Photovoltaic system instability online detection method, system and device based on EEMD-LSTM neural network and medium

Abstract

The invention provides a photovoltaic system instability online detection method, system and device based on an EEMD-LSTM neural network, and a medium. The method comprises the following steps: S1, obtaining voltage information of each node of an actual photovoltaic system; s2, decomposing the voltage signal into a plurality of IMF components through an EEMD method; s3, selecting a plurality of components with the highest amplitude from the IMF components as characteristic signals; s4, extracting time domain features and frequency domain features of the feature signals; s5, constructing a database of the voltage characteristic quantity and the label; s6, taking the database as a training data set, inputting the database into the LSTM model for offline training, and generating a stability / instability classification model; and S7, monitoring the node stability of the photovoltaic system in real time. According to the method, EEMD signal decomposition is adopted, and signal modes under different frequencies can be adaptively decomposed; a training data set of an LSTM model is formed by constructing an EEMD decomposition result and a classification label, and then the method can have the stability online monitoring capability of different types of photovoltaic systems through the LSTM model.

Description

Technical Field

[0001] The present invention belongs to the technical field of photovoltaic system detection, and in particular relates to an online detection method, system, equipment and medium for photovoltaic system instability based on an EEMD-LSTM neural network. Background Art

[0002] To address the climate crisis, building a new power system dominated by renewable energy is a key approach to promoting a clean, low-carbon energy transition. With the continuous development of power systems and renewable energy generation technologies, more and more renewable energy sources such as wind power and photovoltaics, as well as power electronics, are being integrated into the grid. This has significantly increased the degree of power electronics in the power system, forming a power-electronic power system. However, due to the low inertia of power electronics and the frequent disturbances experienced by many parts of the power system, the operating state of the power system can fluctuate significantly when subjected to small disturbances such as random load fluctuations and slowly changing system parameters. Furthermore, the dynamic characteristics of power electronics make the system's stability under small disturbances more prominent than in traditional power systems. In actual photovoltaic systems, the power flow and operating stability of the system are constantly changing as photovoltaic power generation and system load fluctuate. Therefore, even if the system is operating stably at one moment, there is no guarantee that it will not become unstable the next. Therefore, real-time instability monitoring technology for photovoltaic systems is essential. It can immediately issue an instability alarm when the system becomes unstable, prompting the activation of emergency protection mechanisms before the system suffers losses due to the instability.

[0003] Currently, there are two main types of real-time monitoring technologies for photovoltaic system instability: one is to perform online monitoring of the loop gain of the converters in the system, which can determine in real time whether the converters have instability problems; the other is to monitor the entire DC distribution system for instability problems caused by converter interaction mismatch.

[0004] Instability detection techniques based on online monitoring of the converter's loop gain can be categorized as modulated wave perturbation testing or pulse width perturbation testing, depending on the method of disturbance injection. Both methods employ a network analyzer to inject a sinusoidal perturbation signal into the converter's closed-loop control loop. The loop gain is determined by measuring the difference between the injection point and the return signal after the perturbation frequency passes through the loop. The difference lies in that the modulated wave perturbation method injects the perturbation signal before the PWM (Pulse Width Modulation) modulator, while the pulse width perturbation method injects the perturbation signal directly onto the converter's duty cycle. While these techniques can determine converter instability, they require intrusion into the converter's internal control loop, making them difficult to use in photovoltaic systems using "black box" converters. Furthermore, these methods can only detect converter-level instability, limiting their ability to monitor system-level instability.

[0005] The most traditional solution for online monitoring of instability issues in the entire photovoltaic system is to use a network analyzer to measure the port impedance of each converter in the system, and then perform stability analysis based on the impedance criterion of the system. This solution is intuitive, but the operation is time-consuming and the calculations are complex. Therefore, based on the above monitoring solution, an instability monitoring solution has been derived that injects system bus voltage (current) disturbances and quickly calculates the system equivalent loop gain based on bus current (voltage) feedback. The premise of this type of technology is that the equivalent loop gain of the system must be clearly known. However, the operating mode and power flow of the actual photovoltaic system change in real time, and the system equivalent loop gain corresponding to the system at different times is often different. This factor limits the use of this type of technology in actual instability monitoring.

[0006] In power systems, detecting subsynchronous oscillations can be used to determine whether the system is experiencing instability. This approach directly determines system instability by observing whether the system busbar waveform is oscillating. This approach closely matches the "black box" principle and is unconstrained by system operating stability and power flow fluctuations, making it a valuable model. However, direct application of this instability determination method based on busbar oscillation information to photovoltaic systems presents significant limitations. This is primarily due to the fact that busbar voltage in actual DC distribution systems contains a complex information structure. In addition to voltage information caused by system instability, it also includes ripple information introduced by secondary power pulsation in the grid, harmonic information introduced by nonlinear loads, fluctuations introduced by wind / solar power variations, and pulsation information introduced by single-phase AC loads. Further research is needed to directly monitor system instability in real time based on actual DC distribution system busbar information.

[0007] Chinese patent CN117937520A discloses a method and system for analyzing the stability of a new energy multi-machine system. The method first acquires waveform data from the grid-connected point of the new energy multi-machine system during a monitoring period. The waveform data includes the three-phase current and voltage values, as well as the maximum three-phase voltage value, during the monitoring period. The method then calculates the amplitude sensitivity and critical amplitude sensitivity of the new energy multi-machine system based on the waveform data. The method also uses a fast Fourier transform to calculate the voltage and current oscillation frequencies of the new energy multi-machine system under three phases. Based on the amplitude sensitivity, critical amplitude sensitivity, frequency sensitivity, and critical frequency sensitivity, the method determines a strategy for improving the stability of the new energy multi-machine system. This method quantitatively evaluates the small-signal stability of the new energy multi-machine system. However, this method obtains various indicator values ​​for a fixed system, making it difficult to accurately detect stability issues in complex operating conditions of a "black box" photovoltaic system. Summary of the Invention

[0008] The first object of the present invention is to provide an online detection method for photovoltaic system instability based on EEMD-LSTM neural network.

[0009] To this end, the above-mentioned purpose of the present invention is achieved through the following technical solutions:

[0010] An online detection method for photovoltaic system instability based on EEMD-LSTM neural network includes the following steps:

[0011] S1. Obtain voltage information of each node of the actual photovoltaic system;

[0012] S2, decompose the voltage signal into multiple intrinsic mode components (IMFs) through the ensemble empirical mode decomposition method (EEMD);

[0013] S3. Select multiple components with the highest amplitude from the eigenmode components as characteristic signals;

[0014] S4, extracting time domain features and frequency domain features of the characteristic signal;

[0015] S5. Construct the stability detection problem of the photovoltaic system into two labels: "stable" or "instability", and set the "instability" label threshold corresponding to each input feature;

[0016] Obtain node voltage information of actual photovoltaic systems under various working conditions and build a database of voltage characteristics and labels;

[0017] S6. Use the database as a training dataset and input it into a long short-term memory neural network (LSTM) model for offline training to generate a stable / instability classification model;

[0018] S7. Deploy the stability / instability classification model to monitor the node stability of the photovoltaic system in real time.

[0019] While adopting the above technical solutions, the present invention may also adopt or combine the following technical solutions:

[0020] As a preferred technical solution of the present invention: in step S2, the ensemble empirical mode decomposition method includes the following steps:

[0021] S21, adding a white noise sequence to the voltage signal, and performing multiple modal decompositions through empirical mode decomposition (EMD);

[0022] S22. Average the eigenmode component results of multiple decompositions to obtain the final eigenmode component and redundant terms.

[0023] As a preferred technical solution of the present invention: the empirical mode decomposition method includes the following steps:

[0024] S221, performing local maximum search and local minimum search on the voltage signal to find all local maximum points and local minimum points, and fitting the upper envelope of the local maximum points and the lower envelope of the local minimum points respectively;

[0025] S222, obtaining the average of the upper envelope of the local maximum point of the voltage signal and the lower envelope of the local minimum point;

[0026] S223. Extracting the mean value from the voltage signal to obtain a decomposed signal;

[0027] S224, repeating the above signal decomposition process until the decomposed signal obtained meets the conditions of the intrinsic mode function, marking the decomposed signal as a qualified intrinsic mode component;

[0028] S225, separating the qualified intrinsic mode components obtained in step S224 from the voltage signal, calculating a residual signal, and repeating the above signal decomposition process on the residual signal to calculate the remaining intrinsic mode components;

[0029] The EMD method stops when the intrinsic mode component and the residual signal are smaller than a preset threshold or the residual signal is a monotonic function.

[0030] As a preferred technical solution of the present invention: the number of modal decompositions is not less than 2.

[0031] As a preferred technical solution of the present invention: in step S3, the number of components is 1 to 3.

[0032] As a preferred technical solution of the present invention: in step S4, the time domain features include: time domain maximum value and time domain peak-to-peak value.

[0033] As a preferred technical solution of the present invention: in step S4, the frequency domain features include: frequency domain root mean square frequency and frequency domain centroid frequency.

[0034] The second object of the present invention is to provide an online detection system for photovoltaic system instability based on an EEMD-LSTM neural network, comprising the following modules:

[0035] - Voltage information acquisition module, used to obtain the voltage information of each node of the actual photovoltaic system;

[0036] - a voltage signal decomposition module, configured to decompose the voltage signal acquired by the voltage information acquisition module into a plurality of intrinsic mode components (IMFs) by using an ensemble empirical mode decomposition method (EEMD);

[0037] - a characteristic signal screening module, used to select multiple components with the highest amplitudes from the intrinsic mode components decomposed by the voltage signal decomposition module as characteristic signals;

[0038] - A time domain and frequency domain feature extraction module, used to extract the time domain features and frequency domain features of the characteristic signal screened by the characteristic signal screening module;

[0039] A voltage feature and label database construction module is used to construct two labels for PV system stability detection, namely "stable" or "instability", set the "instability" label threshold corresponding to each input feature, and obtain node voltage information of actual PV systems under various working conditions to build a database of voltage feature and labels;

[0040] - A stable / instability classification model generation module is used to use the voltage feature and label database constructed by the voltage feature and label database construction module as a training dataset, input it into a long short-term memory neural network (LSTM) model for offline training, and generate a stable / instability classification model.

[0041] A third object of the present invention is to provide an electronic device comprising a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other via the communication bus, and wherein:

[0042] a memory for storing a computer program;

[0043] A processor is used to execute the computer program stored in the memory to implement the steps of the online detection method for photovoltaic system instability based on the EEMD-LSTM neural network as described above.

[0044] Another object of the present invention is to provide a non-volatile storage medium, characterized in that: the non-volatile storage medium stores an executable program, and when the executable program is executed by a processor, it implements the steps of the online detection method for photovoltaic system instability based on the EEMD-LSTM neural network as described above.

[0045] The present invention provides a method, system, device and medium for online detection of photovoltaic system instability based on an EEMD-LSTM neural network. The method adopts the EEMD method for signal decomposition, which does not require any prior conditions and can adaptively decompose signal modes at different frequencies. In addition, in order to solve the problem that traditional methods can only detect the stability of photovoltaic systems with a fixed structure, an LSTM model is adopted. By constructing EEMD decomposition results and classification labels to form a neural network training data set, the LSTM model can be used to enable this method to have the ability to online monitor the stability of different types of photovoltaic systems. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 This is a flow chart of the online detection method for photovoltaic system instability based on the EEMD-LSTM neural network provided by the present invention.

[0047] Figure 2 This is a schematic diagram of the EEMD decomposition results of the photovoltaic system output voltage provided by the present invention.

[0048] Figure 3 This is the structural diagram of the long short-term memory neural network model. DETAILED DESCRIPTION

[0049] The present invention will be described in further detail with reference to the accompanying drawings and specific embodiments.

[0050] like Figure 1 As shown in Figure 1, an online detection method for photovoltaic system instability based on EEMD-LSTM neural network is proposed. The specific process is as follows:

[0051] Step 1: Measure the bus voltage information of the actual photovoltaic system and input it into the EEMD method for signal decomposition.

[0052] Step 2: Decompose the complex bus voltage into several approximate sinusoidal waveforms containing system instability oscillation information through EEMD.

[0053] The following is the specific calculation process of the EEMD method:

[0054] (1) Input the initial photovoltaic system measured voltage signal x0(t);

[0055] (2) Given a white random white noise sequence ε(t) and the number of repeated decompositions m;

[0056] (3) Integrate the white noise sequence with x0(t) to obtain the sequence x(t) with white noise added;

[0057] (4) Decompose the sequence with white noise by EMD method to obtain n IMF (intrinsic mode component) sublayers c j,i (t) and a redundant term r i (t), such as Figure 2 As shown; wherein n is a positive integer greater than or equal to 2;

[0058] Among them, the decomposition process of EMD is as follows:

[0059] (a) For a given bus voltage signal x(t), perform a local maximum / minimum search. Find all local maxima and connect them using cubic spline interpolation or other interpolation methods to approximate the upper envelope u(t). Repeat the same process for the local minimum to obtain the lower envelope v(t).

[0060] (b) Obtain the mean m1(t) of the upper and lower local extreme envelopes of the given voltage signal x(t). The calculation method is as follows:

[0061]

[0062] (c) Extract the mean signal m1(t) from the given voltage signal x(t) to obtain the decomposed signal h1(t):

[0063] h1(t)=x(t)-m1(t)

[0064] (d) Repeat the above signal decomposition process k times until the decomposition signal h is obtained. k (t) satisfies the conditions of the intrinsic mode function, then h k (t) is recorded as a qualified IMF component.

[0065] (e) Separate the found IMF components from the original signal and calculate the residual signal. Repeat the decomposition process (a) to (d) on the residual signal to calculate the remaining IMF components. The EMD algorithm stops when the final IMF component and the residual signal are less than a preset threshold or the residual signal is a monotonic function.

[0066] (5) Repeat the above steps m times, adding different white noise each time to obtain the corresponding IMF sublayer; where m is a positive integer greater than or equal to 2. The sequence after the i-th decomposition can be expressed as:

[0067]

[0068] Where i represents the result of the i-th decomposition, and the range of i is [1, m]; j represents the j-th IMF sublayer, and the range of j is [1, n].

[0069] (6) The final IMF sub-layer result is obtained by calculating the average value of all IMF sub-layers:

[0070]

[0071] (7) The sequence with white noise added is finally decomposed into:

[0072]

[0073] Among them, c j (t) represents the jth IMF component; r(t) is a redundant term.

[0074] Step 3: To eliminate useless information and select as few input features as possible to improve the stability monitoring speed, the 1-3 IMFs with the highest amplitudes are selected as the main feature signals for processing.

[0075] Step 4: Extract the time domain and frequency domain features of the selected IMF, including the time domain maximum value, time domain peak-to-peak value, frequency domain root mean square frequency and frequency domain centroid frequency as the final input feature quantities.

[0076] Step 5: The stability detection problem of the photovoltaic system is constructed into two labels: "stable" or "instability". According to actual needs, the "instability" label threshold corresponding to each input feature value is set. If the threshold is not met, the system node is judged to be unstable.

[0077] According to the label construction method of input feature quantities, the node voltages of actual photovoltaic systems under various working conditions are measured, and then a database of voltage feature quantities and labels is constructed.

[0078] Step 6: Based on the database constructed in step 5, input it into the LSTM model for offline training. The further description of the LSTM neural network model is as follows:

[0079] The internal structure of the LSTM model is as follows Figure 3 As shown in Figure 2, an LSTM unit consists of a forget gate, an input gate, and an output gate. The forget gate determines which information should be discarded or retained; the input gate updates the unit state; and the output gate determines the value of the next hidden state, which contains the previously input information. The calculation formulas between the variables in the LSTM unit are as follows:

[0080] f t =σ(W f ·[y t-1 ,x t ]+bf )

[0081] i t =σ(W i ·[y t-1 ,x t ]+b i )

[0082] o t =σ(W o ·[y t-1 ,x t ]+b o )

[0083] g t =tanh(W g ·[y t-1 ,x t ]+b g )

[0084] c t =f t c t-1 +i t ·g t

[0085] y t =o t tanh(c t )

[0086] Among them, x t is the current input value; y t-1 and y t Represent the previous state and current state of the hidden layer respectively; c t-1 and c t Represent the previous and current unit memory information respectively; W f ,W i ,W o ,W g is the weight matrix; b f ,b i ,b o ,b g is the bias vector; σ is the activation function; f t is the output of the forget gate; i t is the output of the input gate; o t is the output of the output gate.

[0087] Step 7: Use the trained LSTM model as a classifier for real-time monitoring of photovoltaic system stability. Directly input the actual measured photovoltaic system bus voltage operation data, and output the stability judgment result to complete stability monitoring.

[0088] The present invention also provides an online detection system for photovoltaic system instability based on EEMD-LSTM neural network, which includes the following modules:

[0089] - Voltage information acquisition module, used to obtain the voltage information of each node of the actual photovoltaic system;

[0090] - a voltage signal decomposition module, configured to decompose the voltage signal acquired by the voltage information acquisition module into a plurality of intrinsic mode components (IMFs) by using an ensemble empirical mode decomposition method (EEMD);

[0091] - a characteristic signal screening module, used to select multiple components with the highest amplitudes from the intrinsic mode components decomposed by the voltage signal decomposition module as characteristic signals;

[0092] - A time domain and frequency domain feature extraction module, used to extract the time domain features and frequency domain features of the characteristic signal screened by the characteristic signal screening module;

[0093] A voltage feature and label database construction module is used to construct two labels for PV system stability detection, namely "stable" or "instability", set the "instability" label threshold corresponding to each input feature, and obtain node voltage information of actual PV systems under various working conditions to build a database of voltage feature and labels;

[0094] - A stable / instability classification model generation module is used to use the voltage feature and label database constructed by the voltage feature and label database construction module as a training dataset, input it into a long short-term memory neural network (LSTM) model for offline training, and generate a stable / instability classification model.

[0095] The present invention further provides an electronic device, comprising a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other via the communication bus, and wherein:

[0096] a memory for storing a computer program;

[0097] A processor is used to execute the computer program stored in the memory to implement the steps of the online detection method for photovoltaic system instability based on the EEMD-LSTM neural network as described above.

[0098] The present invention also provides a non-volatile storage medium, characterized in that: the non-volatile storage medium stores an executable program, and when the executable program is executed by a processor, it implements the steps of the online detection method for photovoltaic system instability based on the EEMD-LSTM neural network as described above.

[0099] The above-mentioned specific implementation methods are used to illustrate the present invention and are only preferred embodiments of the present invention, rather than limiting the present invention. Any modifications, equivalent substitutions, improvements, etc. made to the present invention within the spirit of the present invention and the scope of protection of the claims shall fall within the scope of protection of the present invention.

Claims

1. A photovoltaic system instability online detection method based on EEMD-LSTM neural network, characterized by: The steps include: S1. Obtain voltage information of each node of the actual photovoltaic system; S2, decomposing the voltage signal into multiple intrinsic mode components by using the ensemble empirical mode decomposition method; S3. Select multiple components with the highest amplitude from the eigenmode components as characteristic signals; S4, extracting time domain features and frequency domain features of the characteristic signal; S5. Construct the stability detection problem of the photovoltaic system into two labels: "stable" or "instability", and set the "instability" label threshold corresponding to each input feature value; Obtain node voltage information of actual photovoltaic systems under various working conditions and build a database of voltage characteristics and labels; S6. Using the database as a training data set, inputting it into a long short-term memory neural network model for offline training to generate a stable / instability classification model; S7. Deploy the stability / instability classification model to monitor the node stability of the photovoltaic system in real time.

2. The method according to claim 1, wherein: In step S2, the ensemble empirical mode decomposition method includes the following steps: S21, adding a white noise sequence to the voltage signal, and performing multiple modal decompositions through empirical mode decomposition; S22. Average the eigenmode component results of multiple decompositions to obtain the final eigenmode component and redundant terms.

3. The method according to claim 2, wherein: The empirical mode decomposition method includes the following steps: S221, performing local maximum search and local minimum search on the voltage signal to find all local maximum points and local minimum points, and fitting the upper envelope of the local maximum points and the lower envelope of the local minimum points respectively; S222, obtaining the average of the upper envelope of the local maximum point of the voltage signal and the lower envelope of the local minimum point; S223. Extracting the mean value from the voltage signal to obtain a decomposed signal; S224, repeating the above signal decomposition process until the decomposed signal obtained meets the conditions of the intrinsic mode function, marking the decomposed signal as a qualified intrinsic mode component; S225, separating the qualified intrinsic mode components obtained in step S224 from the voltage signal, calculating a residual signal, and repeating the above signal decomposition process on the residual signal to calculate the remaining intrinsic mode components; The EMD method stops when the intrinsic mode component and the residual signal are smaller than a preset threshold or the residual signal is a monotonic function.

4. The method according to claim 2, wherein: The number of modal decompositions is not less than 2.

5. The method according to claim 1, wherein: In step S3, the number of components is 1 to 3.

6. The method according to claim 1, wherein: In step S4, the time domain features include: time domain maximum value and time domain peak-to-peak value.

7. The method according to claim 1, wherein: In step S4, the frequency domain features include: frequency domain root mean square frequency and frequency domain centroid frequency.

8. An online detection system for photovoltaic system instability based on EEMD-LSTM neural network, characterized by: Includes the following modules: - Voltage information acquisition module, used to obtain the voltage information of each node of the actual photovoltaic system; - a voltage signal decomposition module, configured to decompose the voltage signal acquired by the voltage information acquisition module into a plurality of intrinsic mode components by using an ensemble empirical mode decomposition method; - a characteristic signal screening module, used to select multiple components with the highest amplitudes from the intrinsic mode components decomposed by the voltage signal decomposition module as characteristic signals; - A time domain and frequency domain feature extraction module, used to extract the time domain features and frequency domain features of the characteristic signal screened by the characteristic signal screening module; A voltage feature and label database construction module is used to label the stability of the photovoltaic system as "stable" or "instability", set the "instability" label threshold for each input feature, and obtain node voltage information of the actual photovoltaic system under various operating conditions to build a database of voltage feature and labels; - A stable / instability classification model generation module is used to use the voltage feature quantity and label database constructed by the voltage feature quantity and label database construction module as a training data set, input it into the long short-term memory neural network model for offline training, and generate a stable / instability classification model.

9. An electronic device comprising a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other via the communication bus, wherein: a memory for storing a computer program; A processor, wherein the processor is configured to execute a computer program stored in a memory to implement the steps of the method for online detection of photovoltaic system instability based on an EEMD-LSTM neural network as described in any one of claims 1 to 7.

10. A non-volatile storage medium, characterized in that: The non-volatile storage medium stores an executable program, and when the executable program is executed by the processor, the steps of the online detection method for photovoltaic system instability based on the EEMD-LSTM neural network are implemented as described in any one of claims 1 to 7.

Images