A broadband oscillation in-situ monitoring and cooperative defense system

Abstract

The application provides a wide-frequency oscillation on-site monitoring and cooperative defense system, relates to the technical field of power system control, and comprises: a new energy substation used for acquiring electrical parameters of a new energy station, calculating oscillation characteristic quantities of the new energy station based on the acquired electrical parameters, and realizing the functions of wide-frequency oscillation evaluation and early warning and control and removal; a regional master station used for receiving the electrical parameters and the oscillation characteristic quantities sent by the new energy substation, locating an oscillation source, and realizing precise control in different rounds; and a dispatching master station used for monitoring the frequency and voltage level of a power system and the removal amount information of each regional master station in real time, carrying out system safety checking in combination with the system frequency, the voltage level, the removal amount of each master station and the system operation condition, sending a locking instruction to the relevant regional master stations, and realizing effective prevention and control of wide-frequency oscillation on the basis of ensuring the safety of a large power grid. The application realizes precise positioning and cooperative prevention and control of wide-frequency oscillation of a new type of power system.

Description

Technical Field

[0001] This invention relates to the field of power system control technology, and more specifically, to a broadband oscillation local monitoring and collaborative defense system. Background Technology

[0002] With the large-scale grid connection of new energy sources, the formation of high-voltage direct current transmission networks, and the commissioning of power electronic loads, the interaction between power electronic equipment and the power grid in the power system can cause broadband oscillations ranging from a few Hz to several kiloHz. Broadband oscillations can damage power equipment, leading to the shutdown of new energy generating units, seriously affecting equipment safety and threatening the stable operation of the system, becoming a significant factor restricting the efficient absorption of new energy. In recent years, broadband oscillation problems have frequently occurred in regions with a high proportion of new energy sources both domestically and internationally. Some densely integrated new energy areas have experienced oscillation events across multiple frequency ranges during engineering commissioning and operation.

[0003] Due to the multi-timescale, strong nonlinear characteristics of power electronic equipment and the complex coupling between equipment, multiple generator groups in new energy power plants interact with each other. The generation mechanism, oscillation characteristics and propagation range of broadband oscillations are intricate and complex, and they have significant characteristics such as wide frequency domain, strong time-varying, strong nonlinearity, multimodal and wide-area propagation. At present, there is a lack of unified and effective mathematical models and analysis methods, making it difficult to accurately analyze and implement effective prevention and control measures through offline simulation. Online monitoring and prevention and control are urgently needed.

[0004] Traditional broadband oscillation monitoring and control measures focus on online identification of oscillation frequency and amplitude, but they are insufficient in accurately tracing the source. Because the root cause of oscillation problems is difficult to identify, the conventional approach is to "crudely" disconnect the corresponding power plants once the oscillation amplitude and frequency reach a threshold. With the large-scale centralized connection of renewable energy in desert areas, the scale of renewable energy integration in areas with strong electrical coupling exceeds 10 million kilowatts. The number of units affected by broadband oscillations has increased significantly. The traditional method of monitoring and disconnecting oscillating power plants on-site will lead to large-scale disconnection of renewable energy from the grid, seriously affecting the stable operation of the power grid.

[0005] There are currently no effective solutions to the problems in the relevant technologies. Summary of the Invention

[0006] In view of this, the present invention provides a broadband oscillation local monitoring and collaborative defense system to solve the aforementioned problems.

[0007] To solve the above problems, the specific technical solution adopted by the present invention is as follows:

[0008] A broadband oscillation local monitoring and collaborative defense system includes:

[0009] The new energy substation is used to acquire the electrical parameters of the new energy power station, calculate the oscillation characteristic of the new energy power station based on the acquired electrical parameters, and send the electrical parameters and oscillation characteristic to the regional master station when the oscillation characteristic reaches the preset oscillation threshold, and receive and execute the generator switching command issued by the regional master station.

[0010] The regional master station is used to receive electrical parameters and oscillation characteristics sent by the new energy substations and locate the oscillation source; it selects the generator to switch based on the oscillation source location result and sends the generator to switch to the new energy substation according to the generator selection structure;

[0011] The dispatch master station is used to monitor the power system frequency, voltage level, and the amount of generator tripping information of each regional master station in real time. It also performs safety verification based on the power system frequency, voltage level, and generator tripping information of each regional master station, and carries out coordinated power defense based on the verification results.

[0012] Preferably, the new energy substation includes:

[0013] The data acquisition and analysis unit is used to acquire electrical parameters of new energy power plants and identify oscillation events and oscillation event patterns in new energy power plants by analyzing the development trend and distribution of electrical parameters.

[0014] The regional master station transmitting unit is used to send the electrical parameters of the new energy power station to the regional master station after oscillations occur in the new energy power station;

[0015] The receiving and execution unit is used to receive and execute the switching instructions issued by the regional master station.

[0016] Preferably, the data acquisition and analysis unit includes:

[0017] The data monitoring unit is used to monitor and analyze the electrical parameters of wind farms, photovoltaic power station power generation units and feeders in real time using an oscillation online monitoring and analysis device based on time-division multiplexing and adaptive sampling frequency technology. The electrical parameters include voltage, current, active power, reactive power and apparent power.

[0018] The window feature factor extraction unit is used to analyze the voltage-current angle and power curve fluctuation characteristics based on the electrical parameters of the wind farm, photovoltaic power station power generation unit and feeder, and to extract window feature factors using data statistics and extrapolation methods to form oscillation risk assessment elements.

[0019] The oscillation event identification unit is used to identify oscillation events based on oscillation risk assessment elements, through data statistics and trend analysis, and by combining multi-element weighting and fusion luminescence method.

[0020] The oscillation event pattern recognition unit is used to perform data augmentation processing on electrical parameters through a multimodal fusion strategy of data augmentation, and to build an oscillation pattern recognition model by fusing transfer learning and generative adversarial networks, and to identify oscillation event patterns using the oscillation pattern recognition model.

[0021] Preferably, the step of analyzing the voltage-current angle and power curve fluctuation characteristics based on the electrical parameters of the wind farm, photovoltaic power station power generation units and feeders, and extracting window feature factors using data statistics and extrapolation methods to form oscillation risk assessment elements includes:

[0022] Based on the electrical parameters of the wind farm, photovoltaic power station power generation units and feeders, determine the phase angle difference between voltage phase angle and current phase angle, and analyze the changing trend of phase angle difference;

[0023] Fluctuation analysis is performed on the curves of active power, reactive power, and apparent power in the electrical parameters of wind farms, photovoltaic power station generating units, and feeders to identify the fluctuation characteristics of the power curves.

[0024] Based on a pre-set sliding window length, variable window length disturbance analysis is performed on the electrical parameters of wind farms, photovoltaic power station power generation units and feeders to identify the fluctuation characteristics of power disturbance within the window period.

[0025] Based on the changing trend of phase angle difference, the curve fluctuation characteristics of power, and the fluctuation characteristics of power disturbance, window feature factors are extracted using data statistical methods, and these extracted window feature factors are used as elements for oscillation risk assessment.

[0026] Preferably, the method of identifying oscillation events based on oscillation risk assessment factors, through data statistics and trend analysis, and combined with multi-factor weighting and fusion luminescence method, includes:

[0027] Based on the elements of oscillation risk assessment, the importance of each element to the oscillation risk assessment is determined, and according to the determination results, a corresponding weight is assigned to each element.

[0028] The multi-factor weighting fusion light emission method is adopted to fuse the values ​​of each factor with their corresponding weights and present the fusion result in numerical form to form a quantitative risk value.

[0029] Determine whether the quantified risk value exceeds the preset risk threshold. If it does, it indicates that an oscillation event has occurred in the new energy power station; otherwise, it indicates that no oscillation event has occurred in the new energy power station.

[0030] Preferably, the step of performing data augmentation processing on electrical parameters using a data-enhanced multimodal fusion strategy, and fusing transfer learning and generative adversarial networks to construct an oscillation pattern recognition model, and using the oscillation pattern recognition model to identify oscillation event patterns includes:

[0031] The electrical parameter data is transformed using data transformation methods to generate preliminary data augmentation samples.

[0032] Based on the preliminary data augmentation samples, simulated disturbances are performed based on circuit laws such as Kirchhoff's laws and electromagnetic principles to generate new electrical parameter data samples.

[0033] Collect external environmental data related to new energy power stations, and fuse new electrical parameter data samples with external environmental data to form a multimodal dataset;

[0034] By using transfer learning techniques, a pre-trained convolutional neural network model is transferred to the oscillation pattern recognition task to obtain an oscillation pattern recognition model.

[0035] A generative adversarial network model is constructed to generate high-quality simulated oscillation data samples, and the training dataset is expanded. Using the multimodal dataset and the expanded training dataset, an oscillation pattern recognition model is trained, and the trained oscillation pattern recognition model is used to identify oscillation event patterns.

[0036] Preferably, the regional master station includes:

[0037] The oscillation data receiving unit is used to receive the electrical parameters and oscillation characteristics of the new energy power station sent by the new energy substation;

[0038] The oscillation data processing unit is used to calculate the oscillation characteristic quantity based on the electrical parameters of the new energy power station sent by the new energy substation, and to determine the oscillation source discrimination condition based on the oscillation characteristic quantity.

[0039] The vibration source determination unit is used to determine the vibration source in the new energy power station based on the auxiliary discrimination conditions of the vibration source;

[0040] The oscillation source action output unit is used to determine whether the oscillation source has diverged and whether the oscillation duration is greater than a preset time threshold. It sorts the comprehensive characteristic values ​​of the new energy power stations participating in the oscillation, sends a generator cut-off command signal to the new energy substation, and calms the oscillation according to the sorting result of the comprehensive characteristic values.

[0041] The generator tripping quantity uploading unit is used to send generator tripping command signals to the new energy substations and simultaneously send generator tripping quantity information to the dispatching center.

[0042] Preferably, the step of calculating the oscillation characteristic quantity based on the electrical parameters of the new energy power station sent by the new energy substation, and determining the oscillation source discrimination condition based on the oscillation characteristic quantity includes:

[0043] An adaptive normalization algorithm based on data entropy is used to normalize the electrical parameters of new energy power stations to obtain a normalized dataset. Based on the normalized dataset, the amplitude and start-up time of each device in the new energy power station are determined through a time analysis window, and the amplitude and start-up time of each device in the new energy power station are sorted to obtain the order of start-up time of each device.

[0044] Based on the normalized dataset of centralized oscillation mode frequencies, the oscillation mode frequencies are clustered, and the positive-sequence impedance of each oscillation mode is calculated based on the clustering results to obtain the voltage phasor and current phasor at the oscillation mode frequencies. Based on the positive-sequence impedance characteristic calculation formula, the positive-sequence impedance characteristics of the new energy power station equipment under the oscillation modes are calculated.

[0045] The apparent power in the normalized dataset is compared with a preset apparent power threshold, and the oscillation amplitude in the normalized dataset is compared with a preset oscillation amplitude threshold to obtain the comparison results.

[0046] The order of the start-up time of each device, the positive sequence impedance characteristics of the new energy power station equipment under the oscillation mode, and the comparison results are used as auxiliary discrimination conditions for the oscillation source.

[0047] Preferably, the adaptive normalization algorithm based on data entropy normalizes the electrical parameters of the new energy power station to obtain a normalized dataset, which includes:

[0048] For each characteristic quantity in the electrical parameters of a new energy power station, calculate the characteristic quantity information entropy;

[0049] It should be noted that the higher the entropy value, the greater the dispersion of the data. Adaptive scaling of the data is performed based on information entropy. For features with high entropy values ​​(large data dispersion), a smaller scaling factor is used; for features with low entropy values ​​(relatively concentrated data), a larger scaling factor is used.

[0050] The scaling factor is determined based on the information entropy of the feature quantity, and each electrical parameter is normalized using the scaling factor to obtain a normalized dataset.

[0051] Preferably, the clustering process for the oscillation mode frequencies includes:

[0052] The Z-score method is used to standardize the oscillation mode frequency data to obtain standardized oscillation mode frequency data.

[0053] Based on the standardized oscillation mode frequency data, the Euclidean distance between the frequency data of each oscillation mode is calculated using Euclidean distance.

[0054] The Euclidean distance between the frequency data of each oscillation mode is used as the input of the clustering algorithm. The density peak clustering algorithm is used to cluster the Euclidean distance between the frequency data of each oscillation mode to obtain the clustering result.

[0055] The step of clustering the Euclidean distance between the frequency data of each oscillation mode using the density peak clustering algorithm includes: constructing a distance matrix based on the Euclidean distance between the frequency data of each oscillation mode;

[0056] For each oscillation mode frequency data point in the distance matrix, its local density is calculated based on the Gaussian kernel function method;

[0057] For each oscillation mode frequency data point, calculate the minimum distance between it and all points with higher density than it;

[0058] Cluster centers are selected based on local density and minimum distance, and other points are assigned to corresponding clusters according to the density propagation principle to obtain the final clustering result.

[0059] Preferably, the dispatch master station includes: a power system monitoring unit, a generator tripping quantity receiving unit, a safety verification unit, and an instruction sending unit;

[0060] The power system monitoring unit is used to monitor the power system frequency and voltage level within a preset period.

[0061] The machine switching quantity receiving unit is used to receive machine switching quantity information sent by the regional master station;

[0062] The safety verification unit is used to perform safety verification by combining power system frequency, voltage level, generator tripping information, oscillation mode and power system operation status. When the quantity and range of generator tripping exceed the system frequency and voltage support regulation capacity, it generates a regional master station blocking command.

[0063] The instruction sending unit is used to send the generated regional master station blocking instruction to the regional master station.

[0064] The beneficial effects of this invention are as follows:

[0065] 1. This invention monitors and calculates data such as voltage, current, and oscillation frequency of wind farms or photovoltaic power station generating units or feeders, and calculates oscillation characteristic quantities such as oscillation amplitude, oscillation start time, spectrum, and transient energy in real time. For the first time, it adds indicators such as voltage active and reactive power sensitivity to improve the accuracy of oscillation source analysis and location. The main station in the area of ​​the new energy centralized access hub station has the function of receiving various oscillation characteristics sent by each substation and locating the oscillation source. It can accurately identify the oscillation source and key participating stations within a few seconds after the oscillation occurs based on comprehensive comparison of oscillation characteristics. Based on the oscillation source location, it selects the generator to be switched and issues the generator switching command to the new energy substation. The main station is built in the dispatch center, and the system safety verification is carried out in combination with system frequency, voltage level, generator switching amount of each main station and system operation status. When problems such as excessive total generator switching amount or wide range exceeding the system frequency voltage support regulation capacity occur in a short period of time, the blocking command is issued to the relevant regional main station to achieve effective prevention and control of broadband oscillation on the basis of ensuring the safety of the large power grid.

[0066] 2. In order to effectively address the impact of broadband oscillations on the power system, this invention further designs a collaborative defense mechanism. This mechanism not only relies on the monitoring and control of individual stations, but also includes the coordination and cooperation between various new energy power plants. By establishing a regional master station to receive and integrate real-time monitoring data from various substations, the system can quickly locate the oscillation source and, based on the characteristics and impact range of the oscillation source, rationally schedule various equipment in the power system and take coordinated defense measures. Attached Figure Description

[0067] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly described below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. In the drawings:

[0068] Figure 1 This is a schematic diagram of a broadband oscillation local monitoring and collaborative defense system according to an embodiment of the present invention;

[0069] Figure 2 This is a vector diagram of phase angle oscillation in a broadband oscillation local monitoring and collaborative defense system according to an embodiment of the present invention;

[0070] Figure 3 This is a vector diagram of amplitude oscillation in a broadband oscillation local monitoring and collaborative defense system according to an embodiment of the present invention;

[0071] Figure 4 This is a vector diagram of electrical resonance in a broadband oscillation local monitoring and collaborative defense system according to an embodiment of the present invention;

[0072] Figure 5 This is a waveform characteristic diagram of phase angle oscillation in a broadband oscillation local monitoring and collaborative defense system according to an embodiment of the present invention;

[0073] Figure 6 This is a waveform characteristic diagram of amplitude oscillation in a broadband oscillation local monitoring and collaborative defense system according to an embodiment of the present invention;

[0074] Figure 7 This is a waveform characteristic diagram of electrical resonance in a broadband oscillation local monitoring and collaborative defense system according to an embodiment of the present invention;

[0075] Figure 8 This is a trend classification diagram of oscillation characterization features in a broadband oscillation local monitoring and collaborative defense system according to an embodiment of the present invention;

[0076] Figure 9 This is a technical path diagram of an oscillation source based on comprehensive index criteria in a broadband oscillation local monitoring and collaborative defense system according to an embodiment of the present invention;

[0077] Figure 10 This is a flowchart of an auxiliary discrimination process for oscillation energy flow in a broadband oscillation local monitoring and collaborative defense system according to an embodiment of the present invention;

[0078] Figure 11 This is one of the harmonic source analysis results in a case study of a broadband oscillation local monitoring and collaborative defense system according to an embodiment of the present invention;

[0079] Figure 12 This is the second case study harmonic source analysis result in a broadband oscillation local monitoring and collaborative defense system according to an embodiment of the present invention;

[0080] Figure 13 This is a structural framework diagram of a broadband oscillation local monitoring and collaborative defense system according to an embodiment of the present invention;

[0081] Figure 14 This is a communication connection architecture for a broadband oscillation local monitoring and collaborative defense system according to an embodiment of the present invention;

[0082] Figure 15 This is a schematic diagram of a novel power system broadband oscillation mechanism in a broadband oscillation local monitoring and collaborative defense system according to an embodiment of the present invention.

[0083] In the picture:

[0084] 1. New energy substation; 2. Regional main station; 3. Dispatch main station. Detailed Implementation

[0085] To enable those skilled in the art to better understand the technical solutions in this application, the technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of this application.

[0086] According to an embodiment of the present invention, a broadband oscillation local monitoring and collaborative defense system is provided.

[0087] It should be noted that the oscillation mechanisms of new power systems are diverse, such as... Figure 15 As shown, the coupling and correlation are quite complex, affecting frequencies ranging from hundreds to even kilohertz, making the oscillation pattern of the power curve impossible to observe clearly. Clarifying the causes of oscillations and predicting them from the perspective of traditional mechanisms presents certain technical challenges. Since the three core elements for assessing the stability level of key electrical quantities in a power system are "frequency," "waveform," and "amplitude," and the main impact of oscillations is concentrated on the disruption and distortion of these core electrical quantity elements, we can start by analyzing the abrupt changes in these three elements, combining the different characteristics of the time and frequency domains, to analyze the characteristics and trends of the electrical quantities when oscillations occur, and thus provide early warning and location of oscillation occurrences. When measuring frequency, it is generally associated with a vector, as shown in the following formula:

[0088] ;

[0089] In the formula, The expression represents the prediction of oscillation, Δf represents the frequency difference, f0 represents the initial frequency, k represents the current sampling point index, N represents the modulation coefficient, r represents the subcarrier index, and φ represents the frequency difference. r X represents the phase shift of the r-th signal; X represents the signal amplitude.

[0090] As shown in Table 1, the following are: Figure 15 Annotations for Chinese and English phrases;

[0091] Table 1. Annotations for Chinese and English phrases

[0092] The current electrical quantity frequencies are calculated based on voltage waveforms. In power system sampling signal processing, it is assumed that the frequency changes very little within one cycle. Therefore, the frequency difference Δf(t) within the data window time range is represented as a fixed value Δf. The model also assumes that the amplitude is constant over a short period, thus obtaining a phase angle and frequency relationship independent of the amplitude, as shown in the following equation:

[0093] ;

[0094] In the formula, θ represents the phase angle of the power system at time t, Δf represents the frequency difference, f0 represents the initial frequency, and N represents the modulation coefficient.

[0095] Based on the phase angle formula above, it can be deduced that the capture of frequency elements before oscillation occurs can be achieved by analyzing the abrupt changes in phase angle. Frequency oscillation can also evolve into phase angle oscillation. Furthermore, combined with the periodic changes in amplitude, oscillation phenomena can be classified into: phase angle oscillation, amplitude oscillation, and electrical resonance. The vector diagrams for phase angle oscillation, amplitude oscillation, and electrical resonance are shown below. Figure 2-4 As shown, the waveform characteristics of phase angle oscillation, amplitude oscillation, and electrical resonance are as follows: Figure 5-7 As shown, the expressions for phase angle oscillation, amplitude oscillation, and electrical resonance are respectively:

[0096] ;

[0097] ;

[0098] ;

[0099] In the formula, Y p Y represents the electrical vector after phase angle oscillation. m Y represents the electrical vector after amplitude oscillation. c Let y0 represent the electrical vector after electrical resonance, y0 represent the electrical vector before oscillation, j represent the imaginary part, ω0 represent the initial angular frequency, and φ0 represent the initial phase angle. R The change in the electrical vector after oscillation occurs is represented by α, the attenuation factor of the oscillation signal is represented by t, and φ is represented by time. R Indicates y R The phase angle, ω R Indicates y R angular frequency.

[0100] The main forms of oscillation can be broadly categorized into changes in phase angle and amplitude, two quantities that can be intuitively perceived at the message data level. The power curve envelope of this oscillation shows significant changes, allowing for clear definition and analysis of the electrical quantity characteristics prior to oscillation. Based on the severity and sequential trend of changes in oscillation characteristic quantities, oscillation phenomena can be classified into nine main directions, such as... Figure 8 As shown, Figure 8 Among them, ① low-frequency oscillation of synchronous machines, ② subsynchronous oscillation dominated by the shaft system of synchronous machines, ③ oscillation of converter PLL / power synchronization, ④ stability problems dominated by the outer loop of converters, ⑤ stability problems dominated by the outer loop of conventional DC generators, ⑥ stability problems induced by slip of asynchronous machines, ⑦ subsynchronous oscillation caused by series compensation of synchronous machines, ⑧ subsynchronous oscillation caused by series compensation of doubly-fed generators, and ⑨ medium- and high-frequency oscillations dominated by the inner loop / feedforward of converters.

[0101] Depend on Figure 8 As shown, oscillation early warning for thermal power units can be identified by the trend of phase angle deviation, subsynchronous oscillation, etc., which requires superimposed with a trend of increasing oscillation amplitude. For example, in categories ①②③, the trend of phase angle growth will receive key attention, while in categories ④⑤⑥ involving new energy sources, the focus is more on the trend of amplitude changes. This differentiation in trend characteristics can serve as a key basis for trend judgment. Therefore, the main idea of ​​early warning positioning is to identify key characteristic quantities based on mechanism analysis, and analyze the development trend of these characteristic quantities in the time domain and their distribution trend in the frequency domain, such as... Figure 9 As shown. Figure 9 In this context, u1, u2, and un represent electrical quantities such as voltage, current, and power; y1, y2, and yn represent electrical quantity characteristic values ​​obtained through frequency and time domain calculations for oscillation analysis and localization; and z is the result of the comprehensive criterion fusion calculation.

[0102] like Figure 9 As shown, the basic concept of early warning and positioning is based on the frequency analysis of disturbance components, constructing a variable window length disturbance energy integral algorithm, calculating the disturbance energy growth trend, combining the voltage and current angle, the change trend of port impedance, and the fluctuation characteristics of the power curve, and other comprehensive weight indicators. At the regional master station, the characteristic quantities of each substation are comprehensively analyzed and compared, the oscillation source is accurately located and control commands are issued, and the whole network safety is checked at the dispatch master station, forming a large power grid wide-area broadband oscillation early warning, positioning and defense system based on oscillation path deduction and fixed threshold judgment.

[0103] The present invention will now be further described in conjunction with the accompanying drawings and specific embodiments, such as... Figure 1 As shown, the broadband oscillation local monitoring and collaborative defense system according to an embodiment of the present invention includes:

[0104] New Energy Substation 1 is used to acquire electrical parameters of new energy power stations, calculate the oscillation characteristic of new energy power stations based on the acquired electrical parameters, and send the electrical parameters and oscillation characteristic to the regional master station when the oscillation characteristic reaches the preset oscillation threshold, and receive and execute the generator switching command issued by the regional master station.

[0105] Regional master station 2 is used to receive electrical parameters and oscillation characteristics sent by new energy substations and locate the oscillation source; it performs generator switching selection based on the oscillation source location result and sends generator switching instructions to the new energy substations according to the generator switching selection structure;

[0106] Dispatch master station 3 is used to monitor the power system frequency, voltage level and the amount of generator tripping information of each regional master station in real time, and to perform safety verification based on the power system frequency, voltage level and the amount of generator tripping information of each regional master station, and to carry out power coordination defense based on the verification results.

[0107] In a preferred embodiment, the new energy substation 1 includes:

[0108] The data acquisition and analysis unit is used to acquire electrical parameters of new energy power plants and identify oscillation events and oscillation event patterns in new energy power plants by analyzing the development trend and distribution of electrical parameters.

[0109] The regional master station transmitting unit is used to send the electrical parameters of the new energy power station to the regional master station after oscillations occur in the new energy power station;

[0110] The receiving and execution unit is used to receive and execute the switching instructions issued by the regional master station.

[0111] In a preferred embodiment, the data acquisition and analysis unit includes:

[0112] The data monitoring unit is used to monitor and analyze the electrical parameters of wind farms, photovoltaic power station power generation units and feeders in real time using an oscillation online monitoring and analysis device based on time-division multiplexing and adaptive sampling frequency technology. The electrical parameters include voltage, current, active power, reactive power and apparent power.

[0113] The window feature factor extraction unit is used to analyze the voltage-current angle and power curve fluctuation characteristics based on the electrical parameters of the wind farm, photovoltaic power station power generation unit and feeder, and to extract window feature factors using data statistics and extrapolation methods to form oscillation risk assessment elements.

[0114] The oscillation event identification unit is used to identify oscillation events based on oscillation risk assessment elements, through data statistics and trend analysis, and by combining multi-element weighting and fusion luminescence method.

[0115] The oscillation event pattern recognition unit is used to perform data augmentation processing on electrical parameters through a multimodal fusion strategy of data augmentation, and to build an oscillation pattern recognition model by fusing transfer learning and generative adversarial networks, and to identify oscillation event patterns using the oscillation pattern recognition model.

[0116] In a preferred embodiment, the step of analyzing the voltage-current angle and power curve fluctuation characteristics based on the electrical parameters of the wind farm, photovoltaic power station generating units, and feeders, and extracting window feature factors using data statistics and extrapolation methods to form oscillation risk assessment elements includes: determining the phase angle difference between voltage and current phase angles based on the electrical parameters of the wind farm, photovoltaic power station generating units, and feeders, and analyzing the changing trend of the phase angle difference; performing fluctuation analysis on the curves of active power, reactive power, and apparent power in the electrical parameters of the wind farm, photovoltaic power station generating units, and feeders, and identifying the power curve fluctuation characteristics; performing variable window length disturbance analysis on the electrical parameters of the wind farm, photovoltaic power station generating units, and feeders based on a pre-set sliding window length, and identifying the power disturbance fluctuation characteristics within the window period; and extracting window feature factors using data statistics methods based on the changing trend of phase angle difference, the power curve fluctuation characteristics, and the power disturbance fluctuation characteristics, and using the extracted window feature factors as oscillation risk assessment elements.

[0117] It's important to note that the goal of variable window length disturbance analysis is to track the dynamic characteristics of electrical parameters (such as voltage, current, and active power) in real time and accurately extract disturbance components from the raw data stream. Traditional fixed window analysis (such as standard FFT) has limitations: when the window is too long (e.g., 5 seconds), although the frequency resolution is high, it averages out the instantaneous characteristics of sudden disturbances, leading to a significant lag in oscillation initiation timing. When the window is too short (e.g., 100 milliseconds), although the time resolution is high, it cannot accurately resolve the frequency and amplitude of low-frequency oscillations (such as 0.5Hz range oscillations). The variable window length mechanism uses a closed-loop feedback logic to dynamically adjust the analysis window length based on the non-stationarity of the signal itself. .

[0118] The implementation steps for variable window length perturbation analysis are as follows:

[0119] Step 1: Establishing the reference state and sliding detection (steady-state monitoring). When the system starts up, a reference analysis window of medium length is used. (For example, set to) Seconds, enough to cover 2-3 cycles of most low-frequency oscillation patterns, in Within the window, calculate electrical parameters (in terms of active power). (for example) mean and standard deviation .this Defined as the baseline steady-state disturbance level. A very short, rapidly sliding detection window is set. (For example, (Milliseconds or 2-3 power frequency cycles). This window slides continuously at a high sampling rate (e.g., 10kHz). In each... It calculates one or more perturbation indicator factors (PIs). The most commonly used PIs include short-time energy. Short-time zero crossing rate ( Short-term variance Set a dynamic threshold. ,For example ( (Usually 3 or higher). When detected... If the system determines that a potential disturbance event has occurred, it will re-enter the calculation of the mean and standard deviation of electrical parameters.

[0120] Step 2, Disturbance localization and windowing analysis (transient capture), assuming that in Detected at all times If the limit is exceeded, the system will be suspended immediately. Switch the window to high temporal resolution mode. Use a short positioning window. (For example, (milliseconds). Centered on, through Swipe the window back and forth to find The precise sampling point where an indicator (such as energy or slope) begins to increase dramatically. This ensures millisecond-level capture of the oscillation initiation moment, which is a crucial basis for subsequent oscillation initiation time sorting at the regional master station. A short window nearby (like to Within milliseconds, extract transient perturbation features, including:

[0121] Initial slope of disturbance (RoCoP): ;

[0122] Instantaneous impact energy: ;

[0123] Voltage-current angle mutation amount .

[0124] Step 3: Oscillation Mode Identification and Windowing Analysis (Steady-State Capture). The system then enters disturbance tracking mode. The dominant frequency f of the disturbance is determined. dom To set the optimal analysis window, the system employs a fast frequency estimation algorithm (such as wavelet transform). A rough estimate within a short period of time (e.g., 0.5 seconds) afterwards. .once Estimated, Analysis Window Will be dynamically adjusted to ,in This is the number of cycles required to ensure frequency resolution (e.g.) ).like Hz, then The system will use a 10-second long window for Prony analysis or FFT to accurately calculate the amplitude, frequency, and damping ratio of the 0.5Hz mode. Hz, then Milliseconds. The system will use a short window of 62.5 milliseconds for analysis to avoid contamination from power frequency and other low-frequency components. This is the optimal window determined adaptively. Internally, the system performs detailed disturbance analysis and extracts steady-state disturbance characteristics, including: dominant oscillation frequency. and amplitude Oscillation damping ratio Oscillation energy growth trend: by comparing continuous Oscillatory energy within the window. Power curve fluctuation characteristics: such as peaks, troughs, envelope shape, etc.

[0125] Step 4: If the disturbance ends (e.g., Continuous indivual Below the window The system will reset. Then return to the baseline state monitoring in step two.

[0126] Specifically, the main function of on-site monitoring at the site is to analyze the development trend of electrical characteristic quantities in the time domain and the distribution in the frequency domain after the oscillation occurs, to perform real-time monitoring and identification of oscillation modes, to trigger alarms and start recording when the monitoring equipment oscillates, and to transmit the frequency, amplitude, time and other oscillation characteristic quantities of the monitored oscillation to the regional master station.

[0127] To address the challenge of acquiring and transmitting a large volume of broadband electrical quantities for wind farms, photovoltaic power plants, and their feeders, a time-division multiplexing approach is employed to achieve synchronous acquisition and ensure data consistency. Regarding the acquisition frequency, an adaptive sampling frequency technique is used based on the varying characteristics of different electrical parameters. For rapidly changing parameters such as current and voltage, the sampling frequency is automatically increased to above 10kHz during oscillations, while it is reduced to 1kHz during steady-state operation to minimize data volume and processing burden.

[0128] Based on historical operation and oscillation data analysis, an innovative approach is adopted to identify the window characteristics of electrical quantities before oscillation. Primarily employing statistical and extrapolatory methods, window characteristic factors are extracted from changes in disturbance power density, harmonic power growth trends, harmonic frequency distribution changes, and voltage and current phase angle trends. These factors form oscillation risk assessment elements, and a weighted allocation is used to develop an overall oscillation early warning strategy integrating all elements. The window period refers to the period before a significant increase in oscillation power, during which abnormal symptom characteristics such as power disturbances and harmonic frequency points are observed.

[0129] Specifically, in the frequency domain, the analysis primarily relies on frequency analysis of disturbance components and variable-window-length disturbance energy integration algorithms to analyze the growth trend of disturbance energy. For oscillation events in different frequency bands, various spectrum analysis algorithms (such as FFT, PRONY, and transient energy algorithms) are employed to analyze the oscillation time-frequency domain characteristics within the range of 0.1-2500Hz. Combined with electrical quantity spectrum analysis, wide-frequency impedance characteristics are calculated to analyze port impedance change trends. Based on equipment impedance characteristics, oscillation source localization is assisted, and the overall disturbance energy flow forms the oscillation pattern and converter impedance monitoring and early warning. In the time domain, variable-window-length disturbance analysis and calculations are performed for voltage, current, and power, analyzing the voltage-current angle and power curve fluctuation characteristics. In addition to traditional oscillation amplitude and frequency, dynamic electrical characteristic quantities related to oscillation are added. Utilizing the characteristic that equipment voltage fluctuates significantly with active and reactive power before wide-frequency oscillation occurs, dynamic characteristic indicators such as voltage active power sensitivity dv / dp and voltage reactive power sensitivity dv / dq are added to assist in oscillation source early warning and localization. By analyzing data from the online monitoring system, key characteristic quantities for calculating broadband oscillations in a specific frequency band are calculated, followed by comprehensive analysis. The specific calculation details are as follows:

[0130] (1) Oscillation amplitude of active power P, reactive power Q, and apparent power S: Calculate the instantaneous power amplitude P by performing spectrum analysis on the power values ​​of P, Q, and S. max Q max S max The oscillation monitoring activation criterion is that the instantaneous power oscillation amplitude exceeds the preset threshold P. WFO Q WFO S WFO And continue for M seconds, recording the above oscillation amplitude and oscillation start time T. WFO P WFO The value of M can be tuned.

[0131] (2) Integral of P, Q, and S oscillation displacements: Perform spectral analysis on the P, Q, and S power values, and calculate the oscillation displacement corresponding to the time window t for the key oscillation frequency bands detected. , , If the oscillation is of constant amplitude, the integral quantity remains constant within a fixed time window; otherwise, it is an oscillation with decreasing or increasing amplitude. This indicator primarily describes the intensity of rapidly changing disturbance energy from a statistical perspective. Under steady-state operation, the power curve's moving average shows a stable trend across observation windows. As higher-frequency disturbance energy gradually increases, the power curve fluctuates significantly, and the power moving average often crosses and intersects frequently before exhibiting an observable, typical periodic oscillation waveform. This criterion is very suitable for medium-to-high frequency or high-energy oscillation scenarios.

[0132] In the formula, p(t) represents instantaneous active power, q(t) represents instantaneous reactive power, s(t) represents instantaneous apparent power, and P, Q, and S represent active power, reactive power, and apparent power, respectively.

[0133] (3) Energy distribution of oscillation disturbance: The oscillation displacement p(θ) corresponding to the phase angle θ of each power generation element in the new energy power station is calculated within a fixed window. The magnitude of disturbances within the observation window is quantified. This indicator mainly depicts the development trend of the oscillation power curve. By comprehensively comparing the oscillation disturbance energy of multiple devices, the oscillation path can be deduced.

[0134] In the formula, θ is the phase angle of each power generation element, and p(θ) is the oscillation displacement corresponding to the phase angle θ of the power generation element.

[0135] (4) Power Factor Angle Oscillation Amplitude: The power factor angle fluctuation amplitude Δφ is obtained by performing spectrum analysis on the φ value. Power Factor Characteristics: The power factor reflects the development trend of the voltage-current phase angle difference. In steady-state operation, the voltage-current phase angle difference should fluctuate within a certain range. When there is a risk of oscillation or oscillation occurs, the phase angle difference will show a swing-out characteristic. Through power factor angle analysis, the risk of oscillation disturbance in the early warning equipment can be extracted. This indicator mainly describes the development trend of the voltage-current phase angle difference. In steady-state operation, the voltage-current phase angle difference should fluctuate within a certain range, but in the early stage of oscillation, the phase angle difference will show a swing-out characteristic. The trend of the phase angle difference is also one of the main weighting factors for oscillation positioning.

[0136] (5) Voltage U and Current I Oscillation Amplitude: Spectral analysis of U and I values ​​yields the oscillation amplitudes ΔU and ΔI at the current key oscillation frequency band. For the potential occurrence of purely active or purely reactive oscillations at new energy power plants, , , , If the Δφ criterion fails, it is necessary to comprehensively consider the voltage and current oscillation amplitudes of the collector line, SVG, synchronous condenser, and wind turbine as auxiliary basis for oscillation source location. During steady-state operation, voltage and current will be accompanied by a certain proportion of disturbance components. In the early stage of oscillation, the proportion often shows an upward trend. Therefore, the analysis of the voltage and current disturbance trend is one of the main weighting factors for oscillation source location.

[0137] (6) Voltage active power sensitivity and voltage reactive power sensitivity: When the new energy converter operates near the critical point of the PV curve, and the voltage active power sensitivity and voltage reactive power sensitivity are relatively large, wide-frequency oscillation is prone to occur. Based on this characteristic, the electrical characteristic quantities dv / dp and dv / dq under a fixed time window are used as oscillation early warning indicators. At the same time, under different converter control parameters and structures, there are certain differences reflected in dv / dp and dv / dq. Therefore, this characteristic represents a certain converter impedance characteristic, which can reflect whether the device is an oscillation source from the impedance perspective.

[0138] (7) Voltage and current harmonic components: Through spectrum analysis, the proportion and spectrum distribution of non-power frequency components in voltage and current are obtained. The converter impedance has the risk of oscillation and may have a small number of non-power frequency disturbance components, which are on the rise. Based on this characteristic, the impedance model is predicted.

[0139] (8) After the above indicators are calculated, send the above indicators to the regional main station, and the regional main station will identify the oscillation source and select the appropriate measures.

[0140] (9) Communication, time synchronization and waveform recording functions;

[0141] 1) Supports GB / T26865.2 standard, and has the function of transmitting synchronization phasor, interharmonic, harmonic, alarm information and waveform recording files to the outside world; supports DL / T860 standard, and has the function of transmitting alarm information and status events to the outside world; it can also have the function of transmitting fundamental, harmonic and interharmonic measurement data to the outside world.

[0142] 2) It should have a timekeeping function. When the synchronization time signal is lost or abnormal, the device's timekeeping accuracy should be within 60 minutes, and the change in phase angle measurement error should not exceed 1°.

[0143] 3) Supports both triggered waveform recording and continuous waveform recording; triggered waveform recording supports alarm, manual and network triggering methods; the sampling rate of triggered waveform recording should not be lower than 12.8kHz, the recording time should not be less than 60s when the alarm signal is present, the waveform recording file should have no less than 256 entries, and the recording should be cyclic.

[0144] As a preferred embodiment, the step of identifying oscillation events based on oscillation risk assessment elements, through data statistics and trend analysis, and combined with a multi-element weighted fusion luminescence method includes: determining the importance of each element to the oscillation risk assessment based on the elements of oscillation risk assessment, and assigning corresponding weights to each element according to the determination results; using a multi-element weighted fusion luminescence method to fuse the values ​​of each element with their corresponding weights, and displaying the fusion result in numerical form to form a quantitative risk value; determining whether the quantitative risk value exceeds a preset risk threshold, if so, indicating that an oscillation event has occurred in the new energy power station, otherwise, indicating that no oscillation event has occurred in the new energy power station.

[0145] It should be noted that the multi-factor weighting fusion luminescence method is a decision-making algorithm for identifying oscillation events in new energy substations. It is a composite decision-making framework. Multi-factor weighting is its mathematical core (i.e., information fusion), while luminescence method is its engineering application and intuitive representation (similar to a risk traffic light warning). In this invention, the multi-factor weighting fusion luminescence method adopts a fusion framework based on Dempster-Shafer (DS) evidence theory to address the uncertainty and even contradictions (e.g., large power fluctuations but small angle changes) between multi-source window feature factors. The specific implementation steps are as follows:

[0146] Step 1: Define the recognition framework. The ultimate goal of fusion is to determine whether an oscillation event has occurred. Therefore, define the recognition framework. For a set of mutually exclusive and complete hypotheses: in: This indicates that an oscillation event has occurred in the new energy power station. This indicates that no oscillation events have occurred at the new energy power station.

[0147] Step 2: Define the fusion elements (window feature factors). These elements (factors) are the evidence in the DS theory. Based on the aforementioned variable window length perturbation analysis, define a set of key window feature factors. ,For example: Indicates the trend of the voltage-current angle ( ), Indicates the fluctuation characteristics of the power curve (such as the dominant frequency). amplitude ), This indicates the fluctuation characteristics of power disturbances within the window period (such as instantaneous impact energy). ), This indicates an increasing trend in oscillation energy. This indicates the active / reactive voltage sensitivity.

[0148] Step 3: Define element weights (i.e., assigned weights). Assigning appropriate weights is crucial for fusion. This invention employs a combined static and dynamic weight allocation strategy, specifically combining the Analytic Hierarchy Process (AHP) (subjective / static) with the Entropy Weight Method (EWM, objective / dynamic). Static weights (AHP) is based on the prior knowledge of power system experts (experience with actual grid oscillation characteristics) to construct a judgment matrix and determine the prior importance of each factor to oscillation risk (i.e., determine the importance of each factor in oscillation risk assessment). For example, in subsynchronous oscillations, (Angle) and The weight of (sensitivity) may be inherently higher than that of sensitivity. (Power fluctuation). Dynamic weighting (EWM) is a method for calculating each element in real time. The information entropy of the data sequence within the current time period. The smaller the entropy, the more concentrated the value of that element, the greater its information content or degree of variation, and the stronger its indicative power of the current state; therefore, it should be assigned a higher weight. The final weight calculation formula is:

[0149] ;

[0150] In the formula, Indicates the first The final combined weight of each element (evidence), i.e. the reliability discount coefficient of each piece of evidence in the DS theory. This represents a balancing factor (e.g., 0.5), used to adjust the ratio of subjective to objective weights. Indicates the first Static weights of each element (determined by AHP method). Indicates the first The dynamic weights of each element (determined by the entropy weight method).

[0151] Furthermore, the integration process of the DS evidence theory includes the following steps:

[0152] Step one, each element must be... Physical quantities (such as) The amplitude) is converted to the The support of the underlying hypothesis, i.e., the basic probability distribution or function This is achieved through a preset membership function. For example, for elements... (oscillation amplitude) ):if (e.g., <0.01 pu), then (Highly confident of safety), if ,but (Uncertain, but leaning towards oscillation), if (For example, >0.05 pu), then (Highly confident oscillation) This represents uncertainty, which is the key to how DS theory deals with uncertainty. Indicates the lower limit threshold. This indicates the upper limit threshold.

[0153] Step two, applying weights to the evidence: before fusion, the final weights are used. Discounting the BPA reflects the reliability of the evidence:

[0154] ;

[0155] ;

[0156] ;

[0157] Step three, evidence fusion (Dempster's combination rule), involves combining the results of the various elements. Suppose there are two discounted pieces of evidence. and Using Dempster's combination rules ( To merge them and generate a new support level:

[0158]

[0159] In the formula, Indicating fusion evidence and Then, regarding the hypothesis Support level. Representation of recognition framework subsets (e.g.) , or itself). This indicates that the two pieces of evidence are... and The discounted BPA (Basic Probability Allocation). Indicates all Combinations whose intersection is The situation. This represents the conflict coefficient, used for normalization. This indicates a summation. yes subsets of (i.e.) or ). It is the conflict coefficient, which measures the degree of contradiction between two pieces of evidence. The calculation formula is:

[0160] ;

[0161] In the formula, The conflict coefficient represents the ratio of two pieces of evidence. and The degree of contradiction between them. Indicates all Combinations whose intersection is an empty set (i.e., mutually conflicting). This indicates that the two pieces of evidence are... and BPA after discount.

[0162] For example, fusion and right Support level:

[0163] ;

[0164] In the formula, Indicates fusion and Then, regarding the oscillation hypothesis Support level. This indicates the degree of support for the oscillation hypothesis in Evidence 1. This indicates the degree of support for the oscillation hypothesis in Evidence 2. This indicates the degree of support between two pieces of evidence for uncertainty. This represents the conflict coefficient.

[0165] In this example, The calculation simplifies to:

[0166] ;

[0167] In the formula, In this binary ( , ) Conflict coefficient under the framework. This indicates the degree of support for oscillation in Evidence 1. This indicates the degree of support for security in Evidence 2. This indicates the degree of support for security in Evidence 1. This indicates the degree of support for oscillation in Evidence 2.

[0168] Step four involves generating quantified risk values. After fusion, a final BPA is obtained. And calculate the confidence interval for the oscillation event. :

[0169] Confidence level (Belief, Bel): ;

[0170] This is the minimum level of confidence required for all evidence to clearly support the occurrence of oscillations.

[0171] Plausibility (Pls): ;

[0172] This represents the highest level of confidence that all evidence does not oppose the occurrence of oscillations.

[0173] This invention selects As the final quantitative risk value This is in Real numbers that fluctuate between 0 and 1.

[0174] It should be noted that the luminescence method is the engineering visualization of the above fusion results. This quantified risk value is directly mapped to a three-color early warning system (on the monitoring interface of the substation) to achieve luminescence alarm:

[0175] Determine whether the quantified risk value exceeds the preset risk threshold:

[0176] Green (safe) ): (Risk threshold 1);

[0177] Yellow (Warning, uncertain): (Risk threshold 2);

[0178] Red (oscillation) ): (Risk threshold 3);

[0179] when When the system determines that an oscillation event has occurred in the new energy power station, it immediately triggers an alarm and uploads data to the regional master station. As a preferred implementation, the step of using a data-enhanced multimodal fusion strategy to perform data enhancement processing on electrical parameters, and fusing transfer learning and generative adversarial networks to construct an oscillation pattern recognition model, and using the oscillation pattern recognition model to identify oscillation event patterns, includes: using data transformation methods to transform electrical parameter data and generate preliminary data-enhanced samples; based on the preliminary data-enhanced samples, simulating disturbances based on Kirchhoff's laws and other circuit laws and electromagnetic principles to generate new electrical parameter data samples; collecting external environmental data related to the new energy power station and fusing the new electrical parameter data samples with the external environmental data to form a multimodal dataset; using transfer learning technology to transfer a pre-trained convolutional neural network model to the oscillation pattern recognition task to obtain the oscillation pattern recognition model; constructing a generative adversarial network model to generate high-quality simulated oscillation data samples, expanding the training dataset; using the multimodal dataset and the expanded training dataset to train the oscillation pattern recognition model, and using the trained oscillation pattern recognition model to identify oscillation event patterns.

[0180] Specifically, the algorithm model optimization training process using a fusion of transfer learning and generative adversarial networks (GANs) includes: 1) Introducing transfer learning into the training of multi-layer convolutional neural networks (CNNs). Given the high cost of acquiring power system oscillation data, a CNN model is pre-trained from a large amount of labeled data from offline electromagnetic transient simulations to learn general time-frequency domain feature extraction patterns. Then, the pre-trained model is transferred to the power system broadband oscillation monitoring task, and the model is fine-tuned using a small amount of local power system data. This approach accelerates model convergence and reduces dependence on large-scale power system data. 2) Enhancing training data based on generative adversarial networks (GANs): To address the scarcity of data for certain rare oscillation patterns in power system oscillation data, a GAN model is constructed. The generator learns the distribution patterns of existing oscillation data and generates simulated oscillation data samples; the discriminator distinguishes between real and generated data. During training, the generator and discriminator compete against each other and work together to optimize. The generated high-quality simulated data is mixed with real data to expand the training dataset. This not only increases the diversity of data but also enables the model to learn richer oscillation features, improving the ability to identify and predict rare oscillation patterns.

[0181] The data-enhanced multimodal fusion strategy involves multimodal data enhancement of collected electrical parameter data (such as voltage, current, and power). In addition to traditional data transformation methods (such as translation, scaling, and rotation), it incorporates the physical characteristics of the power system and simulates disturbances in the data based on circuit laws and electromagnetic principles. Based on Kirchhoff's laws, it simulates the changes in electrical parameters when a component parameter in the circuit changes, generating new training data. The electrical parameter data is then fused with external environmental data (such as temperature, humidity, and wind speed) and used as input for model training. External environmental factors can affect the performance of power equipment and thus be related to oscillations. By fusing multimodal data, the model can learn more comprehensive features, improving the accuracy of oscillation prediction.

[0182] Specifically, the purpose of transfer learning is to address the core problem of scarce power system oscillation data. Since it is impossible to reproduce thousands of different oscillation patterns on a real power grid to train a deep model, transferring a pre-trained convolutional neural network (CNN) model to the oscillation pattern recognition task can be achieved in the following ways:

[0183] Step 1: Source Domain - Pre-training. A power grid model is constructed using high-precision electromagnetic transient simulation software (such as PSCAD, EMTDC). In this model, millions of parameter scans (such as changing wind turbine control parameters, changing line impedance, and setting different faults) generate massive amounts of labeled oscillation data. A one-second oscillation waveform (such as...) is then used to... The time series data is converted into a 2D time-frequency image, for example, using Continuous Wavelet Transform (CWT) or Short-Time Fourier Transform (STFT). This transforms the time series problem into an image recognition problem. A proven and mature CNN architecture validated in image recognition, such as ResNet-50 or VGG-16, is chosen. This CNN model is trained on the aforementioned million simulated time-frequency images (SourceDomain). After training, the model's convolutional layers become a powerful oscillation feature extractor, learning how to identify general textures, edges, and patterns of oscillations.

[0184] Step 2, Target Domain - Transfer and Fine-Tuning: This involves using rare oscillation event data collected by the new energy substation during actual operation (e.g., 500 real oscillation records accumulated by the Northwest Power Grid over the past 5 years). Transfer operation: Loading the pre-trained ResNet-50 model from the source domain. The model's... The weights of each convolutional block (e.g., the first four blocks of ResNet-50) are locked (W_frozen). It is assumed that the basic features learned by these layers (such as frequency and slope) are generalizable in both simulation and reality. The fully connected layer (ClassifierHead) used for classifying the simulation data in the original model is discarded. A new, randomly initialized fully connected layer is added, whose output matches the target domain task (e.g., outputting four oscillation patterns specific to the Northwest Power Grid). Using 500 rare real time-frequency plots, only the unfrozen layers (i.e., the last block of ResNet-50) and the new classifier head are trained (weights updated). This allows the model to leverage its powerful general feature extraction capabilities while fine-tuning its top-level features to adapt to subtle differences in real-world data. Through this process, a high-precision oscillation event pattern recognition model is obtained with only 500 samples, without having to train from scratch.

[0185] It should be noted that, since the data is a time-frequency image (2D image), a Deep Convolutional Generative Adversarial Network (DCGAN) or a more advanced Wasserstein GAN with Gradient Penalty (WGAN-GP) is constructed to generate the oscillating images. This Deep Convolutional Generative Adversarial Network (DCGAN) includes:

[0186] 1. A generator (G) is used to transform a random noise vector (such as a 100-dimensional one) into a multidimensional one. The data is upsampled to a realistic time-frequency map (e.g., 64x64 pixels). The architecture (deconvolution) includes: input... (100-dimensional) Layer 1, fully connected layer, extended to... Layer 2; Transposed convolution (Conv2DTranspose) + BatchNorm + ReLU (upsampled to Layer 3; Transposed convolution (Conv2DTranspose) + BatchNorm + ReLU (upsampled to) Layer 4; Transposed convolution (Conv2DTranspose) + BatchNorm + ReLU (upsampled to) Layer 5; Transposed convolution (Conv2DTranspose) + Tanh (output) (Image).

[0187] 2. Discriminator (D): Receives a time-frequency graph (real or generated by G) and determines its authenticity (outputting a score). Architecture (standard convolution) includes: input... Image layer 1; Convolution (Conv2D) + LeakyReLU (downsampled to) Layer 2; Convolution (Conv2D) + BatchNorm + LeakyReLU (downsampling to) Layer 3; Convolution (Conv2D) + BatchNorm + LeakyReLU (downsampling to) Layer 4; Convolution (Conv2D) + BatchNorm + LeakyReLU (downsampling to) Layer 5; Flatten Fully connected layer Sigmoid (outputs a single probability value).

[0188] 3. Training process (adversarial), used only with 10 real SSO time-frequency plots ( To train: 1. Train D:G to generate 10 fake images ( D Learning Distinction (Label 1) and (Label 0). 2. Training G: G generates 10 fake images ( Freeze the weights of D. The goal of G is to mislead D into believing that... It is true (Label 1). 3. The two are iterated alternately until the "fake" graph generated by G is visually and statistically indistinguishable from the real graph. At this point, G is the generator of high-quality simulated oscillating data samples.

[0189] In addition, before identifying oscillation event patterns using the oscillation pattern recognition model, it checks whether an oscillation event has occurred. When an oscillation event occurs, a red alarm is issued, and the raw electrical parameters (voltage, current, power) for 2 seconds before and after the alarm time (e.g., 0.5 seconds before and 1.5 seconds after) are captured. These 2-second waveforms are then converted into a 64×64 time-frequency graph using CWT (Continuous Wavelet Transform). This graph is then input into a finely tuned ResNet-50 model. The model outputs a probability vector through the Softmax layer of its classification head.

[0190] Specific application examples (ScenarioExample):

[0191] Scenario: A disturbance occurs on the collection line of a wind farm (new energy substation).

[0192] Event identification within 0.1 seconds: The quantization risk value of the fusion luminescence method spiked to 0.85 (red alert). The system determined that an "oscillation event has occurred".

[0193] Pattern recognition is performed in 0.15 seconds: capturing the line active power from 0.0 to 2.0 seconds. Waveform. The waveform is converted into a CWT time-frequency plot (the image shows significant energy concentration at 15Hz and 50Hz). This plot is then fed into a pattern recognition CNN. The CNN outputs a probability vector:

[0194] Subsynchronous oscillation: Mode_1(SSO): 0.92 (92%);

[0195] High-frequency controlled oscillation: Mode_2(HF-Control): 0.05 (5%);

[0196] Low-frequency range oscillation: Mode_3(LF-InterArea): 0.02 (2%)

[0197] Safe / Transient Process: Mode_4 (Safe / Transient): 0.01 (1%);

[0198] A decision is made within 0.2 seconds: the substation system identifies "Mode 1: SSO". This "Mode ID" along with electrical parameters and oscillation characteristics is immediately sent to the regional master station. In a preferred embodiment, the regional master station 2 includes:

[0199] The oscillation data receiving unit is used to receive the electrical parameters and oscillation characteristics of the new energy power station sent by the new energy substation;

[0200] The oscillation data processing unit is used to calculate the oscillation characteristic quantity based on the electrical parameters of the new energy power station sent by the new energy substation, and to determine the oscillation source discrimination condition based on the oscillation characteristic quantity.

[0201] The vibration source determination unit is used to determine the vibration source in the new energy power station based on the auxiliary discrimination conditions of the vibration source;

[0202] The oscillation source action output unit is used to determine whether the oscillation source has diverged and whether the oscillation duration is greater than a preset time threshold. It sorts the comprehensive characteristic values ​​of the new energy power stations participating in the oscillation, sends a generator cut-off command signal to the new energy substation, and calms the oscillation according to the sorting result of the comprehensive characteristic values.

[0203] The generator tripping quantity uploading unit is used to send generator tripping command signals to the new energy substations and simultaneously send generator tripping quantity information to the dispatching center.

[0204] As a preferred embodiment, the step of calculating oscillation characteristic quantities based on the electrical parameters of the new energy power station transmitted by the new energy substation, and determining the oscillation source discrimination conditions based on the oscillation characteristic quantities, includes: normalizing the electrical parameters of the new energy power station using an adaptive normalization algorithm based on data entropy to obtain a normalized dataset; determining the amplitude and start-up time of each device in the new energy power station through a time analysis window based on the normalized dataset, and sorting the amplitude and start-up time of each device to obtain the sorting of the start-up time of each device; and determining the oscillation source discrimination conditions based on the oscillation mode frequencies in the normalized dataset by analyzing the oscillation modes... The frequencies are clustered, and the positive-sequence impedance of each oscillation mode is calculated based on the clustering results to obtain the voltage phasor and current phasor at the oscillation mode frequency. The positive-sequence impedance characteristics of the new energy power station equipment under the oscillation mode are calculated according to the positive-sequence impedance characteristic calculation formula. The apparent power in the normalized dataset is compared with the preset apparent power threshold, and the oscillation amplitude in the normalized dataset is compared with the preset oscillation amplitude threshold to obtain the comparison results. The oscillation start-up time ranking of each device, the positive-sequence impedance characteristics of the new energy power station equipment under the oscillation mode, and the comparison results are used as auxiliary discrimination conditions for oscillation sources.

[0205] The formula for calculating the positive sequence impedance characteristic is:

[0206] ;

[0207] In the formula, This represents the positive sequence impedance under each oscillation mode. Represents a voltage phasor. Represents the current phasor, j represents the imaginary part, and R m X represents the resistance under each oscillation mode. m This represents the reactance under each oscillation mode.

[0208] As a preferred embodiment, the adaptive normalization algorithm based on data entropy normalizes the electrical parameters of the new energy power station to obtain a normalized dataset by: calculating the feature entropy for each feature quantity in the electrical parameters of the new energy power station; determining the scaling factor based on the feature entropy; and using the scaling factor to normalize each electrical parameter to obtain a normalized dataset.

[0209] The formula for calculating the entropy of the feature quantity information is:

[0210] ;

[0211] In the formula, H(x) represents the information entropy of feature x, and p(x) represents the information entropy of feature x. i Let ) represent the probability of the i-th feature x appearing, and n represent the number of feature types. The higher the entropy value, the greater the dispersion of the data. Adaptive scaling is applied to the data based on information entropy. For features with high entropy (large data dispersion), a smaller scaling factor is used; for features with low entropy (relatively concentrated data), a larger scaling factor is used.

[0212] The scaling factor is calculated using the following formula:

[0213] ;

[0214] In the formula, X norm Let represent the scaling factor, x represent a feature quantity in the electrical parameters, min(x) represent the minimum value of the feature quantity in the electrical parameters, max(x) represent the maximum value of the feature quantity in the electrical parameters, H(x) represent the information entropy of the feature quantity x, and e represent a constant. e is a very small constant (e.g., 10). -8 This prevents the denominator from being zero. This approach considers the discrete nature of the data during normalization, avoiding the dominance of highly discrete data in the normalization result, better preserving the differences between different features, reducing data redundancy, and normalizing based on the inherent structure of the data, thus enhancing data distinguishability.

[0215] In a preferred embodiment, the clustering process for the oscillation mode frequencies includes:

[0216] The Z-score method is used to standardize the oscillation mode frequency data to obtain standardized oscillation mode frequency data. Based on the standardized oscillation mode frequency data, the Euclidean distance between each oscillation mode frequency data is calculated using Euclidean distance. The Euclidean distance between each oscillation mode frequency data is used as the input of a clustering algorithm. The density peak clustering algorithm is used to cluster the Euclidean distance between each oscillation mode frequency data to obtain the clustering results.

[0217] The clustering process of Euclidean distance between oscillation mode frequency data using the density peak clustering algorithm includes: constructing a distance matrix based on the Euclidean distance between each oscillation mode frequency data; calculating the local density of each oscillation mode frequency data point in the distance matrix based on the Gaussian kernel function method; calculating the minimum distance between each oscillation mode frequency data point and all points with higher density than it; selecting cluster centers based on local density and minimum distance, and assigning other points to the corresponding clusters according to the density propagation principle to obtain the final clustering result.

[0218] Specifically, when clustering oscillatory modal frequency data, the Density Peak Clustering (DPC) algorithm is introduced to achieve more efficient and accurate clustering results, enhancing the uniqueness of the solution. The optimized clustering steps of the DPC algorithm are as follows: After calculating based on Euclidean distance, the resulting distance matrix is ​​used as input to the DPC algorithm. The DPC algorithm identifies cluster centers based on two key factors: the local density of data points and the distance to high-density points. For each oscillatory modal frequency data point, its local density is calculated. The local density can be calculated using a Gaussian kernel function-based method, calculating the minimum distance between each point and points with higher densities. After determining the cluster centers, other points are assigned to the corresponding clusters according to the density propagation principle. Compared with traditional clustering algorithms, the DPC algorithm does not require pre-setting the number of clusters, and can more accurately discover naturally existing cluster structures in the dataset. For complex and irregular distributions in oscillatory modal frequency data, it can achieve more reasonable clustering, thus providing a more reliable basis for subsequent oscillation source localization and analysis. The regional master station can also locate the oscillation source based on comprehensive criteria, as follows:

[0219] (1) Oscillation source localization and collaborative defense system based on comprehensive criteria;

[0220] In high-proportion renewable energy power grids, broadband oscillation events are often the result of interactions between nearby equipment. The oscillating equipment is distributed within a certain range near the collection station, and multiple stations in areas with strong electrical coupling can detect oscillation events using broadband monitoring devices. However, relying solely on local monitoring devices cannot accurately determine whether the oscillation is limited to a single device, its extent, or which devices contribute significantly to the oscillation, making effective control decisions difficult. Especially in the case of concentrated renewable energy clusters in desert areas, large-scale renewable energy oscillations may occur. Simply and crudely implementing on-site disconnection measures will lead to large-scale disconnection of renewable energy from the grid, significantly impacting the safety of the main power grid. Therefore, it is necessary to comprehensively analyze oscillation event indicators from downstream renewable energy stations near the collection station to determine the oscillation range, oscillation mode, and the relative magnitude of each device or node's participation in the oscillation. Based on this information, correct control decisions can be made for specific unit outputs or feeders.

[0221] The broadband oscillation on-site monitoring and collaborative defense scheme proposed in this invention can promptly obtain complete information on broadband oscillation events with a large impact area within a region, including detailed oscillation analysis results such as the oscillation range, oscillation generator grouping, oscillation interface, and the relative contribution of each device to the oscillation. For forced oscillations, the oscillation source can be located based on the relative magnitude of the transient energy flow output by new energy units or SVG, or the oscillation outflow area can be located based on the transient energy flow direction of the power grid topology. By comparing and analyzing the key characteristic indicators of broadband oscillations at new energy power plants within the region, comprehensive analysis and monitoring of oscillation conditions can be achieved, including online analysis of oscillation modes, generator grouping, source location, and action exit points, supporting comprehensive monitoring and prevention of broadband oscillation events across power plants.

[0222] (2) Oscillation source localization based on comprehensive oscillation characteristic criterion;

[0223] After receiving the oscillation characteristic quantities calculated by the substations, the regional master station uses power oscillation amplitude and oscillation energy as the main criteria for judgment in the oscillation source algorithm, and comprehensively considers multiple oscillation source judgment indicators. The weights of the judgment indicators are summarized in Table 2.

[0224] Table 2 Summary Table of Weights for Oscillation Source Identification Indicators

[0225]

[0226] The specific calculation method for locating the oscillation source is explained below:

[0227] 1) The main station receives information from each substation regarding the start time of oscillation events (emphasizing the actual impact of the oscillation on the system or the time point that triggers a system response), duration, modal frequency, alarm devices, oscillation amplitude, and the maximum and minimum values ​​of modal voltage / current, and performs comprehensive and coordinated defense control. To facilitate comprehensive comparison of equipment of different models and capacities, normalized calculations are used in the calculation of all indicators.

[0228] 2) Amplitude and start-up time criteria: The amplitude and start-up time (the time when the oscillation criterion is started, i.e. the initial moment when the system enters the oscillation state from the steady state) of each device in a specific time analysis window are sorted, and the sorting of the start-up time of each device is used as an auxiliary discrimination condition for the oscillation source.

[0229] The amplitude comparison of each device uses a normalization algorithm to preserve the original distribution characteristics of the data and better reflect the relative magnitude of the amplitude of each device. After the above processing, the amplitude data can be scaled to the range of 0 to 1.

[0230] In the analysis of start-up time, the main concern is the relative order of the equipment. The purpose of normalization is to convert the data into a uniform scale for comparison.

[0231] 3) Impedance Criteria: Monitor the amplitude, phase angle, and frequency of the three-phase harmonic voltage and current of the monitoring equipment. Calculate the frequency and amplitude data of the inter-harmonic oscillations through data spectrum analysis. Cluster the mode frequencies and calculate the positive-sequence impedance of each mode based on the clustering results to obtain the voltage phasor at the oscillation mode frequency. and current phasor The positive sequence impedance characteristic of the converter in this mode is calculated. When R m X m An alarm will be triggered if the given threshold is exceeded.

[0232] When R m <0 indicates that the component is the active power source for this mode of oscillation;

[0233] When X m <0 indicates that the component is the reactive power source for this mode of oscillation.

[0234] Because phase calculation requires the Prony algorithm, which requires a certain delay, it is not real-time linear and is only used as an auxiliary judgment to give the device / site with the smallest absolute value of negative impedance in rounds.

[0235] The steps involved in clustering modal frequencies are as follows:

[0236] Step 1: To avoid the influence of data with different dimensions, the frequency data needs to be standardized using the Z-score method.

[0237] ;

[0238] In the formula, x i σ represents the i-th frequency value; μ represents the average frequency; σ represents the standard deviation.

[0239] Step 2: Calculate the distance between each modal frequency using Euclidean distance as input for the clustering algorithm;

[0240] ;

[0241] In the formula, d(x,y) represents the Euclidean distance between each modal frequency, and x and y represent the eigenvectors of the modal frequencies;

[0242] Step 3: Select a suitable clustering method based on the data distribution characteristics and actual needs, and input the preprocessed modal frequency data into the selected clustering algorithm. Based on the clustering results, analyze the modal frequencies in each cluster to confirm whether they conform to the characteristics of oscillation modes.

[0243] 4) Condition 1 for a device to participate in the oscillation source calculation: Apparent power is greater than a given threshold; Condition 2 for a device to participate in the oscillation source calculation: Oscillation amplitude is greater than a given threshold.

[0244] 5) Provide the criteria for identifying an oscillation source: if the oscillation amplitude of the device is the largest and the ratio of its amplitude to that of the second smallest is greater than a given threshold, the oscillation source alarm will be activated.

[0245] 6) Provide the following exit conditions for the oscillation source action: The oscillation source diverges (monitoring and acquiring oscillation source divergence relies on in-depth analysis of changes in oscillation characteristic quantities, especially the evolution of oscillation amplitude and frequency. The system uses algorithms to judge the changing trend of oscillation modes and identify whether divergence behavior exists), and the oscillation duration (the time window during which the oscillation source is monitored and continues to exist in the system) is greater than a given threshold TWFO. The oscillation source alarm action is activated when the oscillation source's comprehensive criterion weight ranking is high and the difference between it and the second lowest value is greater than a certain threshold. An exit signal is sent to the substation. The comprehensive characteristic values ​​of the participating new energy power stations are sorted, and the station with the highest ranking is removed in the first round. If the oscillation does not subside, the station (or multiple stations) with the highest comprehensive characteristic value at that time is removed in the second round until the oscillation subsides.

[0246] The comprehensive characteristic value of a new energy power station may be obtained by weighting and summing multiple monitoring indicators (such as voltage, current, frequency, oscillation amplitude, etc.). Each characteristic value may represent a certain aspect of oscillation characteristics, while the comprehensive characteristic value is formed by combining the influence of multiple indicators through a certain weighting method to form an overall assessment.

[0247] 7) Based on the oscillation energy reported by each substation, oscillation source identification is performed based on the oscillation energy flow. Since the calculation of the oscillation energy flow requires a certain delay, this criterion is also used as an auxiliary criterion and can be used for oscillation source analysis. The specific analysis process decision tree is as follows: Figure 10 As shown. Specifically:

[0248] The oscillation modes and transient energy integral values ​​of the unit output and power injection of the synchronous power grid are calculated, and the amplitude and energy values ​​are analyzed:

[0249] When only one unit (or multiple units in the same power plant) meets the oscillation alarm threshold, and the transient energy flow value output by that unit (or the power plant) is the largest, it is judged as forced oscillation, and that unit (or the power plant) is the source of forced oscillation.

[0250] When multiple units meet the oscillation alarm threshold:

[0251] If one of the generating units (or multiple generating units in the same power plant) outputs the largest transient energy flow value, while the transient energy flow values ​​of other generating units are relatively small or negative, it is determined to be a forced oscillation combined with unit resonance, and then that generating unit (or that power plant) is the source of forced oscillation.

[0252] If there is no power plant with a significant transient energy flow output, but a certain substation outputs the largest transient energy flow value, while other units output relatively small transient energy flow values, or negative values, it is judged to be a combination of forced oscillation and unit resonance. In this case, the oscillation source unit is located in the downstream power grid of that substation.

[0253] If the transient energy integral of the injected power from the generating unit and the substation does not show obvious differentiation characteristics and no oscillation source is found, it is judged to be a weakly damped oscillation.

[0254] Oscillatory energy flow analysis is primarily used for locating forced oscillation sources. Most oscillations occurring in power grids are forced oscillations with clearly defined external disturbance sources. When the frequency of this external forced oscillation source happens to match the inherent oscillation mode of the power grid, resonance or harmonic resonance may be triggered, leading to generator clustering and large-scale power fluctuations. This resonance phenomenon can cause continuous grid oscillations, exhibiting characteristics similar to weakly damped oscillations. When analyzing and identifying the oscillation modes of the power grid, a reasonable strategy is to prioritize determining the presence of an external forced oscillation source and then take appropriate measures based on this identification. Furthermore, by analyzing the energy flow of the power grid, utilizing its topology and energy transmission characteristics, and by observing the energy flow and changes within the grid, the specific divergence area of ​​the oscillation energy can be traced, such as a specific regional power grid or a particular transformer, which can also be identified as forced oscillation. When multiple generators oscillate in the same mode and the power grid experiences large-scale power oscillations, and a specific oscillation source or oscillation region cannot be located, it can be considered that a weakly damped oscillation has occurred.

[0255] Typical case data analysis, such as Figure 11-12 The diagram shown is a schematic of a power grid in a typical wind power aggregation area.

[0256] Harmonic source analysis was performed based on continuous waveform data recorded by broadband measurement devices at all nodes during the subsynchronous oscillation period. The original sampled values ​​of three-phase voltage and current at each node were used for feature extraction and modal clustering to obtain information such as the dominant mode frequency, amplitude, phase angle, and start-up time. Modal impedance, complex power, and oscillation energy flow characteristics of each node were calculated. Based on the comprehensive oscillation characteristic values, the Naomohu wind farm had the largest comprehensive oscillation characteristic value. The dominant mode power flow was plotted based on the impedance and complex power calculation results, as shown below. Figure 11 As shown by the arrows, the modal power flow flows from the wind farm to the power grid, which also confirms that the oscillation source is the wind farm, thus verifying the effectiveness and accuracy of the proposed broadband oscillation source comprehensive criterion.

[0257] It should be noted that the above-mentioned oscillation source location is one implementation method. This invention can also use the vibration mode identified by the oscillation identification model for oscillation source location. Specifically, the regional master station receives alarms from 5 substations. Three of them report the oscillation mode as "Mode_1 (SSO)", and two report the oscillation mode as "Mode_2 (HF-Control)". The regional master station immediately determines that this is not a single oscillation, but a "mixed-mode oscillation" with concurrent "SSO" and "HF". Therefore, a source tracing algorithm (such as impedance analysis) will automatically group the oscillations.

[0258] Group 1 (SSO): Calculate the negative damping source in SSO mode using only the data from those 3 substations.

[0259] Group 2 (HF): Using only the data from those two substations, calculate the source of control instability in HF mode.

[0260] This avoids the major flaw of traditional methods that mix data from five stations together to calculate an erroneous average source.

[0261] In a preferred embodiment, the dispatch master station 3 includes: a power system monitoring unit, a generator tripping quantity receiving unit, a safety verification unit, and an instruction sending unit;

[0262] The power system monitoring unit is used to monitor the power system frequency and voltage level within a preset period.

[0263] The machine switching quantity receiving unit is used to receive machine switching quantity information sent by the regional master station 2;

[0264] The safety verification unit is used to perform safety verification by combining power system frequency, voltage level, generator tripping information, oscillation mode and power system operation status. When the quantity and range of generator tripping exceed the system frequency and voltage support regulation capacity, it generates a regional master station blocking command.

[0265] It should be noted that, based on the identified oscillation event patterns, the regional master station can understand the extent of the impact of oscillations on different new energy power generation units. Different oscillation event patterns have different oscillation intensities and development trends. The regional master station can dynamically assess the severity and direction of oscillations based on the identified patterns, thereby dynamically adjusting the amount of power generation offloaded.

[0266] Specifically, for example: the regional master station reports a generator tripping request, Mode_ID="HF-Control" (high-frequency oscillation). The dispatch master station determines that the high-frequency oscillation has a small propagation range and will not cause system-level power angle instability. The safety check passes, but the generator tripping amount is strictly based on 2.1 million kilowatts. Another example: the regional master station reports a generator tripping request, Mode_ID="LF-InterArea" (low-frequency inter-area oscillation). The dispatch master station determines that the low-frequency inter-area oscillation is the "cancer" of system-level instability. It immediately blocks any generator tripping behavior from the regional master station (because disconnecting the unit may worsen power angle stability). The dispatch master station immediately activates the Wide Area Damping Controller (WADC) to provide inter-area damping for this 0.5Hz oscillation by (for example) adjusting the transmission power of the ultra-high voltage direct current (HVDC).

[0267] The instruction sending unit is used to send the generated area master station blocking instruction to area master station 2.

[0268] Specifically, the security verification method for the dispatch master station can be as follows:

[0269] The system integrates frequency, voltage level, generator tripping volume of each master station, and system operation status to conduct system safety verification. When problems such as excessive total generator tripping volume or wide range in a short period of time exceed the system frequency and voltage support regulation capacity occur, such as when the maximum power imbalance of the Northwest Power Grid is 2.1 million kilowatts, and the generator tripping volume issued by each regional master station exceeds 2.1 million kilowatts within a specific time interval, a certain periodic blocking command is issued to the relevant regional master stations. After the frequency is restored, the blocking command is stopped, and the regional master stations continue to implement oscillation defense control to achieve effective prevention and control of broadband oscillations on the basis of ensuring the safety of the large power grid.

[0270] It should be further explained that the functional architecture of the defense system of this invention is as follows: the overall control system is divided into three main parts: the dispatch center, the regional master station, and the new energy substation, such as... Figure 13-14 As shown, the functions of each part are briefly described below:

[0271] Central Dispatch Station: 1) Deployed at the dispatch terminal, it can monitor the system frequency and voltage level in real time over a certain period, as well as the generator tripping information received from each regional master station. 2) Combining system frequency, voltage level, generator tripping volume from each master station, and system operation status, it conducts system safety verification. When problems such as excessive total generator tripping volume or wide range exceeding the system frequency and voltage support regulation capacity occur in a short period, it issues blocking commands to the relevant regional master stations to effectively prevent broadband oscillations while ensuring the safety of the large power grid.

[0272] Regional Master Station: 1) Since broadband oscillations in new power systems are mainly caused by the interaction between power electronic equipment such as wind and solar power, DC power, and the power grid, the propagation range of oscillation events is regional. Therefore, regional master stations are set up in relevant areas with concentrated access to new energy sources, such as important 750kV substations in areas with concentrated access to new energy sources like Hami, Dunhuang, and Xiazhou, or important 220kV new energy collection stations. 2) Organize the research and development of broadband oscillation monitoring and source tracing technology. Based on the key oscillation characteristics transmitted by each substation, the oscillation source can be located and control measures implemented in the shortest possible time. 3) The regional master station has the ability to receive various oscillation characteristics sent by each substation and perform online oscillation monitoring and analysis. It can identify the oscillation source information within 1 second after the oscillation occurs based on the comprehensive comparison of oscillation characteristics, select the generator to switch based on the oscillation source location, and issue generator switching instructions to the new energy substation.

[0273] Before issuing the tripping command, the tripping quantity is first sent to the dispatching center for safety verification. If no blocking command is received from the dispatching center after a certain period of time, the tripping is directly sent to the substation, providing a means for timely and effective handling of broadband oscillation problems.

[0274] New Energy Substation: 1) Deployed at new energy power stations, equipped with an online oscillation monitoring and analysis device with high-precision sampling and millisecond-level computing capabilities (upgraded using a stability control or broadband monitoring device, or a new device can be installed); 2) Can perform broadband oscillation on-site monitoring and analysis, and calculate oscillation characteristics such as oscillation amplitude, start-up time, spectrum, and transient energy in real time by monitoring and analyzing data such as voltage, current, and oscillation frequency of wind farm or photovoltaic power station generating units or feeders, and send relevant signals to the main station when the action threshold is reached; 3) Receives and executes the generator tripping command issued by the regional main station.

[0275] In summary, by utilizing the above-mentioned technical solutions of this invention, this invention monitors and calculates data such as voltage, current, and oscillation frequency of wind farms or photovoltaic power station generating units or feeders, and calculates oscillation characteristic quantities such as oscillation amplitude, oscillation start time, spectrum, and transient energy in real time. It also adds indicators such as voltage active and reactive power sensitivity for the first time to improve the accuracy of oscillation source analysis and location. The main station in the area of ​​the new energy centralized access hub has the capability to receive various oscillation characteristics sent by each substation and perform oscillation source location, enabling online oscillation monitoring and analysis. Based on a comprehensive comparison of oscillation characteristics, it can accurately identify the oscillation source and key participating stations within seconds of the oscillation occurring. Based on the oscillation source location, it selects which generators to trip and issues tripping commands to the new energy substations. A main station is built in the dispatch center to conduct system safety checks based on system frequency, voltage level, the number of generators tripped by each main station, and system operation. When problems such as excessive total generator tripping or a wide range exceeding the system frequency and voltage support regulation capacity occur in a short period, a blocking command is issued to the relevant regional main stations, achieving effective prevention and control of broadband oscillations while ensuring the safety of the large power grid. To effectively address the impact of broadband oscillations on the power system, this invention further designs a collaborative defense mechanism. This mechanism not only relies on the monitoring and control of individual stations, but also includes the coordinated cooperation between various new energy power plants. By establishing a regional master station to receive and integrate real-time monitoring data from various substations, the system can quickly locate the oscillation source and, based on the characteristics and impact range of the oscillation source, rationally schedule various equipment within the power system and take coordinated defense measures.

[0276] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, optical storage, etc.) containing computer-usable program code.

[0277] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A broadband oscillation local monitoring and collaborative defense system, characterized in that, include: The new energy substation is used to acquire the electrical parameters of the new energy power station and calculate the oscillation characteristic of the new energy power station based on the acquired electrical parameters. When the oscillation characteristic reaches the preset oscillation threshold, the electrical parameters and oscillation characteristic are sent to the regional master station, and the machine switching command issued by the regional master station is received and executed. The regional master station is used to receive electrical parameters and oscillation characteristics sent by the new energy substations and to locate the oscillation source. Based on the oscillation source location results, select the generator to switch over, and based on the generator selection structure, issue a generator switching command to the new energy substation; The dispatch master station is used to monitor the power system frequency, voltage level and the amount of generator tripping information of each regional master station in real time, and to perform safety verification based on the power system frequency, voltage level and the amount of generator tripping information of each regional master station, and to carry out power coordination defense based on the verification results; The regional master station includes: an oscillation data processing unit, used to calculate oscillation characteristic quantities based on the electrical parameters of the new energy power stations sent by the new energy substations, and to determine the oscillation source discrimination conditions based on the oscillation characteristic quantities; The process of calculating oscillation characteristic quantities based on the electrical parameters of the new energy power station sent by the new energy substation, and determining the oscillation source discrimination conditions based on the oscillation characteristic quantities, includes: An adaptive normalization algorithm based on data entropy is used to normalize the electrical parameters of new energy power stations to obtain a normalized dataset. Based on the normalized dataset, the amplitude and start-up time of each device in the new energy power station are determined through a time analysis window, and the amplitude and start-up time of each device in the new energy power station are sorted to obtain the order of start-up time of each device. Based on the normalized dataset of centralized oscillation mode frequencies, the oscillation mode frequencies are clustered, and the positive-sequence impedance of each oscillation mode is calculated based on the clustering results to obtain the voltage phasor and current phasor at the oscillation mode frequencies. Based on the positive-sequence impedance characteristic calculation formula, the positive-sequence impedance characteristics of the new energy power station equipment under the oscillation modes are calculated. The apparent power in the normalized dataset is compared with a preset apparent power threshold, and the oscillation amplitude in the normalized dataset is compared with a preset oscillation amplitude threshold to obtain the comparison results. The order of oscillation start-up time of each device, the positive sequence impedance characteristics of new energy power station equipment in oscillation mode and the comparison results are used as auxiliary discrimination conditions for oscillation sources. The adaptive normalization algorithm based on data entropy normalizes the electrical parameters of the new energy power station, resulting in a normalized dataset including: For each characteristic quantity in the electrical parameters of a new energy power station, calculate the characteristic quantity information entropy; The scaling factor is determined based on the information entropy of the feature quantity, and each electrical parameter is normalized using the scaling factor to obtain a normalized dataset. Specifically, for features with high entropy, i.e., large data dispersion, a small scaling factor is used; for features with low entropy, i.e., concentrated data, a large scaling factor is used.

2. The broadband oscillation local monitoring and collaborative defense system according to claim 1, characterized in that, The new energy substations include: The data acquisition and analysis unit is used to acquire electrical parameters of new energy power plants and identify oscillation events and oscillation event patterns in new energy power plants by analyzing the development trend and distribution of electrical parameters. The regional master station transmitting unit is used to send the electrical parameters of the new energy power station to the regional master station after oscillations occur in the new energy power station; The receiving and execution unit is used to receive and execute the switching instructions issued by the regional master station.

3. The broadband oscillation local monitoring and collaborative defense system according to claim 2, characterized in that, The data acquisition and analysis unit includes: The data monitoring unit is used to monitor and analyze the electrical parameters of wind farms, photovoltaic power station power generation units and feeders in real time using an oscillation online monitoring and analysis device based on time-division multiplexing and adaptive sampling frequency technology. The electrical parameters include voltage, current, active power, reactive power and apparent power. The window feature factor extraction unit is used to analyze the voltage-current angle and power curve fluctuation characteristics based on the electrical parameters of the wind farm, photovoltaic power station power generation unit and feeder, and to extract window feature factors using data statistics and extrapolation methods to form oscillation risk assessment elements. The oscillation event identification unit is used to identify oscillation events based on oscillation risk assessment elements, through data statistics and trend analysis, and by combining multi-element weighting and fusion luminescence method. The oscillation event pattern recognition unit is used to perform data augmentation processing on electrical parameters through a multimodal fusion strategy of data augmentation, and to build an oscillation pattern recognition model by fusing transfer learning and generative adversarial networks, and to identify oscillation event patterns using the oscillation pattern recognition model.

4. A broadband oscillation local monitoring and collaborative defense system according to claim 3, characterized in that, The method analyzes the voltage-current angle and power curve fluctuation characteristics based on the electrical parameters of wind farms, photovoltaic power station generating units and feeders, and extracts window characteristic factors using data statistics and extrapolation methods to form oscillation risk assessment elements, including: Based on the electrical parameters of the wind farm, photovoltaic power station power generation units and feeders, determine the phase angle difference between voltage phase angle and current phase angle, and analyze the changing trend of phase angle difference; Fluctuation analysis is performed on the curves of active power, reactive power, and apparent power in the electrical parameters of wind farms, photovoltaic power station generating units, and feeders to identify the fluctuation characteristics of the power curves. Based on a pre-set sliding window length, variable window length disturbance analysis is performed on the electrical parameters of wind farms, photovoltaic power station power generation units and feeders to identify the fluctuation characteristics of power disturbance within the window period. Based on the changing trend of phase angle difference, the curve fluctuation characteristics of power, and the fluctuation characteristics of power disturbance, window feature factors are extracted using data statistical methods, and these extracted window feature factors are used as elements for oscillation risk assessment.

5. A broadband oscillation local monitoring and collaborative defense system according to claim 3, characterized in that, The method for identifying oscillation events based on oscillation risk assessment factors, through data statistics and trend analysis, and combined with a multi-factor weighted fusion luminescence method, includes: Based on the elements of oscillation risk assessment, the importance of each element to the oscillation risk assessment is determined, and according to the determination results, a corresponding weight is assigned to each element. The multi-factor weighting fusion light emission method is adopted to fuse the values ​​of each factor with their corresponding weights and present the fusion result in numerical form to form a quantitative risk value. Determine whether the quantified risk value exceeds the preset risk threshold. If it does, it indicates that an oscillation event has occurred in the new energy power station; otherwise, it indicates that no oscillation event has occurred in the new energy power station.

6. A broadband oscillation local monitoring and collaborative defense system according to claim 3, characterized in that, The process involves data augmentation of electrical parameters using a multimodal fusion strategy, followed by the integration of transfer learning and generative adversarial networks to construct an oscillation pattern recognition model. The model is then used to identify oscillation event patterns. The electrical parameter data is transformed using data transformation methods to generate preliminary data augmentation samples. Based on the preliminary data augmentation samples, simulated disturbances are performed based on Kirchhoff's laws and electromagnetic principles to generate new electrical parameter data samples. Collect external environmental data related to new energy power stations, and fuse new electrical parameter data samples with external environmental data to form a multimodal dataset; By using transfer learning techniques, a pre-trained convolutional neural network model is transferred to the oscillation pattern recognition task to obtain an oscillation pattern recognition model. A generative adversarial network model is constructed to generate high-quality simulated oscillation data samples. The training dataset is expanded, and the oscillation pattern recognition model is trained using the multimodal dataset and the expanded training dataset. The trained oscillation pattern recognition model is then used to identify oscillation event patterns.

7. A broadband oscillation local monitoring and collaborative defense system according to claim 1, characterized in that, The regional main station also includes: The oscillation data receiving unit is used to receive the electrical parameters and oscillation characteristics of the new energy power station sent by the new energy substation; The vibration source determination unit is used to determine the vibration source in the new energy power station based on the auxiliary discrimination conditions of the vibration source; The oscillation source action output unit is used to determine whether the oscillation source diverges and whether the oscillation duration exceeds a preset time threshold. It also sorts the comprehensive characteristic values ​​of the new energy power stations participating in the oscillation, sends a generator trip command signal to the new energy substation, and calms the oscillation based on the sorting result of the comprehensive characteristic values. The generator tripping quantity uploading unit is used to send generator tripping command signals to the new energy substations and simultaneously send generator tripping quantity information to the dispatching center.

8. A broadband oscillation local monitoring and collaborative defense system according to claim 7, characterized in that, The clustering process for the oscillation mode frequencies includes: The Z-score method is used to standardize the oscillation mode frequency data to obtain standardized oscillation mode frequency data. Based on the standardized oscillation mode frequency data, the Euclidean distance between the frequency data of each oscillation mode is calculated using Euclidean distance. The Euclidean distance between the frequency data of each oscillation mode is used as the input of the clustering algorithm. The density peak clustering algorithm is used to cluster the Euclidean distance between the frequency data of each oscillation mode to obtain the clustering result. The step of clustering the Euclidean distance between the frequency data of each oscillation mode using the density peak clustering algorithm includes: constructing a distance matrix based on the Euclidean distance between the frequency data of each oscillation mode; For each oscillation mode frequency data point in the distance matrix, its local density is calculated based on the Gaussian kernel function method; For each oscillation mode frequency data point, calculate the minimum distance between it and all points with higher density than it; Cluster centers are selected based on local density and minimum distance, and other points are assigned to corresponding clusters according to the density propagation principle to obtain the final clustering result.

Images