Photovoltaic power station abnormity early warning method and system based on wavelet packet-BP neural network

Abstract

The invention relates to a photovoltaic power station abnormity early warning method and system based on a wavelet packet-BP neural network. The method comprises the following steps: step 1, abnormal signal feature extraction: extracting an abnormal signal of a current power system through wavelet analysis; 2, building a photovoltaic power station fault simulation model according to the related characteristics of the non-standard current; 3, calculating power quality power supply voltage deviation and frequency deviation; step 4, adopting a photovoltaic power station abnormity early warning method based on a BP neural network to realize photovoltaic power station abnormity early warning; step 5, constructing a photovoltaic system electric energy quality abnormity intelligent diagnosis monitoring system; 6, early fault notification and early warning are completed, and remote system state monitoring and control are achieved; the method has the advantages that output electric energy monitoring is achieved, the monitoring precision is high, and the electric energy quality is accurately predicted.

Description

Technical Field

[0001] The invention belongs to the technical field of photovoltaic power station early warning, and in particular relates to a photovoltaic power station abnormality early warning method and system based on wavelet packet-BP neural network. Background Art

[0002] Under the background of "dual carbon", the construction of photovoltaic power generation projects has been continuously promoted, and the installed capacity of photovoltaic power generation has repeatedly set new highs. With the promotion of distributed photovoltaic power generation, the installed capacity of distributed photovoltaic power generation has surpassed the installed capacity of photovoltaic power stations. Photovoltaic power station power generation has the characteristics of 1) being affected by environmental factors, and the power generation has the characteristics of random fluctuations; 2) the electricity generated by the photovoltaic array is connected to the grid through the inverter without rotational inertia; 3) the inverter in the power station usually has the typical characteristics of four-quadrant control and decoupling control capabilities. Therefore, after the photovoltaic power station is connected to the power grid, the structure of the distribution network is changed, affecting the power quality of the power grid, thereby affecting the power supply safety and electricity consumption of the power grid. The power safety of the equipment poses a potential threat; on the other hand, due to the increasing penetration rate of photovoltaic power generation in the power grid, large-scale photovoltaic power stations may cause serious power quality problems due to the uncertainty of light resources or the operating quality of the converter, affecting the normal operation of the public power grid and bringing related power quality problems. Therefore, it is necessary to monitor the power quality of photovoltaic power stations to ensure that the grid-connected current meets the relevant national power quality standards; therefore, it is very necessary to provide a photovoltaic power station abnormal warning method and system based on wavelet packet-BP neural network that can realize output power monitoring, high monitoring accuracy, and accurate prediction of power quality. Summary of the invention

[0003] The purpose of the present invention is to overcome the deficiencies of the prior art and to provide a photovoltaic power station abnormality warning method and system based on wavelet packet-BP neural network that can realize output power monitoring, high monitoring accuracy, and accurate prediction of power quality.

[0004] The object of the present invention is achieved by: a photovoltaic power station abnormality early warning method based on wavelet packet-BP neural network, the method comprising the following steps:

[0005] Step 1: Abnormal signal feature extraction: Extract the abnormal signal of the current power system through wavelet analysis;

[0006] Step 2: Establish a fault simulation model of the photovoltaic power station based on the relevant characteristics of the non-standard current;

[0007] Step 3: Calculation of power quality supply voltage deviation and frequency deviation;

[0008] Step 4: Use the photovoltaic power station abnormality warning method based on BP neural network to realize photovoltaic power station abnormality warning;

[0009] Step 5: Construct an intelligent diagnosis and monitoring system for abnormal power quality of photovoltaic systems;

[0010] Step 6: Complete early fault notification and warning to achieve remote system status monitoring and control.

[0011] The step 1 of extracting abnormal signals of the current power system by wavelet analysis specifically includes the following steps:

[0012] Step 1.1: Power system singular signals;

[0013] Step 1.2: Singularity detection of fault transient signal.

[0014] The power system singular signal in step 1.1 specifically includes the following steps:

[0015] Step 1.11: Basic definition of singular features: In order to clarify the singularity of the signal, the signal is divided into slowly changing and suddenly changing singular signals, and the following definitions are derived: ① Suppose function f:[a,b]→R,x0∈[a,b], let α0=SUP{a,f is Lipschitz a at x0}, the Lipschitz singularity of f at x0 is called α0; ② Suppose function f:[a,b]→R at x0∈[a,b] is a singular signal, if α0<0, f at x0 is called a suddenly changing singular signal; if α0>0, f at x0 is called a slowly changing signal; From the above definition, we can get: if function f:[a,b]→R at x0∈[a,b] has an nth-order derivative, then the singularity of this point α0≥n; if the nth-order derivative does not exist, then the singularity of this point is α0<n;

[0016] Step 1.12: When a fault occurs in the power system, the fault signal components are very complex. In the actual protection fault signal analysis, the time segment of the signal used is very short, and it is generally described by the following fault signal model: Where ω is the fundamental frequency of the power system; Ae -λt is the decaying DC component; It is the fundamental frequency and high-order harmonic components; from the analysis of the circuit equivalent principle, it can be seen that the transient signal of the power system fault is continuous in nature, and the fault signal model is continuous at the fault time t=0, so the fault transient signal is a continuous signal. If the signal is singular, it is a slowly varying singular signal.

[0017] The singularity detection of the fault transient signal in step 1.2 specifically includes the following steps:

[0018] Step 1.21: Singularity detection of fault transient signal Definition: In order to detect the singularity of slowly varying singular signals, a wavelet function with compact support and vanishing moments of sufficient order is required. Its definition is given as follows: The wavelet Ψ(x) is called a wavelet with vanishing moments of order , such as for all positive integers k, 0≤k≤n: The method for determining the singular points of slowly varying singular signals is as follows: Assume that the wavelet Ψ(x) has compact support and n-order vanishing moments, and is continuously differentiable n times, where n is a positive integer. If the Lipschitz degree of the function f at x0 is α0 (α0 < 0), and it is continuously differentiable n times near the point x0, then |W f (x,s) reaches a maximum value at x0, which can accurately detect the position of the singular point of the slowly varying singular signal;

[0019] Step 1.22: Singularity detection algorithm of fault transient signal: The detection algorithm of the singular point position of the slowly varying singular signal is summarized as follows: ① Perform discrete wavelet transform on f using wavelet, and use Mallat algorithm to obtain {W(x j ,s,Ψ)j=1,2,...,N};②For data {W(x j ,s,Ψ)j=1,2,...,N}, and record the corresponding maximum point, then the maximum point is the singular point of the slowly varying singular signal; the wavelet transform in the algorithm is defined by the following formula: For any function f(t)∈l 2 (R), its wavelet transform is:

[0020] In step 2, a photovoltaic power station fault simulation model is established according to the non-standard current related characteristics, specifically: the fault signal model is set as: It can be verified that the derivative of the signal at point t=0 does not exist, and the singularity is less than 1, so a wavelet with a vanishing moment of 1 corresponding to the 4th-order transfer function is selected for singularity detection; the specific steps of detecting the moment of fault occurrence include:

[0021] Step 2.1: The sampling sequence is given by the fault signal. The number of sampling points per week is 320. Half a week is taken before and after the fault, that is, 160 points each;

[0022] Step 2.2: Use FFT to analyze the fault signal and obtain the decomposition sequence;

[0023] Step 2.3: Use the wavelet corresponding to the selected 4th-order transfer function and the two-scale sequence of the scaling function {g n},{h n}, implement the Mallat algorithm of wavelet transform to obtain the decomposition sequence of fault signal;

[0024] Step 2.4: Observe the decomposed information sequence of the fault signal sampling sequence to determine whether there is a maximum value, that is, to determine whether there is a fault, thereby determining the location of the fault point;

[0025] Step 2.5: Draw the fault signal detection graph based on the calculated data.

[0026] The calculation of power quality supply voltage deviation and frequency deviation in step 3 is as follows: voltage deviation refers to the difference between the actual voltage of a node in the system and the rated voltage of the system under normal working conditions; frequency deviation refers to the difference between the actual frequency of the power system operation and the rated frequency of the system; the calculation of voltage deviation is: assuming that the voltage deviation on the power supply bus is δU A %, the voltage loss of high-voltage line l1 is The voltage deviation caused by the transformer is δU T %, the voltage loss of low voltage line l2 is Then the voltage deviations at points B, C, and D are: Extending the above concept to any power supply system, if there are multiple levels and voltages or voltage regulating equipment from the power supply to a specified point, the voltage deviation at the specified point can be calculated by the following formula: δU E % = ∑δU% - ∑ΔU%, where ∑δU% is the sum of all voltage deviations from the power supply to the specified point; ∑ΔU% is the sum of all voltage losses from the power supply to the specified point.

[0027] In step 4, a photovoltaic power station abnormality warning method based on BP neural network is used to implement photovoltaic power station abnormality warning, including photovoltaic power station island detection and photovoltaic power station self-diagnosis monitoring; wherein photovoltaic power station island detection is implemented by a method based on wavelet transform and BP neural network, specifically including the following steps:

[0028] Step 4.1: The central controller of the photovoltaic power station collects the current i flowing from the inverter power supply into the common coupling point PCC PCC and the voltage at the common coupling point V PCC , and then use wavelet transform to transform the collected current signal i PCC And the voltage signal V PCC Processing is performed to obtain the corresponding wavelet coefficients;

[0029] Step 4.2: Process the obtained wavelet coefficients by algorithm to obtain a feature vector reflecting the characteristics of the collected signal;

[0030] Step 4.3: These feature vectors are provided to the BP neural network, and the trained BP neural network performs pattern recognition and thereby determines the operating status of the photovoltaic power station;

[0031] Step 4.4: When the BP neural network detects that the photovoltaic power station is in an island operation state, the central controller takes island protection measures to stop the photovoltaic power generation system from supplying power to the local load. When the photovoltaic power station is not in an island operation state, the island protection does not work.

[0032] The photovoltaic power station self-diagnosis monitoring in step 4 adopts the photovoltaic power station self-diagnosis monitoring based on BP neural network, specifically: the BP neural network is divided into three layers of input layer, hidden layer and output layer. Suppose the fault input mode x of the network is (T c ,I m ,U m ), the fault output of the hidden layer y=(y1,y2,y3); the expected output values ​​of various fault types: normal (0,0,0), panel cracking (1,0,0), panel aging short circuit (0,1,0), local shadow covering (0,0,1), the fault input mode function of the BP network X=(T c ,I m ,U m ) T , the fault output mode function of the hidden layer y=(y1,y2,y3) T , in order to build the mapping relationship between the input and output of the neural network, in the process of transmitting from the input layer to the hidden layer of the BP network, the output value formula of each neuron in the hidden layer is b i ={exp[-(∑a i ·W ij -θ j )r]+1} -1 , where a i Represents the input value of each neuron in the input layer; W ij Represents the connection weight from the input layer to the hidden layer; θ j represents the hidden layer threshold; r represents the correction coefficient of the S-type function; in the process of transmitting from the hidden layer to the output layer of the BP network, the output value formula of each neuron in the output layer is c i ={exp[-(∑b j ·V ji -γ j )r]+1} -1 , where b j Represents the input value of each neuron in the hidden layer; V ji represents the connection weight from the hidden layer to the output layer; γ j represents the threshold of the output layer; r represents the correction coefficient of the S-type function.

[0033] The photovoltaic power station abnormality warning system based on wavelet packet-BP neural network includes a photovoltaic power station real-time monitoring system, which includes a single-chip microcomputer, a data acquisition module, a monitoring module, a wireless communication module, a local SD card and a cloud platform; the photovoltaic power station abnormality warning system based on wavelet packet-BP neural network is used for the above-mentioned photovoltaic power station abnormality warning method based on wavelet packet-BP neural network.

[0034] The monitoring module collects and monitors the power quality data information of the photovoltaic power station in real time; the single-chip microcomputer effectively manages the large-scale power quality data collected by the monitoring module in real time, controls the monitoring module to realize uninterrupted monitoring of the power station, and sends the data to the cloud platform when the conditions for data transmission are met; the local SD card is used to store the data information collected by the monitoring module in real time; the communication module is used to perform remote data transmission to realize remote real-time monitoring of power-related data of the photovoltaic power station; the cloud platform is used to receive relevant data information of the single-chip microcomputer to form a complete monitoring system.

[0035] Beneficial effects of the present invention: The present invention is a photovoltaic power station abnormal warning method and system based on wavelet packet-BP neural network. In use, the present invention uses wavelet analysis to process the original model, has the local analysis characteristics of variable two windows in the time and frequency domain, can analyze and amplify the signal details on any frequency band, and make the system monitoring accuracy higher. The present invention deeply extracts the characteristics of abnormal electric energy signals through wavelet analysis, and can provide more information compared with other monitoring methods. Deeply restore the real situation of the power quality fluctuation of the photovoltaic system; the present invention is based on a neural network, has autonomous learning ability and adaptive ability, and has a strong information integration ability, can process quantitative and qualitative information at the same time, can coordinate a variety of input information relationships well, has higher portability and stability, and as the system runs, the more data, the higher the efficiency of the system, and the stability will increase accordingly. The present invention applies the BP neural network to the prediction of power quality. When the amount of data is sufficient, the accurate prediction of power quality can be achieved, and the control accuracy and stability of the entire photovoltaic power generation system can be prompted; the present invention has the advantages of realizing output power monitoring, high monitoring accuracy, and accurate prediction of power quality. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 It is a schematic diagram of FFT transformation and wavelet transformation of the fault transient signal of the present invention.

[0037] Figure 2 It is a schematic diagram of calculating voltage deviation of the power supply system of the present invention.

[0038] Figure 3 It is a schematic diagram of voltage deviation adjustment of the present invention.

[0039] Figure 4 This is a schematic diagram of the island detection principle of the present invention.

[0040] Figure 5 Schematic diagram of the four-layer wavelet decomposition tree of the present invention.

[0041] Figure 6 It is a schematic diagram of the BP neural network topology structure of the present invention.

[0042] Figure 7 It is a schematic diagram of the basic structure of the BP neural network of the present invention.

[0043] Figure 8 It is a flow chart of BP neural network fault diagnosis of the present invention.

[0044] Fig. 9 The system structure of the present invention is shown in FIG. Figure 1 .

[0045] Fig.10 The system structure of the present invention is shown in FIG. Figure 2 . DETAILED DESCRIPTION

[0046] The present invention will be further described below in conjunction with the accompanying drawings.

[0047] Example 1

[0048] like Figure 1-10 As shown, a photovoltaic power station abnormality early warning method based on wavelet packet-BP neural network includes the following steps:

[0049] Step 1: Abnormal signal feature extraction: Since the inverter output voltage is connected to the grid through a transformer, the fault information contained in it is offset by the grid voltage. Using the three-phase grid-connected current waveform output by the inverter to extract effective abnormal feature information is the core of solving the problem of abnormal warning of power stations. Therefore, wavelet analysis is used to extract abnormal signals of the current power system;

[0050] Step 2: Establish a fault simulation model of the photovoltaic power station based on the relevant characteristics of the non-standard current;

[0051] Step 3: Calculation of power quality supply voltage deviation and frequency deviation;

[0052] Step 4: Use the photovoltaic power station abnormality warning method based on BP neural network to realize photovoltaic power station abnormality warning;

[0053] Step 5: Construct an intelligent diagnosis and monitoring system for abnormal power quality of photovoltaic systems;

[0054] Step 6: Complete early fault notification and warning to achieve remote system status monitoring and control.

[0055] The wavelet analysis in step 1 is used to extract the abnormal signal of the current power system, which specifically includes the following steps:

[0056] Step 1.1: Power system singular signals;

[0057] In this embodiment, the power system singular signal in step 1.1 specifically includes the following steps:

[0058] Step 1.11: Basic definition of singular features: In order to clarify the singularity of the signal, the signal is divided into slow-changing and sudden-changing singular signals, and the following definition is derived: ① Assume function f:[a,b]→R,x0∈[a,b], let α0=SUP{a,f is Lipschitz at x0 a} The Lipschitz singularity of f at x0 is called α0; ② Assume that the function f:[a,b]→R at x0∈[a,b] is a singular signal. If α0<0, f at x0 is called a sudden change singular signal; if α0>0, f at x0 is called a slow change signal; From the above definition, we can get: if the function f:[a,b]→R at x0∈[a,b] has an n-order derivative, then the singularity of this point is α0≥n; if the n-order derivative does not exist, then the singularity of this point is α0<n; For example: the singularity of f(t)=(t-t0) at t0 is 1, which is a slow change singular signal; and the singularity of f(t)=δ(t-t0) at t0 is -1, which is a sudden change singular signal;

[0059] Step 1.12: When a fault occurs in the power system, the fault signal components are very complex. In the actual protection fault signal analysis, the time segment of the signal used is very short, and it is generally described by the following fault signal model: Where ω is the fundamental frequency of the power system; Ae -λt is the decaying DC component; It is the fundamental frequency and high-order harmonic components; from the analysis of the circuit equivalent principle, it can be known that the transient signal of the power system fault is continuous in nature, and the fault signal model is continuous at the fault time (reference point) t=0, so the fault transient signal is a continuous signal. If the signal has singularity, it is a slowly varying singular signal;

[0060] By examining the derivative of the fault transient signal at t = 0, we can calculate the left derivative of the signal f-(0) = aωcosψ, and the right derivative is When the left derivative is not equal to the right derivative, the fault transient signal is not differentiable at t = 0, and the fault transient signal has singularity, with a singularity of α0≤1; when t = 0, the left derivative is equal to the right derivative, that is, When , the fault signal is differentiable, and the singularity α0≥1; at this time, the derivative function of f(t) is: Similarly, after making similar left and right derivative calculations on f(t), it can be obtained that f″(t) of the fault transient signal at t=0 is generally not true, but it may be true under certain conditions and cannot be completely ruled out; similarly, the 2N+2nd order derivative of f(t) at point t=0 generally does not exist, but it still has the possibility of existence; therefore, the singularity of the fault transient signal at the moment of fault is uncertain, and the singularity of different fault transient signals is different; in summary, the fault transient signal of the power system is a slowly varying singular signal, and its singularity is uncertain.

[0061] Step 1.2: Singularity detection of fault transient signal.

[0062] In this embodiment, the singularity detection of the fault transient signal in step 1.2 specifically includes the following steps:

[0063] Step 1.21: Singularity detection of fault transient signal Definition: In order to detect the singularity of slowly varying singular signals, a wavelet function with compact support and vanishing moments of sufficient order is required. Its definition is given as follows: The wavelet Ψ(x) is called a wavelet with vanishing moments of order , such as for all positive integers k, 0≤k≤n: If the wavelet has compact support and n-th order vanishing moment, and is n-times continuously differentiable, n is a positive integer; let the function and a<n, then f is uniformly Lipschitz a on [a,b] if and only if: for any ξ>0 there exists Aξ>0 such that for any x∈[a+ξ,b-ξ] and s>0, there is always: |W f (x,s)|≤A ξ s a (4) Assume that the wavelet Ψ(x) has compact support and n-th order vanishing moments and is continuously differentiable n times, n is a positive integer, and if f is Lipschitza at x0, then there exists A>0 such that: |W f (x,s)|≤A(s a +|x-x0| a )(5), there exists B>0, such that holds true, then f is Lipschitz a at x0; from this, the method for determining the singular points of slowly varying singular signals is as follows: Assume that the wavelet Ψ(x) has compact support and n-th order vanishing moments, and is continuously differentiable n times, where n is a positive integer. If the Lipschitz degree of the function f at x0 is α0 (α0 < 0), and it is continuously differentiable n times near the point x0, then |W f (x,s)| reaches a maximum value at x0, which can accurately detect the position of the singular point of the slowly varying singular signal;

[0064] Step 1.22: Singularity detection algorithm of fault transient signal: The detection algorithm of the singular point position of the slowly varying singular signal is summarized as follows: ① Perform discrete wavelet transform on f using wavelet, and use Mallat algorithm to obtain {W(x j ,s,Ψ)|j=1,2,...,N};②For data {W(x j ,s,Ψ)|j=1,2,...,N}, and record the corresponding maximum point, then the maximum point is the singular point of the slowly varying singular signal; the wavelet transform in the algorithm is defined by the following formula: For any function f(t)∈l 2 (R), its wavelet transform is:

[0065] In step 2, a photovoltaic power station fault simulation model is established according to the non-standard current related characteristics, specifically: the fault signal model is set as: It can be verified that the derivative of the signal at point t=0 does not exist, and the singularity is less than 1, so a wavelet with a vanishing moment of 1 corresponding to the 4th-order transfer function is selected for singularity detection; the specific steps of detecting the moment of fault occurrence include:

[0066] Step 2.1: The sampling sequence is given by the fault signal. The number of sampling points per week is 320. Half a week is taken before and after the fault, that is, 160 points each;

[0067] Step 2.2: Use FFT to analyze the fault signal and obtain the decomposition sequence;

[0068] Step 2.3: Use the wavelet corresponding to the selected 4th-order transfer function and the two-scale sequence of the scaling function {g n},{h n}, implement the Mallat algorithm of wavelet transform to obtain the decomposition sequence of fault signal;

[0069] Step 2.4: Observe the decomposed information sequence of the fault signal sampling sequence to determine whether there is a maximum value, that is, to determine whether there is a fault, thereby determining the location of the fault point;

[0070] Step 2.5: Draw the fault signal detection graph based on the calculated data.

[0071] In this embodiment, the fault signal detection graph is as follows: Figure 1 As shown, it is obvious that the signal has singularity at t = 0 (160 points) (because the wavelet transform adopts the two-decimation method, so Figure 1 According to the above method, the singular point is detected, but the fault moment cannot be extracted by FFT; it should be noted that the obtained singular point is fixedly delayed by one sampling interval from the real singular point.

[0072] The calculation of the power quality supply voltage deviation and frequency deviation in step 3 is specifically as follows: voltage deviation refers to the difference between the actual voltage of a node in the system and the rated voltage of the system under normal working conditions; frequency deviation refers to the difference between the actual frequency of the power system operation and the rated frequency of the system; the calculation of voltage deviation is as follows: Figure 1 As shown, let the voltage deviation on the power supply bus be δU A %, the voltage loss of high-voltage line l1 is The voltage deviation caused by the transformer is δU T %, the voltage loss of low voltage line l2 is ΔU l2 %, then the voltage deviations of points B, C, and D are: Extending the above concept to any power supply system, if there are multiple levels and voltages or voltage regulating equipment from the power supply to a specified point, the voltage deviation at the specified point can be calculated by the following formula: δU E % = ∑δU% - ∑ΔU%, where ∑δU% is the sum of all voltage deviations from the power supply to the specified point; ∑ΔU% is the sum of all voltage losses from the power supply to the specified point; the frequency deviation is calculated using one of the classical system model, complex speed regulator and prime mover model, simplified speed regulator and prime mover model, and minimum frequency model.

[0073] In this embodiment, the voltage loss of the power supply system is: For high-voltage power supply systems, the system equivalent reactance is much larger than the system equivalent resistance. Therefore, ignoring the resistance, the voltage loss of the power supply system is: This formula shows that the main factors affecting voltage quality are: ① load reactive power or reactive power change; ② grid short-circuit capacity or grid equivalent reactance; voltage deviation refers to the degree to which the grid voltage deviates from the rated voltage of the grid. Changes in system operation mode or changes in user load will cause the actual voltage at a certain point on the grid to deviate from its rated voltage. We define the difference between the actual voltage and the rated voltage as a percentage of the rated voltage as voltage deviation: Among them, δU% is the voltage deviation percentage of a certain point on the power grid; U is the actual voltage at that point; U N is the rated voltage of the grid.

[0074] In step 4, a photovoltaic power station abnormality warning method based on BP neural network is used to implement photovoltaic power station abnormality warning, including photovoltaic power station island detection and photovoltaic power station self-diagnosis monitoring; wherein photovoltaic power station island detection is implemented by a method based on wavelet transform and BP neural network, specifically including the following steps:

[0075] Step 4.1: Figure 4 As shown, first, the central controller of the photovoltaic power station collects the current i flowing from the inverter power supply into the common coupling point PCC PCC and the voltage at the common coupling point V PCC , and then use wavelet transform to transform the collected current signal i PCC And the voltage signal V PCC Processing is performed to obtain the corresponding wavelet coefficients;

[0076] Step 4.2: Secondly, the obtained wavelet coefficients are processed by algorithm to obtain the feature vector reflecting the characteristics of the collected signal;

[0077] In this embodiment, the present invention selects db4 wavelet as the wavelet basis to perform the wavelet transform required for algorithm processing, and uses the average value of the absolute value of the wavelet coefficients to construct the characteristic vector of the signal, which can accurately reflect the change characteristics of the island signal. The present invention first calculates the average value of the absolute value of the 7 layers of wavelet coefficients obtained in the non-island and island states, and obtains 14 characteristic quantities related to the voltage and current signals, and compares them. Among them, the characteristic quantities constructed by the detail components of the lowest 4 layers of wavelet coefficients have a large difference before and after in the island and non-island states, which is conducive to the neural network to perform pattern recognition of the island state; for this reason, the characteristic quantities of the lowest 4 layers of voltage and current signals are selected to construct the characteristic vector of the signal, and it is used as the input signal of the BP neural network input layer; after the collected voltage and current signals are subjected to 4-layer wavelet decomposition, the high-frequency components D,-D of each layer are selected for further processing, such as Figure 5 As shown;

[0078] The present invention adopts the following method to construct the characteristic vector of the signal: First, the collected voltage V PCC and current i PCC Four layers of wavelet decomposition are performed respectively to obtain the wavelet coefficients of the high-frequency signal components of each layer of the two signals; since the signal sampling frequency is set to 10kHz, 200 wavelet coefficients can be obtained in one voltage cycle (0.02s); secondly, the absolute values ​​of the 200 wavelet coefficients of each layer are calculated respectively, and then all these values ​​are added and divided by 200, so as to obtain the characteristic quantities of each layer of voltage and current signals after wavelet transformation; among them, the voltage V PCC The corresponding characteristic quantity is u TZ1 、u TZ2 、u TZ3 、u TZ4 represents, and the current i PCC The corresponding feature quantity is i TZ1 、i TZ2 、i TZ3 、i TZ4 Indicates that, therefore, there are a total of 8 feature quantities that constitute the feature vector required by the neural network.

[0079] Step 4.3: Finally, these feature vectors are provided to the BP neural network, and the trained BP neural network performs pattern recognition and thus determines the operating status of the photovoltaic power station;

[0080] In this embodiment, the present invention collects i PCC and V PCC These two signals are used as feature vectors to determine whether an island has appeared. Under each load condition, at least one signal can be guaranteed to have a significant change. The typical topological structure of the BP neural network includes a total of three layers of networks: input layer, hidden layer and output layer. The Mallat algorithm based on error gradient descent is used to train the weights and thresholds of each layer of the network.

[0081] The feature vector constructed by the present invention includes 8 signal feature quantities, so the number of neuron nodes of the input layer network is designed to be 8; the number of nodes of the BP network output layer depends on the number of recognition states, and the neural network in the present invention only needs to recognize and classify two operating states of the photovoltaic power station, namely, non-islanding and islanding, so the number of neuron nodes of the output layer network is designed to be 1; if the number of nodes of the hidden layer network is too large, the amount of calculation will increase and the learning time of the network will be prolonged; if the number of nodes is too small, the ability to recognize untrained samples will be poor or even errors will occur; the number of hidden layer nodes is usually designed according to the following empirical formula: Where, l is the number of hidden layer nodes; m is the number of input nodes; n is the number of output nodes; α is a constant between 1 and 10; in the present invention, when the number of hidden layer network nodes is 12, the convergence speed and classification effect can reach the best, so the number of neuron nodes in the hidden layer network is designed to be 12; the BP neural network topology is as follows Figure 6 As shown in the figure, w and v are the weights between neuron nodes, and the weights are trained by the reverse error propagation algorithm; the input vectors x1~x8 are the aforementioned voltage and current characteristic quantities; and the output vector y1 is the island recognition state quantity.

[0082] Step 4.4: When the BP neural network detects that the photovoltaic power station is in an island operation state, the central controller takes island protection measures to stop the photovoltaic power generation system from supplying power to the local load. When the photovoltaic power station is not in an island operation state, the island protection does not work.

[0083] The photovoltaic power station self-diagnosis monitoring in step 4 adopts the photovoltaic power station self-diagnosis monitoring based on BP neural network, specifically: the BP neural network is divided into three layers of input layer, hidden layer and output layer. Suppose the fault input mode x of the network is (T c ,I m ,U m ), the fault output of the hidden layer y = (y1, y2, y3); the expected output values ​​of various fault types: normal (0, 0, 0), battery panel cracking (1, 0, 0), battery panel aging short circuit (0, 1, 0), local shadow covering (0, 0, 1), the basic structure of the BP neural network is as follows Figure 7 As shown, the fault input mode function of BP network is X = (T c ,I m ,U m ) T , the fault output mode function of the hidden layer y=(y1,y2,y3) T , in order to build the mapping relationship between the input and output of the neural network, in the process of transmitting from the input layer to the hidden layer of the BP network, the output value formula of each neuron in the hidden layer is b i ={exp[-(∑a i ·Wij -θ j )r]+1} -1 , where a i Represents the input value of each neuron in the input layer; W ij Represents the connection weight from the input layer to the hidden layer; θ j represents the hidden layer threshold; r represents the correction coefficient of the S-type function; in the process of transmitting from the hidden layer to the output layer of the BP network, the output value formula of each neuron in the output layer is c i ={exp[-(Σb j ·V ji -γ j )r]+1} -1 , where b j Represents the input value of each neuron in the hidden layer; V ji represents the connection weight from the hidden layer to the output layer; γ j represents the threshold of the output layer; r represents the correction coefficient of the S-type function; the program of the BP algorithm is as follows Figure 8 shown.

[0084] In this embodiment, step 5 constructs an intelligent diagnosis and monitoring system for abnormal power quality of photovoltaic systems and step 6 completes early fault notification and warning to achieve remote system status monitoring and control, specifically: the information detected by the intelligent diagnosis system for abnormal power quality of photovoltaic systems is transmitted to the background, and an evaluation system of an expert library is established using existing photovoltaic power quality monitoring methods and corresponding solution information; an intelligent diagnosis system for abnormal status of transmission lines is built, and the power quality of photovoltaic system power generation is evaluated based on the real-time transmitted data of online monitoring and the evaluation system of the expert library, and corresponding strategies are given.

[0085] The present invention is a photovoltaic power station abnormal warning method and system based on wavelet packet-BP neural network. In use, the present invention uses wavelet analysis to process the original model, has the local analysis characteristics of variable two windows in the time and frequency domain, can analyze and amplify the signal details on any frequency band, and make the system monitoring accuracy higher. The present invention deeply extracts the characteristics of abnormal electric energy signals through wavelet analysis, and can provide more information compared with other monitoring methods. Deeply restore the real situation of the power quality fluctuation of the photovoltaic system; the present invention is based on a neural network, has autonomous learning ability and adaptive ability, and has a strong information integration ability, can process quantitative and qualitative information at the same time, can coordinate a variety of input information relationships well, has higher portability and stability, and as the system runs, the more data, the higher the efficiency of the system, and the stability will increase accordingly. The present invention applies the BP neural network to the prediction of power quality. When the amount of data is sufficient, the accurate prediction of power quality can be achieved, and the control accuracy and stability of the entire photovoltaic power generation system can be prompted; the present invention has the advantages of realizing output power monitoring, high monitoring accuracy, and accurate prediction of power quality.

[0086] Example 2

[0087] like Figure 1-10 As shown, the photovoltaic power station abnormality warning system based on wavelet packet-BP neural network includes a photovoltaic power station real-time monitoring system, and the photovoltaic power station real-time monitoring system includes a single-chip microcomputer, a data acquisition module, a monitoring module, a wireless communication module, a local SD card and a cloud platform; the photovoltaic power station abnormality warning system based on wavelet packet-BP neural network is used for the above-mentioned photovoltaic power station abnormality warning method based on wavelet packet-BP neural network.

[0088] In this embodiment, the photovoltaic power station abnormality warning system based on wavelet packet-BP neural network of the present invention can: ① quickly and accurately detect, determine and record the fluctuation of photovoltaic power quality; ② quickly classify and process the current power quality abnormality type; ③ effectively monitor the fluctuation abnormality of the entire system; ④ predict the possibility of power quality abnormality in the future and issue a warning in advance; ⑤ improve the stability of the entire photovoltaic power generation control system, and facilitate personnel to inspect and monitor the photovoltaic power generation system; ⑥ send the monitoring information to the dispatching center through remote communication; ⑦ continuously collect information to eliminate the transient signal "recording dead zone"; ⑧ reliable operation, stable performance, simple installation and disassembly, and little impact on the photovoltaic power generation system; ⑨ can help to timely monitor the status of the entire photovoltaic power station, ensure the output power quality, reduce the economic losses caused by power quality factors, and can greatly save the workload of the staff, achieve real-time monitoring, early warning, and can be technically It can ensure the safe, stable and economical operation of photovoltaic power stations; ⑩ It has good economic and social benefits: Economic benefits: It can greatly improve the detection accuracy of abnormal power monitoring in photovoltaic power stations, can effectively reduce manual workload, reduce labor cost investment, and at the same time help to improve the safety and stability of photovoltaic power station systems and the ability to output high-quality electricity. At the same time, it can significantly reduce economic losses and electrical accidents caused by power quality fluctuations. The economic benefits are very significant; Social benefits: It can provide a new, efficient and highly accurate real-time monitoring solution for abnormal power quality monitoring in photovoltaic power stations, which can greatly improve the quality of abnormal power quality monitoring in photovoltaic power stations, and bring more guarantees for the normal operation of photovoltaic power stations, thereby ensuring residents' satisfaction with electricity use. It is of great significance to improving the service quality and image of power supply companies and ensuring the normal and stable operation of the entire power system. The social benefits are very significant.

[0089] The monitoring module collects and monitors the power quality data information of the photovoltaic power station in real time; the single-chip microcomputer effectively manages the large-scale power quality data collected by the monitoring module in real time, controls the monitoring module to realize uninterrupted monitoring of the power station, and sends the data to the cloud platform when the conditions for data transmission are met; the local SD card is used to store the data information collected by the monitoring module in real time; the communication module is used to perform remote data transmission to realize remote real-time monitoring of power-related data of the photovoltaic power station; the cloud platform is used to receive relevant data information of the single-chip microcomputer to form a complete monitoring system.

[0090] In this embodiment, the data acquisition module includes parameter acquisition sensors and terminal nodes to realize the acquisition of various parameters of the photovoltaic power station, including temperature sensors, light intensity sensors, etc.; the wireless communication module includes a router grounding and a ZigBee wireless module, and the ZigBee wireless network module sends the parameters collected by the ZigBee terminal node to the LabVIEW monitoring platform through the ZigBee coordinator node; the monitoring module includes a coordinator node and a LabVIEW monitoring platform, and the LabVIEW monitoring module displays the parameter information collected by the serial port in a digital or graphical form on the monitoring interface, which is convenient for management personnel to monitor and query historical status information;

[0091] The photovoltaic power station real-time monitoring system also includes a JN5139 wireless microprocessor module, a peripheral application expansion module and a power supply module; the parameter acquisition sensor includes an SHT11 temperature sensor, a TSL2550 light intensity sensor and reserved interfaces for other parameter monitoring sensors.

[0092] In this embodiment, the system of the present invention has the following advantages: 1. Realize unattended photovoltaic substation: By adopting the real-time monitoring system of photovoltaic power station, operation and maintenance personnel can understand the real-time operation status of photovoltaic power station through remote monitoring, especially for photovoltaic power stations with relatively harsh environment that are not suitable for long-term deployment of technical personnel; 2. Improve work efficiency and reduce labor costs: Distributed photovoltaic system sites are mostly scattered, and photovoltaic power station manufacturers are diverse. Although the functions are generally the same, the models are different. By adopting the real-time monitoring system of photovoltaic power station, it is convenient to realize distributed monitoring and centralized management of distributed photovoltaic power station, thereby improving work efficiency; 3. Assist in finding fault points and improving maintenance efficiency: By Collect equipment and system operation data, connect fault signals to the real-time monitoring system of the photovoltaic power station, and assist in finding the fault point, which helps maintenance personnel to grasp the operation status of the entire photovoltaic power station in real time and carry out fault maintenance in time; ④ Online monitoring to improve the operation reliability of distributed photovoltaic power stations: The photovoltaic power station data collected by the photovoltaic power station monitoring system is real-time and accurate, and the data collection process is information-based and intelligent, which can feedback alarm information in time. Operation and maintenance personnel can check information at any time, and remotely operate and control, and report faults in time. Manufacturers can check the database of the monitoring system to find equipment defects or even family defects and repair them in time, jointly ensuring the reliability of power station operation;

[0093] In summary, the present invention monitors the output power of the photovoltaic power station, promptly identifies power quality abnormalities, and locates the causes of the abnormalities, which is of great significance for reducing the operation and maintenance costs of the photovoltaic power station and improving the operational reliability of the photovoltaic power station.

[0094] The present invention is a photovoltaic power station abnormality warning method and system based on wavelet packet-BP neural network. In use, the present invention uses wavelet analysis to process the original model, has the local analysis characteristics of two windows that are variable in time and frequency domains, can analyze and amplify the extraction of signal details on any frequency band, and make the system's monitoring accuracy higher. The present invention deeply extracts the characteristics of abnormal electric energy signals through wavelet analysis, and can provide more information compared with other monitoring methods. Deeply restore the true situation of power quality fluctuations in photovoltaic systems; the present invention is based on neural networks, has autonomous learning and adaptive capabilities, and has strong information integration capabilities, can process quantitative and qualitative information at the same time, can well coordinate the relationship between multiple input information, has higher portability and stability, and as the system runs, the more data, the higher the efficiency of the system will be, and the stability will increase accordingly. The present invention applies BP neural networks to the prediction of power quality. When the amount of data is sufficient, accurate prediction of power quality can be achieved, which can prompt the control accuracy and stability of the entire photovoltaic power generation system; the present invention provides a new power quality monitoring and prediction system to improve the reliability of the entire photovoltaic system application, and facilitate staff to timely monitor and predict the actual situation of the current photovoltaic system; the present invention has the advantages of realizing output power monitoring, high monitoring accuracy, and accurate prediction of power quality.

Claims

1. A photovoltaic power station abnormality early warning method based on wavelet packet-BP neural network, characterized by: The method comprises the following steps: Step 1: Abnormal signal feature extraction: Extract the abnormal signal of the current power system through wavelet analysis; Step 2: Establish a fault simulation model of the photovoltaic power station based on the relevant characteristics of the non-standard current; Step 3: Calculation of power quality supply voltage deviation and frequency deviation; Step 4: Use the photovoltaic power station abnormality warning method based on BP neural network to realize photovoltaic power station abnormality warning; Step 5: Construct an intelligent diagnosis and monitoring system for abnormal power quality of photovoltaic systems; Step 6: Complete early fault notification and warning to achieve remote system status monitoring and control.

2. The abnormal early warning method for photovoltaic power station based on wavelet packet-BP neural network according to claim 1, characterized in that: The step 1 of extracting abnormal signals of the current power system by wavelet analysis specifically includes the following steps: Step 1.1: Power system singular signals; Step 1.2: Singularity detection of fault transient signal.

3. The abnormal early warning method for photovoltaic power station based on wavelet packet-BP neural network as claimed in claim 2 is characterized by: The power system singular signal in step 1.1 specifically includes the following steps: Step 1.11: Basic definition of singular features: In order to clarify the singularity of the signal, the signal is divided into slowly changing and suddenly changing singular signals, and the following definitions are derived: ① Suppose function f:[a,b]→R,x0∈[a,b], let α0=SUP{a,f is Lipschitz a at x0}, the Lipschitz singularity of f at x0 is called α0; ② Suppose function f:[a,b]→R at x0∈[a,b] is a singular signal, if α0<0, f at x0 is called a suddenly changing singular signal; if α0>0, f at x0 is called a slowly changing signal; From the above definition, we can get: if function f:[a,b]→R at x0∈[a,b] has an nth-order derivative, then the singularity of this point α0≥n; if the nth-order derivative does not exist, then the singularity of this point is α0<n; Step 1.12: When a fault occurs in the power system, the fault signal components are very complex. In the actual protection fault signal analysis, the time segment of the signal used is very short, and it is generally described by the following fault signal model: Where ω is the fundamental frequency of the power system; Ae -λt is the decaying DC component; It is the fundamental frequency and high-order harmonic components; from the analysis of the circuit equivalent principle, it can be seen that the transient signal of the power system fault is continuous in nature, and the fault signal model is continuous at the fault time t=0, so the fault transient signal is a continuous signal. If the signal is singular, it is a slowly varying singular signal.

4. The abnormal early warning method for photovoltaic power station based on wavelet packet-BP neural network as claimed in claim 2 is characterized by: The singularity detection of the fault transient signal in step 1.2 specifically includes the following steps: Step 1.21: Singularity detection of fault transient signal Definition: In order to detect the singularity of slowly varying singular signals, a wavelet function with compact support and vanishing moments of sufficient order is required. Its definition is given as follows: The wavelet Ψ(x) is called a wavelet with vanishing moments of order , such as for all positive integers k, 0≤k≤n: The method for determining the singular points of slowly varying singular signals is as follows: Assume that the wavelet Ψ(x) has compact support and n-order vanishing moments, and is continuously differentiable n times, where n is a positive integer. If the Lipschitz degree of the function f at x0 is α0 (α0 < 0), and it is continuously differentiable n times near the point x0, then |W f (x,s) reaches a maximum value at x0, which can accurately detect the position of the singular point of the slowly varying singular signal; Step 1.22: Singularity detection algorithm of fault transient signal: The detection algorithm of the singular point position of the slowly varying singular signal is summarized as follows: ① Perform discrete wavelet transform on f using wavelet, and use Mallat algorithm to obtain {W(x j ,s,Ψ)j=1,2,...,N};②For data {W(x j ,s,Ψ)j=1,2,...,N}, and record the corresponding maximum point, then the maximum point is the singular point of the slowly varying singular signal; the wavelet transform in the algorithm is defined by the following formula: For any function f(t)∈l 2 (R), its wavelet transform is:

5. The abnormal early warning method for photovoltaic power station based on wavelet packet-BP neural network according to claim 1, characterized in that: In step 2, a photovoltaic power station fault simulation model is established according to the non-standard current related characteristics, specifically: the fault signal model is set as: It can be verified that the derivative of the signal at point t = 0 does not exist, and the singularity is less than 1, so the wavelet with a vanishing moment of 1 corresponding to the 4th-order transfer function is selected for singularity detection; The specific steps of detecting the time when a fault occurs include: Step 2.1: The sampling sequence is given by the fault signal. The number of sampling points per week is 320. Half a week is taken before and after the fault, that is, 160 points each; Step 2.2: Use FFT to analyze the fault signal and obtain the decomposition sequence; Step 2.3: Use the wavelet corresponding to the selected 4th-order transfer function and the two-scale sequence of the scaling function {g n },{h n }, implement the Mallat algorithm of wavelet transform to obtain the decomposition sequence of fault signal; Step 2.4: Observe the decomposed information sequence of the fault signal sampling sequence to determine whether there is a maximum value, that is, to determine whether there is a fault, thereby determining the location of the fault point; Step 2.5: Draw the fault signal detection graph based on the calculated data.

6. The abnormal early warning method for photovoltaic power station based on wavelet packet-BP neural network according to claim 1, characterized in that: The calculation of power quality supply voltage deviation and frequency deviation in step 3 is as follows: voltage deviation refers to the difference between the actual voltage of a node in the system and the rated voltage of the system under normal working conditions; frequency deviation refers to the difference between the actual frequency of the power system operation and the rated frequency of the system; the calculation of voltage deviation is: assuming that the voltage deviation on the power supply bus is δU A %, the voltage loss of high-voltage line l1 is The voltage deviation caused by the transformer is δU T %, the voltage loss of low voltage line l2 is Then the voltage deviations at points B, C, and D are: Extending the above concept to any power supply system, if there are multiple levels and voltages or voltage regulating equipment from the power supply to a specified point, the voltage deviation at the specified point can be calculated by the following formula: δU E % = ∑δU% - ∑ΔU%, where ∑δU% is the sum of all voltage deviations from the power supply to the specified point; ∑ΔU% is the sum of all voltage losses from the power supply to the specified point.

7. The abnormal early warning method for photovoltaic power station based on wavelet packet-BP neural network according to claim 1, characterized in that: In step 4, a photovoltaic power station abnormality warning method based on BP neural network is used to implement photovoltaic power station abnormality warning, including photovoltaic power station island detection and photovoltaic power station self-diagnosis monitoring; wherein photovoltaic power station island detection is implemented by a method based on wavelet transform and BP neural network, specifically including the following steps: Step 4.1: The central controller of the photovoltaic power station collects the current i flowing from the inverter power supply into the common coupling point PCC PCC and the voltage at the common coupling point V PCC , and then use wavelet transform to transform the collected current signal i PCC And the voltage signal V PCC Processing is performed to obtain the corresponding wavelet coefficients; Step 4.2: Process the obtained wavelet coefficients by algorithm to obtain a feature vector reflecting the characteristics of the collected signal; Step 4.3: These feature vectors are provided to the BP neural network, and the trained BP neural network performs pattern recognition and thereby determines the operating status of the photovoltaic power station; Step 4.4: When the BP neural network detects that the photovoltaic power station is in an island operation state, the central controller takes island protection measures to stop the photovoltaic power generation system from supplying power to the local load. When the photovoltaic power station is not in an island operation state, the island protection does not work.

8. The abnormal early warning method for photovoltaic power station based on wavelet packet-BP neural network according to claim 7, characterized in that: The photovoltaic power station self-diagnosis monitoring in step 4 adopts the photovoltaic power station self-diagnosis monitoring based on BP neural network, specifically: the BP neural network is divided into three layers of input layer, hidden layer and output layer. Suppose the fault input mode x of the network is (T c ,I m ,U m ), the fault output of the hidden layer y=(y1,y2,y3); the expected output values ​​of various fault types: normal (0,0,0), panel cracking (1,0,0), panel aging short circuit (0,1,0), local shadow covering (0,0,1), the fault input mode function of the BP network X=(T c ,I m ,U m ) T , the fault output mode function of the hidden layer y=(y1,y2,y3) T , in order to build the mapping relationship between the input and output of the neural network, in the process of transmitting from the input layer to the hidden layer of the BP network, the output value formula of each neuron in the hidden layer is b i ={exp[-(∑a i ·W ij -θ j )r]+1} -1 , where a i Represents the input value of each neuron in the input layer; W ij Represents the connection weight from the input layer to the hidden layer; θ j represents the hidden layer threshold; r represents the correction coefficient of the S-type function; in the process of transmitting from the hidden layer to the output layer of the BP network, the output value formula of each neuron in the output layer is c i ={exp[-(∑b j ·V ji -γ j )r]+1} -1 , where b j Represents the input value of each neuron in the hidden layer; V ji represents the connection weight from the hidden layer to the output layer; γ j represents the threshold of the output layer; r represents the correction coefficient of the S-type function.

9. The photovoltaic power station abnormality early warning system based on wavelet packet-BP neural network as claimed in claim 1, comprising a photovoltaic power station real-time monitoring system, characterized in that: The photovoltaic power station real-time monitoring system includes a single-chip microcomputer, a data acquisition module, a monitoring module, a wireless communication module, a local SD card and a cloud platform; the photovoltaic power station abnormality warning system based on wavelet packet-BP neural network is used to execute the photovoltaic power station abnormality warning method based on wavelet packet-BP neural network described in any one of claims 1-8.

10. The photovoltaic power station abnormality early warning system based on wavelet packet-BP neural network according to claim 9, characterized in that: The monitoring module collects and monitors the power quality data information of the photovoltaic power station in real time; the single-chip microcomputer effectively manages the large-scale power quality data collected by the monitoring module in real time, controls the monitoring module to realize uninterrupted monitoring of the power station, and sends the data to the cloud platform when the conditions for data transmission are met; the local SD card is used to store the data information collected by the monitoring module in real time; the communication module is used to perform remote data transmission to realize remote real-time monitoring of power-related data of the photovoltaic power station; the cloud platform is used to receive relevant data information of the single-chip microcomputer to form a complete monitoring system.

Images