V c is the phase c voltage, n1 is the modal coefficient of phase a current, I a is the phase a current, n2 is the modal coefficient of phase b current, I b is the phase b current, n3 is the modal coefficient of phase c current, I c is the phase c current.

[0041] In step T203, the PCC point voltage signal and the converter output current signal are decomposed into a series of finite-bandwidth intrinsic mode functions through VMD using Equation (2), and the amplitude and frequency of the mode functions are oscillating signals that vary with time.

[0042]

[0043] In the formula, u k (t) is the mode function, A k(t) is the instantaneous amplitude, and φ(t) is the phase. For a signal f, its decomposition process is a constrained variational problem, which can be expressed as:

[0044]

[0045] In the formula, δ(t) is the Dirac function, is the square of the two-norm, and u k = {u1, u2..u k} is the IMF component after decomposition, and w k = {w1, w2..w k} is the center frequency of each component, is the gradient operation, and * is the convolution operation. To solve Equation (3), a quadratic penalty factor a and a Lagrange multiplier λ are introduced to convert the constrained problem into a corresponding unconstrained problem as shown in Equation (4).

[0046]

[0047] To solve the saddle point of the augmented Lagrangian function in Equation (4), the alternating direction multiplier method is used to update u k , w k and λ, and the update formulas are (5)-(7).

[0048]

[0049]

[0050]

[0051] In the formula, n is the number of iterations, and τ is the fidelity coefficient. Repeat Equations (5)-(7) until the condition in Equation (8) is satisfied, then stop the iteration.

[0052]

[0053] In the formula, ε > 0 is the discrimination accuracy.

[0054] In steps T201 and T202, the relative entropy of each IMF component is calculated through Equation (9), and the decomposition layer and penalty factor corresponding to the minimum relative entropy are determined as the optimal solutions.

[0055]

[0056] In the formula, p(x i ) is the target probability distribution, q(x i ) is the theoretical probability distribution, N is the distribution length, x i is the discrete random variable, and i is the permutation serial number of the discrete random variable.

[0057] In step T204, the expression for the Teager energy operator to solve a continuous signal s(t) is:

[0058]

[0059]

[0060] In the formula, ψ[s(t)] is the instantaneous change energy value of the continuous signal, and are respectively and t is time, f(t) is the frequency of the continuous signal, and a(t) is the amplitude of the continuous signal. After discretizing the continuous signal s(t), the discrete signal s(n) is obtained, and the corresponding Teager energy operator expression is as shown in Equation (12).

[0061] ψ[s(n)] = s 2 (n) - s(n + 1)s(n - 1) (12)

[0062] In step T301, in the convolutional layer of the 1D-CNN, the convolutional kernel convolves the input, extracts the features of the local area, and constructs the output features using a non-linear activation function. The output of each layer is the convolution result of multiple input features. The advantage of the convolutional kernel is that it can obtain rotation-invariant features, and its mathematical expression is described as:

[0063]

[0064]

[0065] In the formula, and respectively represent the weight and bias of the i-th convolutional kernel in the l-th layer, x l (j) represents the i-th area in the l-th layer, * represents the convolution operation, represents the j-th feature map of x l (j). To enhance the non-linear expression ability of the input signal and make the learned features clearer, the parameters are adjusted in combination with the backpropagation learning method to accelerate the convergence speed. In this paper, a tanh activation function is added after the convolutional layer. In Equation (14), f(·) is the activation function.

[0066] The pooling layer samples the large matrix into a small matrix through data sampling, reduces the parameters and computational amount of the neural network, thereby avoiding overfitting. In practical applications, the commonly used one is Max-pooling, and its expression is:

[0067]

[0068] In the formula, represents the value of the t-th neuron in the i-th feature at layer l, and W is the width of the pooling layer.

[0069] The fully connected layer can integrate the local information of the convolutional layer and the pooling layer, and construct the output of the previous pooling layer into a one-dimensional vector as the input of the fully connected layer. Its expression is:

[0070]

[0071] In the formula, and are the weight matrix and bias of the fully connected layer, is the output of the pooling layer.

[0072] In step T302, a multi-class Softmax classifier is used as the output layer. It is an extension of Logistic regression, and its expression is:

[0073]

[0074] In the formula, z o (j) takes the logarithm of the output of the j-th neuron in the output layer, and M represents the total number of categories.

[0075] To verify the performance of the proposed detection method under different working conditions, the VMD-Teager energy operator feature extraction is illustrated by taking voltage as an example. The tripping states of the PCC point circuit breakers with quality factors of 1 and 2.5 are set to obtain training and test data, and the algorithm performance is verified at the same time.

[0076] As Figure 2 shown is the tripping characteristic waveform of the PCC point circuit breaker with a quality factor of 1. When the circuit breaker trips at 0.8 s, the voltage at the PCC point does not change significantly. At this time, the load quality factor is 1, and the load resonance frequency is exactly equal to the working frequency of 50 Hz. All the power required by the load is provided by the distributed power system. At this time, both the over / under voltage and over / under frequency detection methods cannot effectively identify the islanding event, and it is difficult for the test system to detect the islanding, and the conditions are the most stringent. Figure 2 It can be seen that the mode functions IMF2 and IMF3 decomposed by VMD are slightly distorted before and after tripping at 0.8 s, but the mode functions cannot accurately characterize the islanding characteristics. To detect the transient impact of the mode function IMF2, the Teager energy operator is used to obtain the energy sequence as Figure 3 shown, and the energy sequence is significantly distorted at 0.8 s. By changing the RLC load parameters to change the load quality factor, Figure 4 shown is the tripping characteristic waveform of the PCC point circuit breaker with a quality factor of 2.5. As Figure 4 when the circuit breaker is disconnected at 0.8 s, the voltage at the PCC point changes significantly and shows a three-phase unbalanced state. Figure 5The VMD-Teager energy operator shown above can well characterize the island feature.

[0077] Those skilled in the art should understand that the embodiments of the present application can be provided as methods, systems, or computer program products. Therefore, the present application can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code.

[0078] The present application is described with reference to the flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each flow and / or block in the flowchart and / or block diagram, as well as the combination of flows and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing devices generate means for implementing the functions specified in Figure 1 one or more of the processes or multiple processes and / or blocks Figure 1 one or more of the blocks.

[0079] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer-readable memory generate a manufactured article including instruction means, and the instruction means implements the functions specified in Figure 1 one or more of the processes or multiple processes and / or blocks Figure 1 one or more of the blocks.

[0080] These computer program instructions can also be loaded onto a computer or other programmable data processing device, so that a series of operation steps are executed on the computer or other programmable device to generate a computer-implemented process, and thus the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in Figure 1 one or more of the processes or multiple processes and / or blocks Figure 1 one or more of the blocks.

[0081] Those of ordinary skill in the art can understand that all or part of the steps in implementing the above facts and methods can be completed by instructing relevant hardware through a program. The program involved or the said program can be stored in a computer-readable storage medium. When the program is executed, it includes the following steps: At this time, the corresponding method steps are introduced. The storage medium can be ROM / RAM, magnetic disk, optical disc, etc.

[0082] The specific embodiments described above further elaborate on the purpose, technical solutions, and beneficial effects of the present invention. It should be understood that the above description is only the specific embodiments of the present invention and is not used to limit the protection scope of the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A detection method for islanding faults of distributed power sources, characterized in that, It includes the following steps: S1: Under the conditions of non-island fault conditions of distributed power sources and three island fault conditions of the circuit breaker tripping at the point of common coupling, three-phase short circuit, and local load mutation, extract the modal components of the voltage signal at the point of common coupling and the modal components of the output current signal of the converter respectively; S2: Using the variational mode decomposition parameters, perform variational mode decomposition on the modal components of the voltage signal and the current signal collected in step S1. Perform variational mode decomposition on the a, b, and c phases of the common coupling point voltage signal and the converter output current signal respectively to form the IMF feature vectors of the three-phase voltage signal at the common coupling point and the IMF feature vectors of the three-phase current signal output by the converter; By calculating the relative entropy D of each IMF feature vector KL , determine that the decomposition layer number and penalty factor corresponding to the minimum relative entropy are the optimal solutions. In the formula, p(x i ) is the target probability distribution, q(x i ) is the theoretical probability distribution, N is the distribution length, x i is a discrete random variable, and i is the arrangement serial number of the discrete random variable; perform Teager energy operator demodulation to obtain the feature space of the islanding fault characteristic signal; S3: Divide the feature space obtained in step S2 into a training set and a test set, which are used as the input signals of the neural network model respectively. The training set is used for training the neural network recognition model for island fault states, and the test set is used to verify the classification performance of the neural network recognition model for island fault states; S4: Use the trained neural network recognition model for island fault states to detect the island faults of distributed power sources.

2. The detection method for islanding faults of distributed power sources according to claim 1, characterized in that, In step S1, the modal components of the voltage signal at the point of common coupling and the modal components of the output current signal of the converter are calculated according to the following formula Where, I m is the modal current, V m is the modal voltage, m1 is the voltage modal coefficient of phase a, V a is the voltage of phase a, m2 is the voltage modal coefficient of phase b, V b is the voltage of phase b, m3 is the voltage modal coefficient of phase c, V c is the voltage of phase c, n1 is the current modal coefficient of phase a, I a is the current of phase a, n2 is the current modal coefficient of phase b, I b is the current of phase b, n3 is the current modal coefficient of phase c, I c is the current of phase c.

3. A detection method for islanding faults of distributed power sources according to claim 1, characterized in that, In step S2, the Teager energy operator demodulation has the following expression: Where ψ[s(t)] is the instantaneous change energy value of the continuous signal, and are respectively and t is time, f(t) is the frequency of the continuous signal, a(t) is the amplitude of the continuous signal. After discretizing the continuous signal s(t), the discrete signal s(n) is obtained. The corresponding Teager energy operator expression is as in the formula ψ[s(n)] = s 2 (n) - s(n + 1)s(n - 1).

4. A detection method for islanding faults of distributed power sources according to claim 1, characterized in that In step S2, the variational mode decomposition parameters include the decomposition layer number and the penalty factor. Among them, the preset value range of the decomposition layer number is an integer from 1 to 6, and the preset value range of the penalty factor is from 200 to 3000.

5. A detection method for islanding faults of distributed power sources according to claim 1, characterized in that, In step S3, the neural network model uses a one-dimensional convolutional neural network.

6. A detection method for islanding faults of distributed power sources according to claim 1, characterized in that In step S3, the neural network model sets the number of input layer nodes according to the number of selected signal feature vector spaces.

7. A detection method for islanding faults of distributed power sources according to claim 1, characterized in that In step S3, the number of output layer nodes is set to 2 for classification and recognition of island fault states and non-island fault states.

Images