A distributed power supply island detection method based on information entropy

Abstract

The present invention relates to a power equipment detection technology, and in particular to a distributed power supply island detection method based on information entropy. The method models a power distribution system including a distributed power supply that needs island detection through a real-time digital simulation system, and obtains an instantaneous active power data set by multiple simulations under different conditions; the data set is split into a training set and a test set, and two support vector machines are trained with a data length of 40 milliseconds and 60 milliseconds respectively; after the training is completed, the probability density curves of the classification accuracy of the two support vector machines are obtained based on the time when the traveling wave enters the detection window pane; the probability density curves are fitted and the information entropies of the two support vector machines are obtained respectively; and the information entropies are compared to obtain the final island detection result. The method performs on-site island detection on the distributed power supply in the microgrid, and ensures the normal and stable operation of the microgrid. The accuracy of distributed power supply island detection is improved, and the classification accuracy is high, the detection time is short, and the credibility is strong.

Description

Technical Field

[0001] The present invention belongs to the technical field of power equipment detection, and in particular relates to a distributed power supply island detection method based on information entropy. Background Art

[0002] The continuous development of distributed energy has brought about the microgrid operation mode of the distribution system. Among them, for unplanned island operation, the distributed power supply quickly and accurately identifies the island operation state and exits the power supply, which is a necessary condition for ensuring the safety of grid equipment and personnel. A fast, reliable, advanced and effective distributed power supply island detection method helps to achieve stable and reliable operation and efficient management of microgrids, improve the overall safety and stability of microgrids, and promptly eliminate the safety threats caused by unplanned islands to grid operation and personal equipment.

[0003] The existing distributed power supply island detection methods mainly include active method and passive method. The active method is not suitable for all types of distributed power supplies because the injection signal will cause the power quality to deteriorate, and the engineering implementation relies on power electronic equipment. The passive method detects by comparing the electrical quantity signal and the threshold value. Overall, it has the disadvantages of not being applicable to all distributed power supplies, a large undetectable area, false detection, missed detection, and low detection efficiency. Summary of the invention

[0004] In view of the problems existing in the background technology, the present invention provides a distributed power supply island detection method based on information entropy.

[0005] In order to solve the above technical problems, the present invention adopts the following technical solution: a distributed power supply island detection method based on information entropy, comprising the following steps:

[0006] Step 1: Use a real-time digital simulation system to model the distributed power sources that need to be detected, perform n simulations under different islanding and non-islanding conditions, and take the instantaneous active power of the monitoring point to obtain a distributed generation islanding detection data set;

[0007] Step 2: split the distributed generation island detection data set into a training set and a test set, and in the training set, use the sampling time of 40 milliseconds and the sampling time of 60 milliseconds as the data length, and select the Gaussian kernel function to train the first support vector machine and the second support vector machine respectively;

[0008] Step 3, respectively obtaining the probability density functions of the classification accuracy of the first support vector machine and the second support vector machine at different traveling wave entry times;

[0009] Step 4: Use a polynomial to fit the probability density function curve to obtain the information entropy of the first support vector machine and the second support vector machine, and compare the smaller information entropy to obtain the island detection result.

[0010] In the above-mentioned distributed power supply island detection method based on information entropy, the implementation of step 1 includes the following steps:

[0011] Step 1.1: The instantaneous active power at the monitoring point is sampled through the island detection point voltage u a ,u b ,u c and current i a ,i b ,i c Calculated; the Clark transform process is as follows:

[0012]

[0013]

[0014]

[0015] p km =u αkm i αkm +u βkm i βkm

[0016] Among them, k∈[1,n], m∈[1,L], L represents the total simulation time, n represents the total number of simulations, p km represents the instantaneous active power at the mth millisecond of the kth simulation, u akm ,u bkm ,u ckm They represent the A phase voltage, B phase voltage, and C phase voltage at the mth millisecond of the kth simulation, respectively. akm ,i bkm ,i ckm They represent the A-phase current, B-phase current, and C-phase current at the mth millisecond of the kth simulation, respectively. αkm ,u βkm They represent the α-phase voltage and β-phase voltage after Clark transformation, respectively. αkm ,i βkm They represent the α-phase current and β-phase current after Clark transformation respectively;

[0017] Step 1.2: Perform n simulations under different island and non-island conditions for the following situations:

[0018] I. Island operation:

[0019] 1) Active power mismatch:

[0020]

[0021] Among them, ΔP represents the active power shortage, P DG Indicates the active power output of distributed generation;

[0022] 2) Reactive power mismatch:

[0023]

[0024] Among them, ΔQ represents the reactive power shortage, Q DG Represents the reactive power output of distributed generation;

[0025] II. Non-island operation:

[0026] 1) Normal operation;

[0027] 2) Fault: including single-phase grounding fault, phase-to-phase fault, two-phase grounding fault, three-phase fault, and three-phase grounding fault; during the fault process:

[0028] R f =1~50Ω

[0029] Among them, R f Indicates the fault transition resistance;

[0030] 3) Load switching: line load switching during operation;

[0031] 4) Capacitor switching: capacitor switching during operation;

[0032] Step 1.3, the distributed generation island detection data set comes from the distributed generation output side that needs island detection, the simulation of the first support vector machine and the second support vector machine are consistent, and the time window length is different;

[0033] The distributed generation island detection dataset is represented as:

[0034]

[0035] Where, Γ represents the instantaneous active power data set obtained by simulation, which is used for the first support vector machine. The detection window duration is two cycles of 40ms, and each vector has 20 data points. m_i represents the instantaneous active power value of the monitoring point in the i-th case at the m-th millisecond, i∈[1,l], l represents the total number of simulation cases;

[0036]

[0037] Among them, Ψ represents the instantaneous active power data set obtained by simulation, which is used for the second support vector machine. The detection window length is three cycles of 60ms, and each vector has 30 data points. m_i represents the instantaneous active power value of the monitoring point in the i-th case at the m-th millisecond, i∈[1,l], and l represents the total number of simulation cases.

[0038] In the above-mentioned distributed power supply island detection method based on information entropy, the implementation of step 2 includes the following steps:

[0039] Step 2.1, Gaussian kernel function belongs to radial basis function, that is:

[0040] K(x i ,x j )=exp(-γ||x i -x j || 2 )

[0041] where i,j∈[1,n], x i ,x j represents the eigenvector, K(x i ,x j ) represents the kernel function, γ∈(0,+∞) is the kernel function parameter of the support vector machine;

[0042] Step 2.2: The training process of the first support vector machine and the second support vector machine is as follows:

[0043] Solve the weight and bias of the separating hyperplane, seek the maximum margin, and introduce the Lagrange multiplier to express it as the following dual problem:

[0044]

[0045] Among them, ω represents the hyperplane weight vector, b represents the hyperplane bias, and y i ∈{-1,1} represents the manual calibration result corresponding to island detection, -1 represents non-island, 1 represents island, and the Lagrange multiplier ξ is introduced i , partial differentials can be used to solve the extreme value problem, the process is as follows:

[0046]

[0047]

[0048] The final support vector machine solution is as follows:

[0049]

[0050] Where ω represents the hyperplane weight vector, b represents the hyperplane bias, i,j∈[1,n], x i ,x j represents the feature vector, y i ,y j ∈{-1,1} represents the manual calibration result corresponding to island detection, -1 represents non-island, 1 represents island, ξ i ,ξ j represents the Lagrange multiplier, l represents the total number of simulation cases;

[0051] Relying on the radial basis kernel function, the decision function of the separation hyperplane can be expressed as:

[0052]

[0053] Among them, θ i is the coefficient of the i-th support vector, x i is the i-th support vector;

[0054] Subsequent support vector classification relies on the trained decision function part for classification output.

[0055] In the above-mentioned distributed power supply island detection method based on information entropy, the probability density functions of the classification accuracy of the first support vector machine and the second support vector machine in step 3 are respectively denoted as ζ Γ (x) and ζ Ψ (x) represents the probability density function of the recognition accuracy of the first support vector machine and the second support vector machine during the island detection process over time, which is fitted by a polynomial equation and the fitting index does not exceed 4.

[0056] In the above-mentioned distributed power supply island detection method based on information entropy, the implementation of step 4 includes the following steps:

[0057] Step 4.1: Use differential entropy to calculate information entropy. The differential entropy formula of the first support vector machine is as follows:

[0058]

[0059] Where h(Γ) represents the entropy value of the first support vector machine, ζ Γ (x) is the probability density function of the classification accuracy of the first support vector machine;

[0060] Similarly, the differential entropy formula of the second support vector machine is as follows:

[0061]

[0062] Among them, h(Ψ) represents the entropy value of the second support vector machine, ζ Ψ (x) is the probability density function of the classification accuracy of the second support vector machine;

[0063] Step 4.2: When the classification result of the first support vector machine changes, the system detects island operation and waits for 20ms before outputting the result of the second support vector machine; if the results of the first support vector machine and the second support vector machine are the same, it is judged as island operation; if the judgment results are different, h(Γ) and h(Ψ) are compared. If h(Γ) is smaller, the result of the first support vector machine is output; if h(Ψ) is smaller, the result of the second support vector machine is output. Compared with the prior art, the beneficial effects of the present invention are:

[0064] The present invention performs on-site island detection on distributed power sources in a microgrid by using a support vector machine method based on information entropy, thereby ensuring the normal and stable operation of the microgrid.

[0065] The present invention evaluates the credibility of the results of data-driven algorithms (support vector machines) by comparing and quantifying information entropy, providing a reference for the application of data-driven algorithms in the field of engineering practice.

[0066] The present invention effectively improves the accuracy of island detection classification problems through complementary detection of two support vector machines, and the detection blind area can reach zero.

[0067] Compared with the existing frequency difference island detection method and active injection method, the method of the present invention has the advantages of short detection time, high detection accuracy, and can be applied to synchronous machine distributed power supplies, which can effectively solve the problem that distributed power supplies cannot identify island operation in time. BRIEF DESCRIPTION OF THE DRAWINGS

[0068] Figure 1 : This is a framework diagram of the distributed power generation island detection process according to an embodiment of the present invention;

[0069] Figure 2 : A schematic diagram of three-phase voltage and current sampling in the detection window of the first support vector machine model and the second support vector machine model according to an embodiment of the present invention;

[0070] Figure 3 : It is a framework diagram of the support vector machine training process according to an embodiment of the present invention. DETAILED DESCRIPTION

[0071] The technical solutions in the embodiments of the present invention will be described clearly and completely below in combination with the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0072] It should be noted that, in the absence of conflict, the embodiments of the present invention and the features in the embodiments may be combined with each other.

[0073] The present invention will be further described below in conjunction with specific embodiments, but the present invention is not limited thereto.

[0074] This embodiment uses a real-time digital simulation system to model a power distribution system containing distributed power sources that require island detection, and simulates multiple times under different conditions to obtain an instantaneous active power data set; split the data set into a training set and a test set, and train two support vector machines with a data length of 40 milliseconds and 60 milliseconds respectively; after the training is completed, the probability density curve of the classification accuracy of the two support vector machines is obtained based on the time when the traveling wave enters the detection pane; fit the probability density curve and obtain the information entropy of the two support vector machines respectively; compare the information entropy to obtain the final island detection result. Not only does it improve the accuracy of distributed power supply island detection and the safety of microgrid distributed power supplies, but it also has high classification accuracy, short detection time, and strong credibility.

[0075] This embodiment is implemented by the following technical solution, a distributed power supply island detection method based on information entropy, comprising the following steps:

[0076] S1: Model the distributed power sources that need to be detected through a real-time digital simulation system, perform n simulations under different islanding and non-islanding conditions, and take the instantaneous active power of the monitoring point to obtain a distributed generation islanding detection data set;

[0077] S2: Split the distributed generation island detection data set into a training set and a test set. In the training set, the first support vector machine and the second support vector machine are trained respectively with a sampling time of 40 milliseconds and a sampling time of 60 milliseconds as the data length and a Gaussian kernel function;

[0078] S3: Obtaining probability density functions of the classification accuracy of the first support vector machine and the second support vector machine at different traveling wave entry times;

[0079] S4: polynomial fitting probability density function curve is used to obtain the information entropy of the first support vector machine and the second support vector machine, and the smaller information entropy is compared to obtain the island detection result.

[0080] Moreover, the instantaneous active power is determined by the voltage u sampled at the island detection point. a ,u b ,u c and current i a ,i b ,i c The Clark transform process is as follows:

[0081]

[0082]

[0083]

[0084] p km =u αkm i αkm +u βkm i βkm

[0085] Among them, k∈[1,n], m∈[1,L], L represents the total simulation time, n represents the total number of simulations, p km represents the instantaneous active power at the mth millisecond of the kth simulation, u akm ,u bkm ,u ckm represents the voltage of phase A, phase B, and phase C at the mth millisecond of the kth simulation, i akm ,i bkm ,i ckm represents the A phase current, B phase current, and C phase current at the mth millisecond of the kth simulation, u αkm ,u βkm represents the α-phase voltage and β-phase voltage after Clark transformation, i αkm ,i βkm Represents the α-phase current and β-phase current after Clark transformation.

[0086] The simulations were performed n times with different islanding and non-islanding conditions under the following conditions:

[0087] Isolated operation:

[0088] 1) Active power mismatch:

[0089]

[0090] Among them, ΔP represents the active power shortage, P DG Represents the active power output of distributed generation.

[0091] 2) Reactive power mismatch:

[0092]

[0093] Among them, ΔQ represents the reactive power shortage, Q DG Represents the reactive power output of distributed generation.

[0094] Non-island operation:

[0095] 1) Normal operation.

[0096] 2) Fault: including single-phase grounding fault, phase-to-phase fault, two-phase grounding fault, three-phase fault, three-phase grounding fault.

[0097] R f=1~50Ω

[0098] Among them, R f Represents the fault transition resistance.

[0099] 3) Load switching: The line load is switched during operation.

[0100] 4) Capacitor switching: Capacitor switching during operation.

[0101] The distributed power data set comes from the output side of distributed generation that requires island detection. The simulations of the first support vector machine and the second support vector machine are consistent, with only the time window length being different.

[0102] S1 The dataset is represented as:

[0103]

[0104] Where Γ represents the instantaneous active power data set obtained through simulation, which is used for the first support vector machine. The detection window duration is two cycles of 40ms, and each vector has 20 data points. m_i represents the instantaneous active power value of the monitoring point in the i-th case at the m-th millisecond, i∈[1,l], and l represents the total number of simulation cases.

[0105]

[0106] Where Ψ represents the instantaneous active power data set obtained through simulation, which is used for the second support vector machine. The detection window length is three cycles of 60ms, and each vector has 30 data points. m_i represents the instantaneous active power value of the monitoring point in the i-th case at the m-th millisecond, i∈[1,l], and l represents the total number of simulation cases.

[0107] Moreover, the Gaussian kernel function in S2 belongs to the radial basis function, that is,

[0108] K(x i ,x j )=exp(-γ||x i -x j || 2 )

[0109] where i,j∈[1,n], x i ,x j represents the eigenvector, K(x i ,x j ) represents the kernel function, and γ∈(0,+∞) is the kernel function parameter of the support vector machine.

[0110] The support vector training process is mainly to solve the weight and bias of the separation hyperplane, seek the maximum interval, and introduce the Lagrange multiplier to express it as the following dual problem:

[0111]

[0112] Where ω represents the hyperplane weight vector, b represents the hyperplane bias, and y i ∈{-1,1} represents the manual calibration result of the corresponding island detection (-1 represents non-island, 1 represents island), and the Lagrange multiplier ξ is introduced i , partial differentials can be used to solve the extreme value problem, the process is as follows:

[0113]

[0114]

[0115] The final support vector machine solution can be expressed as follows:

[0116]

[0117] Where ω represents the hyperplane weight vector, b represents the hyperplane bias, i,j∈[1,n], x i ,x j represents the eigenvector, y i ,y j ∈{-1,1} represents the manual calibration result of the corresponding island detection (-1 represents non-island, 1 represents island), ξ i ,ξ j represents the Lagrange multiplier and l represents the total number of simulation cases.

[0118] Relying on the radial basis kernel function, the decision function of the separation hyperplane can be re-expressed as follows:

[0119]

[0120] where θ i is the coefficient of the i-th support vector, x i is the i-th feature support vector. The subsequent support vector classification solution mainly relies on the trained decision function part for classification output.

[0121] Moreover, the probability density functions of the classification accuracy of the first support vector machine and the second support vector machine are respectively denoted as ζ Γ (x) and ζ Ψ (x) represents the probability density function of the recognition accuracy of support vector machine A and support vector machine B during island detection. The fitting process should adopt a polynomial equation and the fitting index should not exceed 4.

[0122] Moreover, the information entropy of the first support vector machine and the second support vector machine is preferably calculated using differential entropy. The differential entropy formula of the first support vector machine is as follows:

[0123]

[0124] Where h(Γ) represents the entropy value, ζ Γ (x) is the probability density function of the classification accuracy of the first support vector machine.

[0125] Similarly, the differential entropy formula of the second support vector machine can be written as follows:

[0126]

[0127] Where h(Ψ) represents the entropy value of the support vector machine B, ζ Ψ (x) is the probability density function of the classification accuracy of the second support vector machine.

[0128] In specific implementation, the island operation condition is For example, it represents the island operation status being detected.

[0129] like Figure 1 As shown, it is a basic framework diagram of the distributed generation island detection process, including simulation model construction, data set generation, model training completion, and island operation result detection.

[0130] like Figure 2 As shown, it is a schematic diagram of three-phase voltage and current sampling.

[0131] like Figure 3 The figure shows the framework diagram of the support vector machine training process.

[0132] Combine the following Figures 1 to 3 The specific implementation of the present invention is an island detection method based on instantaneous active power and support vector machine, which specifically includes the following steps:

[0133] First, the distributed power sources that need to be detected are modeled through a real-time digital simulation system, and n simulations are performed under different islanding and non-islanding conditions. The instantaneous active power of the monitoring point is taken to obtain a distributed generation islanding detection data set.

[0134] The simulation model should simulate islanding and non-islanding events under different conditions, where the range of islanding operation conditions is as follows:

[0135] 1) Active power mismatch:

[0136]

[0137] Among them, ΔP represents the active power shortage, P DG Represents the active power output of distributed generation.

[0138] 2) Reactive power mismatch:

[0139]

[0140] Among them, ΔQ represents the reactive power shortage, Q DG Represents the reactive power output of distributed generation.

[0141] The value range of non-island operation conditions is as follows:

[0142] 1) Normal operation.

[0143] 2) Fault: including single-phase grounding fault, phase-to-phase fault, two-phase grounding fault, three-phase fault, three-phase grounding fault.

[0144] R f =1~50Ω

[0145] Among them, R f Represents the fault transition resistance.

[0146] 3) Load switching: The line load is switched during operation.

[0147] 4) Capacitor switching: Capacitor switching during operation.

[0148] The instantaneous active power of the monitoring point is obtained by sampling the voltage u at the island detection point. a ,u b ,u c and current i a ,i b ,i c The Clark transform process is as follows:

[0149]

[0150]

[0151]

[0152] p km =u αkm i αkm +u βkm i βkm

[0153] Among them, k∈[1,n], m∈[1,L], L represents the total simulation time, n represents the total number of simulations, p km represents the instantaneous active power at the mth millisecond of the kth simulation, u akm ,u bkm ,u ckm represents the voltage of phase A, phase B, and phase C at the mth millisecond of the kth simulation, i akm ,i bkm,i ckm represents the A-phase current, B-phase current, and C-phase current at the mth millisecond of the kth simulation, u αkm ,u βkm represents the α-phase voltage and β-phase voltage after Clark transformation, i αkm ,i βkm Represents the α-phase current and β-phase current after Clark transformation.

[0154] The dataset is represented as:

[0155]

[0156] Where Γ represents the instantaneous active power data set obtained through simulation, which is used for the first support vector machine. The detection window duration is two cycles of 40ms, and each vector has 20 data points. m_i represents the instantaneous active power value of the monitoring point in the i-th case at the m-th millisecond, i∈[1,l], and l represents the total number of simulation cases.

[0157]

[0158] Where Ψ represents the instantaneous active power data set obtained through simulation, which is used for the second support vector machine. The detection window length is three cycles of 60ms, and each vector has 30 data points. m_i represents the instantaneous active power value of the monitoring point in the i-th case at the m-th millisecond, i∈[1,l], and l represents the total number of simulation cases.

[0159] Second, the distributed power supply island detection data set is split into a training set and a test set. In the training set, the sampling time is 40 milliseconds and the sampling time is 60 milliseconds as the data length, and the Gaussian kernel function is selected to train the first support vector machine model and the second support vector machine model respectively;

[0160] The training process of the first support vector machine and the second support vector machine is the same, only the data sets are different. The main training process is as follows:

[0161] To find the maximum interval, the Lagrange multiplier is introduced to express it as the following dual problem:

[0162]

[0163] Where ω represents the hyperplane weight vector, b represents the hyperplane bias, and y i ∈{-1,1} represents the manual calibration result of the corresponding island detection (-1 represents non-island, 1 represents island), and the Lagrange multiplier ξ is introduced i , partial differentials can be used to solve the extreme value problem, the process is as follows:

[0164]

[0165]

[0166] The final SVM solution result can be expressed as follows:

[0167]

[0168] Where ω represents the hyperplane weight vector, b represents the hyperplane bias, i,j∈[1,n], x i ,x j represents the eigenvector, y i ,y j ∈{-1,1} represents the manual calibration result of the corresponding island detection (-1 represents non-island, 1 represents island), ξ i ,ξ j represents the Lagrange multiplier and l represents the total number of simulation cases.

[0169] Relying on the radial basis kernel function, the decision function of the separation hyperplane can be re-expressed as follows:

[0170]

[0171] where θ i is the coefficient of the i-th support vector, x i is the i-th feature support vector. The subsequent support vector classification solution mainly relies on the trained decision function part for classification output.

[0172] Third, the probability density functions of the classification accuracy of the first support vector machine and the second support vector machine are obtained respectively with different traveling wave entry times;

[0173] The probability density functions of the classification accuracy of the first support vector machine and the second support vector machine are denoted as ζ Γ (x) and ζ Ψ (x), respectively represent the probability density functions of the recognition accuracy of the first support vector machine and the second support vector machine during the island detection process. The fitting process is preferably performed using a polynomial equation and the fitting index should not exceed 4.

[0174] Fourth, the information entropy of the first support vector machine and the second support vector machine is obtained by fitting the probability density function curve with a polynomial, and the smaller information entropy is compared to obtain the island detection result.

[0175] Information entropy is calculated preferentially using differential entropy. The differential entropy formula of the first support vector machine is as follows:

[0176]

[0177] Where h(Γ) represents the entropy value. Similarly, the differential entropy formula of the second support vector machine can be written as follows:

[0178]

[0179] Where h(Ψ) represents the entropy value of support vector machine B.

[0180] When the classification result of the first support vector machine changes, the system detects island operation and waits for 20ms before outputting the result of the second support vector machine. If the results of the first and second support vector machines are the same, it is finally judged as island operation. If the judgment results are different, h(Γ) and h(Ψ) are compared. If h(Γ) is smaller, the result of the first support vector machine is output. If h(Ψ) is smaller, the result of the second support vector machine is output.

[0181] The above are only preferred embodiments of the present invention, and are not intended to limit the implementation methods and protection scope of the present invention. Those skilled in the art should be aware that all solutions obtained by equivalent substitutions and obvious changes made using the contents of the specification of the present invention should be included in the protection scope of the present invention.

Claims

1. A distributed power supply island detection method based on information entropy, characterized by: The following steps are involved: Step 1: Use a real-time digital simulation system to model the distributed power sources that need to be detected, perform n simulations under different islanding and non-islanding conditions, and take the instantaneous active power of the monitoring point to obtain a distributed generation islanding detection data set; Step 2: split the distributed generation island detection data set into a training set and a test set, and in the training set, use the sampling time of 40 milliseconds and the sampling time of 60 milliseconds as the data length, and select the Gaussian kernel function to train the first support vector machine and the second support vector machine respectively; Step 3, respectively obtaining the probability density functions of the classification accuracy of the first support vector machine and the second support vector machine at different traveling wave entry times; Step 4: Use a polynomial to fit the probability density function curve to obtain the information entropy of the first support vector machine and the second support vector machine, and compare the smaller information entropy to obtain the island detection result; The implementation of step 4 includes the following steps: Step 4.1: Use differential entropy to calculate information entropy. The differential entropy formula of the first support vector machine is as follows: Where h(Γ) represents the entropy value of the first support vector machine, ζ Γ (x) is the probability density function of the classification accuracy of the first support vector machine; Similarly, the differential entropy formula of the second support vector machine is as follows: Among them, h(Ψ) represents the entropy value of the second support vector machine, ζ Ψ (x) is the probability density function of the classification accuracy of the second support vector machine; Step 4.2: When the classification result of the first support vector machine changes and the system detects island operation, it waits for 20ms and then outputs the result of the second support vector machine; if the results of the first support vector machine and the second support vector machine are the same, it is judged to be island operation; if the judgment results are different, compare the sizes of h(Γ) and h(Ψ). If h(Γ) is smaller, output the result of the first support vector machine; if h(Ψ) is smaller, output the result of the second support vector machine.

2. The distributed power supply island detection method based on information entropy according to claim 1 is characterized in that: The implementation of step 1 includes the following steps: Step 1.1: The instantaneous active power at the monitoring point is sampled through the island detection point voltage u a ,u b ,u c and current i a ,i b ,i c Calculated; the Clark transform process is as follows: p km =in αkm and αkm +in βkm and βkm Among them, k∈[1,n], m∈[1,L], L represents the total simulation time, n represents the total number of simulations, p km represents the instantaneous active power at the mth millisecond of the kth simulation, u akm ,u bkm ,u ckm They represent the A phase voltage, B phase voltage, and C phase voltage at the mth millisecond of the kth simulation, respectively. akm ,i bkm ,i ckm They represent the A-phase current, B-phase current, and C-phase current at the mth millisecond of the kth simulation, respectively. αkm ,u βkm They represent the α-phase voltage and β-phase voltage after Clark transformation, respectively. αkm ,i βkm They represent the α-phase current and β-phase current after Clark transformation respectively; Step 1.2: Perform n simulations under different island and non-island conditions for the following situations: I. Island operation: 1) Active power mismatch: Among them, ΔP represents the active power shortage, P DG Indicates the active power output of distributed generation; 2) Reactive power mismatch: Among them, ΔQ represents the reactive power shortage, Q DG Represents the reactive power output of distributed generation; II. Non-island operation: 1) Normal operation; 2) Fault: including single-phase grounding fault, phase-to-phase fault, two-phase grounding fault, three-phase fault, and three-phase grounding fault; during the fault process: R f =1~50Ω Among them, R f Indicates the fault transition resistance; 3) Load switching: line load switching during operation; 4) Capacitor switching: capacitor switching during operation; Step 1.3, the distributed generation island detection data set comes from the distributed generation output side that needs island detection, the simulation of the first support vector machine and the second support vector machine are consistent, and the time window length is different; The distributed generation island detection dataset is represented as: Where, Γ represents the instantaneous active power data set obtained through simulation, which is used for the first support vector machine. The detection window duration is two cycles of 40ms, and each vector has 20 data points. m_i represents the instantaneous active power value of the monitoring point in the i-th case at the m-th millisecond, i∈[1,l], l represents the total number of simulation cases; Among them, Ψ represents the instantaneous active power data set obtained by simulation, which is used for the second support vector machine. The detection window length is three cycles of 60ms, and each vector has 30 data points. m_i represents the instantaneous active power value of the monitoring point in the i-th case at the m-th millisecond, i∈[1,l], and l represents the total number of simulation cases.

3. The distributed power supply island detection method based on information entropy according to claim 1 is characterized in that: The implementation of step 2 includes the following steps: Step 2.1, Gaussian kernel function belongs to radial basis function, that is: K(x i ,x j )=exp(-γ||x i -x j || 2 ) where i,j∈[1,n], x i ,x j represents the eigenvector, K(x i ,x j ) represents the kernel function, γ∈(0,+∞) is the kernel function parameter of the support vector machine; Step 2.2: The training process of the first support vector machine and the second support vector machine is as follows: Solve the weight and bias of the separating hyperplane, seek the maximum margin, and introduce the Lagrange multiplier to express it as the following dual problem: Among them, ω represents the hyperplane weight vector, b represents the hyperplane bias, and y i ∈{-1,1} represents the manual calibration result corresponding to island detection, -1 represents non-island, 1 represents island, and the Lagrange multiplier ξ is introduced i , partial differentials can be used to solve the extreme value problem, the process is as follows: The final support vector machine solution is as follows: Where ω represents the hyperplane weight vector, b represents the hyperplane bias, i,j∈[1,n], x i ,x j represents the feature vector, y i ,y j ∈{-1,1} represents the manual calibration result corresponding to island detection, -1 represents non-island, 1 represents island, ξ i ,ξ j represents the Lagrange multiplier, l represents the total number of simulation cases; Relying on the radial basis kernel function, the decision function of the separation hyperplane can be expressed as: Among them, θ i is the coefficient of the i-th support vector, x i is the i-th support vector; Subsequent support vector classification relies on the trained decision function part for classification output.

4. The distributed power supply island detection method based on information entropy according to claim 1 is characterized in that: Step 3: The probability density functions of the classification accuracy of the first support vector machine and the second support vector machine are respectively denoted as ζ Γ (x) and ζ Ψ (x) represents the probability density function of the recognition accuracy of the first support vector machine and the second support vector machine during the island detection process over time, which is fitted by a polynomial equation and the fitting index does not exceed 4.

Images