Wind turbine voltage sag compensation method and system based on deep learning

Abstract

The application discloses a wind turbine voltage sag compensation method and system based on deep learning, and relates to the technical field of low-voltage ride-through control, which comprises the following steps: collecting wind turbine three-phase voltage output and operation state data in real time; judging whether a voltage sag occurs, and if so, extracting voltage sag element features; combining the unit operation state data and the voltage sag element features to judge and generate trigger source information; performing compensation analysis according to a voltage sag compensator to output a compensation decision; and performing voltage sag compensation control. The application solves the technical problem that the traditional voltage sag compensation technology cannot accurately identify the trigger source and element features of the wind turbine voltage sag, the compensation decision lacks pertinence, the control and compensation effect of voltage stability in the circuit system are poor, and the demand cannot be met, and achieves the technical effect of accurately identifying the trigger source and element features of the wind turbine voltage sag and effectively improving the control and compensation of voltage stability.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of low-voltage ride-through control, in particular to a wind turbine voltage sag compensation method and system based on deep learning. BACKGROUND

[0002] During the operation of a wind turbine, voltage sag phenomena occur from time to time, which significantly affect the stable operation of the circuit system. In the prior art, the compensation for wind turbine voltage sag is mostly achieved by using traditional circuit regulation methods, which have played a certain role in relatively stable power grid environments or simple operation scenarios. However, with the expansion of wind power grid connection scale, the power grid environment is becoming increasingly complex, and these traditional compensation technologies have exposed limitations in application: due to the inability to accurately identify the trigger source and element characteristics of voltage sag, the compensation decision lacks pertinence, resulting in poor compensation effect, which is difficult to meet the control and compensation requirements of the power supply / distribution circuit system for voltage stability. SUMMARY

[0003] The application provides a wind turbine voltage sag compensation method and system based on deep learning, which is used to solve the technical problem that traditional voltage sag compensation technologies cannot accurately identify the trigger source and element characteristics of wind turbine voltage sag, the compensation decision lacks pertinence, resulting in poor control and compensation effect of voltage stability in the circuit system, and it is difficult to meet the requirements.

[0004] In a first aspect, the application provides a wind turbine voltage sag compensation method based on deep learning, which comprises: collecting three-phase voltage output data and unit operation state data of a wind turbine in real time; determining whether a voltage sag behavior occurs based on the three-phase voltage output data, and if so, extracting sag elements to generate voltage sag element characteristics; judging fault triggering and fluctuation triggering based on the unit operation state data and the voltage sag element characteristics to generate trigger source information; performing compensation analysis by a voltage sag compensator according to the trigger source information and the voltage sag element characteristics, and outputting a compensation decision; and performing voltage sag compensation control according to the compensation decision.

[0005] In a second aspect of the present application, a wind turbine voltage sag compensation system based on deep learning is provided, which comprises: a wind turbine data acquisition module for acquiring real-time three-phase voltage output data and unit operating state data of a wind turbine; a voltage sag behavior judgment module for judging whether a voltage sag behavior occurs based on the three-phase voltage output data, and if so, extracting a sag element to generate a voltage sag element feature; a trigger source information acquisition module for judging fault triggering and fluctuation triggering based on the unit operating state data and the voltage sag element feature to generate trigger source information; a compensation decision output module for performing compensation analysis by a voltage sag compensator according to the trigger source information and the voltage sag element feature to output a compensation decision; and a voltage sag compensation control execution module for performing voltage sag compensation control according to the compensation decision.

[0006] The one or more technical solutions provided in the present application have at least the following technical effects or advantages:

[0007] In the present application, real-time three-phase voltage output data and unit operating state data of a wind turbine are acquired, voltage sag related information is obtained through sag element extraction, trigger source judgment and other processes, a compensation decision is generated by a voltage sag compensator in combination with trigger source information and sag element features, and then global collaborative analysis and local optimization are performed for adjustment, so as to accurately compensate for voltage sag of a wind turbine, so that the control and compensation effect of voltage stability in a power supply / distribution circuit system is more optimal, the stable operation demand of a wind turbine is met, and the technical effect of accurately identifying the trigger source and element feature of voltage sag of a wind turbine is achieved, and the control and compensation of voltage stability are effectively improved. BRIEF DESCRIPTION OF DRAWINGS

[0008] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0009] Figure 1 is a flowchart of a wind turbine voltage sag compensation method based on deep learning provided by the embodiments of the present application.

[0010] Figure 2 is a structural schematic diagram of a wind turbine voltage sag compensation system based on deep learning provided by the embodiments of the present application.

[0011] Legend of the drawings: wind turbine data acquisition module 1, voltage sag behavior judgment module 2, trigger source information acquisition module 3, compensation decision output module 4, voltage sag compensation control execution module 5. DETAILED DESCRIPTION

[0012] The application provides a wind turbine voltage sag compensation method and system based on deep learning, which is used to solve the technical problem that the traditional voltage sag compensation technology cannot accurately identify the trigger source and element characteristics of the wind turbine voltage sag, the compensation decision lacks pertinence, the control and compensation effect of voltage stability in the circuit system is poor, and the demand cannot be met.

[0013] The technical solutions in the embodiments of the application will be clearly and completely described below with reference to the drawings in the embodiments of the application. Obviously, the described embodiments are only part of the embodiments of the application, not all the embodiments. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the application.

[0014] It should be noted that the terms "first", "second", etc. in the specification and the above drawings of the application are used to distinguish similar objects, and do not necessarily mean a specific order or sequence. It should be understood that the data thus used can be interchanged under appropriate circumstances, so that the embodiments of the application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "include" and "have" and any variations thereof are intended to cover non-exclusive inclusion, for example, a process, method, system, product or server including a series of steps or units does not have to be limited to those steps or units clearly listed, but can include other steps or modules that are not clearly listed or inherent to these processes, methods, products or devices.

[0015] Embodiment one, as shown in the figure, a wind turbine voltage sag compensation method based on deep learning, wherein the method comprises: Figure 1

[0016] Step A100: Real-time acquisition of three-phase voltage output data and unit operation state data of the wind turbine.

[0017] In the embodiments of the application, the three-phase voltage is the voltage of the three-phase alternating current circuit used in the power generation and power transmission process of the wind turbine.

[0018] Specifically, the real-time acquisition of the three-phase voltage output data of the wind turbine is mainly realized by deploying high-precision voltage sensors at the output end and grid-connected interface of the wind turbine. These sensors need to cover the A, B and C three-phase circuits, real-time capture the instantaneous value (instantaneous voltage fluctuation data) and effective value (root mean square value) of each phase voltage, and record the accurate time stamp of each acquisition synchronously.

[0019] ​The collection of the operating state data of the unit involves various special sensors and data interfaces. The wind speed data are collected in real time by a wind speed meter installed on the top of the cabin, which records the instantaneous wind speed and average wind speed once per second, and is used to assist in determining whether the voltage sag is triggered due to wind speed fluctuation; the blade speed and the generator speed are obtained by deploying speed sensors at the blade hub and the shaft end of the generator, which record the speed change data per minute in real time, reflecting the mechanical operating state of the unit; the reactive power and active power are collected by the power sensor at the output end of the generator, which is updated once every 0.5 seconds, and is used to evaluate the power output stability of the unit.

[0020] The working temperature of the generator, the transformer, the inverter and other equipment is collected by the temperature sensor attached to the surface of the equipment, which monitors whether the equipment is affected by the voltage output due to overheating in real time; the internal voltage, current and frequency data of the wind turbine are collected synchronously by the current sensor and the frequency meter in the control cabinet of the unit, which generates a set of data every 0.1 seconds, reflecting the internal circuit state of the unit; the start / stop state of the unit is recorded by the state monitoring module in the control cabinet, which is updated in real time with the on-off signal starting as 1 and stopping as 0; the operating data of the connected power grid are obtained by the special data acquisition module of the grid connection interface, including the grid voltage, frequency and fault signal, which is synchronized once every 0.1 seconds, and is used to determine whether the voltage sag is triggered due to power grid failure.

[0021] All the operating state data of the unit are attached with the time stamp consistent with the three-phase voltage output data, ensuring the matching of different types of data in time sequence, and providing a complete state background for subsequent trigger source judgment.

[0022] Step A200: judging whether a voltage sag behavior occurs based on the three-phase voltage output data, if yes, performing sag element extraction to generate voltage sag element features.

[0023] In the embodiment of the application, the voltage sag is a phenomenon that the output voltage of the wind turbine temporarily decreases due to power grid failure, short circuit, load fluctuation and other reasons.

[0024] Optionally, based on the three-phase voltage output data collected in the foregoing, first, the core threshold for voltage sag judgment is set: 80% of the rated voltage is taken as the voltage amplitude threshold, and 10 milliseconds is taken as the duration threshold. The real-time three-phase voltage effective value is continuously monitored by the data processing module, and each phase voltage is compared with the amplitude threshold.

[0025] Then, when it is monitored that the effective value of any one phase voltage is first lower than the voltage amplitude threshold, the duration timing is started, and the time stamp of this moment is recorded synchronously. If the voltage of this phase continues to be lower than the voltage amplitude threshold within the next 10 milliseconds, and no fluctuation that restores to above the threshold occurs, it is preliminarily determined that a voltage sag may occur; if the voltage rises above the threshold during this period, it is determined as transient fluctuation, and the timing is terminated.

[0026] To avoid single phase misjudgment, the system will check other two phase voltage states at the same time. If only one phase voltage is abnormal and the duration does not reach the threshold value, it may be sensor failure or local interference; if at least one phase voltage meets the amplitude lower than the threshold value and the duration exceeds 10 milliseconds, the voltage sag behavior is finally confirmed to occur, and then the subsequent voltage sag element feature extraction process is triggered.

[0027] Finally, the voltage drop amplitude, phase identified by Fourier transform, and duration are calculated based on the three-phase voltage output data, and the three are used to construct the voltage sag element feature, which is described in detail in steps A210-A240.

[0028] By setting double thresholds of voltage amplitude and duration, combined with continuous monitoring and multi-dimensional verification of high-frequency collected three-phase voltage data, it can accurately determine whether the wind turbine generator has voltage sag, and provide reliable basis for subsequent voltage sag element analysis and compensation control.

[0029] Step A300: Based on the unit operating state data and the voltage sag element feature, fault trigger and fluctuation trigger judgment is performed to generate trigger source information.

[0030] In an embodiment of the present application, by constructing a voltage sag judgment sample library, the samples and the unit operating state data, the voltage sag element feature are input into the twin network for similarity identification, the corresponding historical trigger factors are obtained according to the qualified similarity, and the trigger source information is generated, which is described in detail in A310-A330.

[0031] Among them, the fault trigger is the voltage sag caused by the internal fault of the wind turbine generator or the power grid, and the fluctuation trigger is the voltage sag caused by the wind speed fluctuation, which is described in detail in A340.

[0032] Step A400: According to the trigger source information and the voltage sag element feature, compensation analysis is performed by the voltage sag compensator to output compensation decision.

[0033] In the embodiment of the present application, the voltage sag compensator is a device constructed by constructing an integrated twin unit model containing the wind turbine generator itself, the connected power grid and the weather conditions, loading the historical voltage sag recovery record set to learn the relationship between the voltage compensation parameters and the voltage sag elements and the trigger source.

[0034] Specifically, by constructing a twin unit model containing the unit, the power grid and the weather, the voltage sag compensator is trained by historical records, and after inputting the trigger source and the sag element feature, the model is verified to output the compensation decision, which is described in detail in A410-A440.

[0035] Step A500: Perform voltage sag compensation control according to the compensation decision.

[0036] Specifically, when performing voltage sag compensation control according to the compensation decision, the generated compensation decision parameters are first converted into specific control instructions. The compensation decision parameters include reactive power compensation amount, converter modulation ratio, compensation execution time, etc. These parameters will be issued to the corresponding execution equipment, such as converters and static var generators (SVGs), in the form of digital signals through the control system bus of the wind turbine. For example, after receiving the instruction of a modulation ratio of 0.75, the converter will adjust the switching frequency and conduction angle of the internal power module to change the amplitude and phase of the output voltage. According to the instruction of 120 kvar, the SVG will inject specified reactive power into the grid by controlling the bridge arm switching state to support voltage recovery.

[0037] During execution, the system will collect wind turbine output voltage data in real time, updating every 10 ms to monitor the compensation effect. If the compensation decision is aimed at voltage sag caused by grid faults, the voltage recovery speed needs to be focused on, requiring the voltage to recover from the sag within 200 ms. If it is aimed at sag triggered by wind speed fluctuations, the voltage fluctuation amplitude after recovery needs to be controlled to ensure that it is within ±3% of the rated voltage. When it is monitored that the voltage does not recover as expected, it will be immediately fed back to the control system to temporarily increase the compensation amount, such as additional reactive power compensation, until the voltage meets the standard.

[0038] At the same time, the control process needs to be coordinated with other protection mechanisms of the wind turbine. For example, when the generator speed exceeds the safe range during compensation, such as exceeding 1800 r / min, the compensation strength will be temporarily reduced, and the blade angle will be adjusted through the variable pitch system to stabilize the speed. After the speed is restored to below 1500 r / min, the original compensation decision will be continued to be executed, avoiding equipment overload caused by a single compensation target.

[0039] By converting the compensation decision into specific control instructions and driving the execution equipment, combined with real-time monitoring and coordinated protection mechanisms, the voltage sag is effectively alleviated, the wind turbine output voltage is quickly restored to the stable range, and the safe operation of the unit and the power supply quality of the grid are ensured.

[0040] Further, the method provided in the embodiment of the application comprises the following steps:

[0041] A210: Perform voltage drop amplitude calculation based on the three-phase voltage output data to generate voltage sag amplitude information.

[0042] A220: Perform Fourier transform on the three-phase voltage output data, identify the phase characteristics based on the transform result, and generate voltage sag phase information.

[0043] A230: based on the time stamp in the three-phase voltage output data, the duration of the voltage sag is extracted, and the duration information of the voltage sag is generated.

[0044] A240: the voltage sag element feature is constructed with the voltage sag amplitude information, the voltage sag phase information and the duration information of the voltage sag.

[0045] In the embodiments of the present application, the duration information of the voltage sag is the time during which the voltage sag has lasted, rather than the time from reduction to recovery.

[0046] Specifically, when the voltage sag element is extracted, the voltage drop amplitude is first calculated based on the three-phase voltage output data. Taking the rated voltage of the wind turbine generator 690V as an example, the effective value of the current three-phase voltage is obtained in real time, such as A phase 510V, B phase 505V, C phase 515V, and the drop amplitude of A phase is about 26.1%, B phase is about 26.8%, and C phase is about 25.4% by formula (rated voltage-current voltage) / rated voltage. The voltage sag amplitude information is generated by integration, reflecting the degree of voltage drop of each phase.

[0047] Subsequently, the three-phase voltage output data is subjected to Fourier transform, and the voltage signal of each phase continuously changing with time in the time domain is obtained, such as the instantaneous voltage value sequence of A phase from t0 to t1. These time domain signals are converted into frequency domain signals by Fourier transform algorithm, so that the phase components of the fundamental wave (50Hz or 60Hz) and 2nd, 3rd and other harmonics in each phase voltage are decomposed. Then the fundamental wave phase is extracted. The fundamental wave is the main component of the voltage signal, and its phase change is more critical to the analysis of voltage sag. The fundamental wave phase is compared with the rated fundamental wave phase of the wind turbine generator in normal operation (such as the rated phase of A phase is 0°), and the phase offset value of each phase is obtained, for example, the fundamental wave phase of A phase lags behind the rated phase by 5°, the fundamental wave phase of B phase leads by 2°, and the fundamental wave phase of C phase has no offset. Finally, these phase offset data are integrated to generate the voltage sag phase information, which clearly reflects the difference in phase of A, B and C three phases during voltage sag, and provides key basis for subsequent judgment of voltage sag characteristics in phase dimension.

[0048] At the same time, the duration based on the time stamp in the three-phase voltage output data is extracted. From the time t1 when the voltage sag occurs to the current data collection time t2, the time difference between the two is calculated to obtain the duration of the voltage sag, and the duration information of the voltage sag is generated. The duration information of the voltage sag records the duration from the occurrence to the current, rather than the time from reduction to recovery.

[0049] Finally, the voltage sag amplitude information, the voltage sag phase information and the duration information of the voltage sag obtained above are integrated, as shown in Table 1, to construct a complete voltage sag element feature, which fully characterizes the key characteristics of the voltage sag.

[0050] By extracting the amplitude, phase and duration information of the voltage sag in steps and integrating, the accurate voltage sag element characteristics are constructed, which provides detailed and key basis for subsequent trigger source judgment and compensation decision based on these characteristics.

[0051] Table 1: Voltage sag element characteristic parameter table

[0052]

[0053] Further, step A300 in the method provided by the embodiment of the application comprises:

[0054] A310: Construct a voltage sag judgment sample library.

[0055] A320: Traverse each voltage sag judgment sample in the voltage sag judgment sample library, and input the unit operation state data, the voltage sag element characteristics and the voltage sag judgment sample into a twin network for similarity identification to obtain similarity output information.

[0056] A330: If the similarity output information meets a preset threshold, collect the historical voltage sag trigger factors corresponding to the voltage sag judgment sample, complete the fault trigger and fluctuation trigger judgment, and generate the trigger source information.

[0057] In the embodiment of the application, the voltage sag judgment sample library is a sample set containing historical voltage sag events, each sample is associated with unit operation state data, voltage sag element characteristics and corresponding historical voltage sag trigger factors, and is used for similarity comparison with current data to judge the trigger source. The twin network is a network for similarity identification, which receives the samples in the voltage sag judgment sample library, the unit operation state data and the voltage sag element characteristics, performs similarity calculation and outputs similarity information to assist in completing the fault trigger and fluctuation trigger judgment.

[0058] Optionally, when constructing the voltage sag judgment sample library, all kinds of voltage sag event data occurring in the historical operation of the wind turbine are collected, covering complete three-phase voltage output data of each event, including instantaneous value, effective value and time stamp of each phase voltage; corresponding unit operation state data, such as wind speed, blade speed, generator power, equipment temperature, etc.; and explicit trigger factors recorded by the system or manually annotated, such as power grid short circuit, unit inverter fault, sudden wind speed rise, etc.

[0059] The raw data is pre-processed, including removing outliers such as sensor false alarm extreme voltage values; standardization, converting voltage amplitude, speed and other data of different orders of magnitude to the same interval; time alignment, ensuring that the timestamps of three-phase voltage data and operating state data are accurately matched, and then classified and stored according to the trigger factor type of fault trigger or fluctuation trigger, forming a voltage sag judgment sample library containing thousands to tens of thousands of valid samples, each sample is associated with complete feature data and trigger factor label.

[0060] After completing the construction of the sample library, each voltage sag judgment sample in the library is traversed, and the feature data of the sample, i.e. historical three-phase voltage features and historical unit operating state data, are input into the pre-trained twin network together with the current detected unit operating state data and voltage sag element features. The twin network extracts the features of the historical sample and the current data through two sub-networks with the same structure, generates high-dimensional feature vectors, and then calculates the cosine similarity between the vectors to obtain similarity output information, such as the similarity of a sample to the current data is 0.85, and the other sample is 0.62.

[0061] Further, the training of the twin network needs to first select data from the voltage sag judgment sample library, construct positive sample pairs composed of different historical events of the same trigger source (such as voltage sag events caused by two power grid faults) and negative sample pairs composed of events of different trigger sources (such as sag events caused by an internal fault and a wind speed fluctuation); then standardize, time align, etc. Preprocess the three-phase voltage features and unit operating state data in the sample to unify the data format and dimension; then build two sub-networks with the same structure to extract the features of the historical sample and the current data, and through training, the sub-network learns the key features that can distinguish different trigger sources; In the training, a suitable loss function is used to make the feature vectors of the same type of samples closer and the feature vectors of different types of samples farther apart, and the network parameters are optimized through multiple iterations until the similarity judgment accuracy of the model on the validation set is stable at a high level, at which point the twin network training is complete.

[0062] Then, a preset threshold (such as 0.8) is set, when the similarity output information of a sample reaches or exceeds the threshold, it is determined that the current voltage sag event has a high degree of similarity with the historical sample, and the corresponding historical voltage sag trigger factor of the sample is collected. If the trigger factor of the sample with the highest similarity is a power grid fault, it is determined that the current event is fault triggered; if it is a wind speed fluctuation, it is determined to be fluctuation triggered, and finally the explicit trigger source information is generated, such as "trigger source: fault triggered (power grid short circuit)".

[0063] By constructing a sample library containing rich historical data, combined with twin network for similarity matching, the trigger source type of voltage sag is accurately identified, which provides key basis for subsequent targeted compensation decision and improves the accuracy and reliability of trigger judgment.

[0064] Further, the step A300 in the method provided by the embodiment of the application comprises:

[0065] A340: The fault trigger refers to the voltage sag caused by the internal fault of the wind turbine generator or the power grid fault, and the fluctuation trigger refers to the voltage sag caused by the wind speed fluctuation.

[0066] Specifically, during the operation of the wind turbine generator, when the voltage sag phenomenon is detected, the trigger type thereof, i.e., the fault trigger or the fluctuation trigger, needs to be further distinguished.

[0067] For the judgment of the fault trigger, the wind turbine generator internal equipment state data and the power grid operation data need to be combined: if the generator winding temperature is collected to be suddenly increased from the normal working temperature 70℃ to 105℃ within 10 seconds, the inverter output current is distorted, the total harmonic distortion rate exceeds 5%, and the three-phase voltage unbalance degree exceeds 2%, it is possible that the voltage sag is caused by the internal fault (such as inverter fault) of the wind turbine generator; if the voltage of the point of common coupling connected to the power grid is monitored to be suddenly decreased from the rated voltage within 50 milliseconds, for example, from 690V to 480V, and the voltage sag occurs in other wind turbine generators in the region at the same time, and the power grid frequency fluctuation range exceeds ±0.5Hz, it can be initially determined that the voltage sag is caused by the power grid fault (such as line short circuit).

[0068] For the judgment of the fluctuation trigger, the correlation analysis of the wind speed and the wind turbine generator operation state is mainly relied on: for example, the wind speed is collected by the anemometer to be suddenly increased from 12m / s to 20m / s within 3 seconds, which exceeds the normal fluctuation range ±3m / s, at the same time, the blade rotating speed is increased from 18rpm to 25rpm, the generator active power is suddenly increased from 800kW to 1200kW, and the wind turbine generator internal equipment temperature, current and power grid voltage are all in the normal range, at this time, it can be initially determined that the voltage sag is caused by the wind speed fluctuation.

[0069] Further, in the actual judgment process, the system will traverse the voltage sag judgment sample library, compare the real-time collected internal fault characteristic data such as temperature and current distortion rate, power grid fault characteristic data such as point of common coupling voltage and frequency, wind speed fluctuation characteristic data such as wind speed change rate and speed correlation degree with the corresponding characteristic data of all historical samples in the library, and calculate the similarity of each sample and the current data. A preset threshold such as 0.8 is set, when there is a sample with similarity exceeding the threshold, the sample with the highest similarity is selected, if the historical trigger factor corresponding to the sample is wind turbine internal fault or power grid fault, it is determined that the current event is fault trigger; if it is wind speed fluctuation, it is determined to be fluctuation trigger, so as to determine the specific trigger source of the voltage sag.

[0070] By determining the judgment basis and data characteristics of fault trigger and fluctuation trigger, the different causes of voltage sag are accurately distinguished, which provides clear basis for subsequent development of targeted compensation strategy according to the trigger source, and ensures the effectiveness and accuracy of the compensation measures.

[0071] Further, the step A400 in the method provided by the embodiment of the application comprises:

[0072] A410: performing digital simulation modeling on the wind turbine to construct a twin unit model, wherein the modeling comprises integrated modeling of the unit itself, the connected power grid and the weather conditions.

[0073] A420: performing trigger source statistics based on historical voltage sag record information, generating a plurality of preset trigger sources, and collecting a historical voltage sag recovery record set.

[0074] A430: loading the historical voltage sag recovery record set to the twin unit model, learning the relationship between the voltage compensation parameters and the voltage sag elements and the trigger sources, and constructing the voltage sag compensator, wherein the voltage sag compensator is connected with the twin unit model.

[0075] A440: inputting the trigger source information and the voltage sag element characteristics into the voltage sag compensator, performing compensation parameter analysis, calling the twin unit model for compensation verification after the verification, and outputting the compensation decision after the verification.

[0076] In the embodiment of the application, the twin unit model is a model constructed by performing digital simulation modeling on the wind turbine, wherein the modeling comprises integrated modeling of the unit itself, the connected power grid and the weather conditions.

[0077] Specifically, first, a digital simulation modeling of the wind turbine is performed to build a twin unit model, which covers integrated data integration and model building of the unit itself, connection to the power grid and weather conditions. The modeling of the wind turbine itself requires the input of core parameters by those skilled in the art, including hardware parameters such as rated power, blade length, generator rated voltage, and converter capacity, and operating data such as the corresponding relationship between blade pitch angle and wind speed, and the characteristic curve of generator speed and output power; the modeling of the connection to the power grid needs to include the line impedance from the point of common coupling to the substation, the transformer capacity and ratio, and the short-circuit capacity of the power grid, to simulate the voltage transfer characteristics during power grid faults; the modeling of the weather conditions needs to associate environmental factors such as wind speed, air density, and temperature with the mapping relationship of unit output, and finally form a digital twin that can reproduce the voltage output characteristics of the unit under different working conditions, i.e., the twin unit model.

[0078] Next, when the trigger source statistics are based on historical voltage sag record information, all voltage sag events in the past 5 years need to be sorted and classified according to the trigger factors: for example, power grid faults, such as line short circuit and transformer fault, account for 60%, wind turbine internal faults, such as inverter fault and bearing overheating, account for 25%, and wind speed fluctuation, ±5m / s or more sudden change, accounts for 15%, from which multiple preset trigger sources such as power grid short circuit, inverter fault, and wind speed sudden rise are generated. At the same time, the corresponding historical voltage sag recovery record set is collected, each record containing the trigger source type, voltage sag elements such as amplitude drop, phase lag, and duration; compensation parameters such as reactive power compensation amount and converter modulation ratio; and recovery effect such as voltage recovery time and post-recovery fluctuation, forming thousands of effective records.

[0079] Then, when the historical voltage sag recovery record set is loaded to the twin unit model to construct the voltage sag compensator, the unstructured data in the record set needs to be converted into a feature vector recognizable by the model through data preprocessing: the trigger source type is converted into one-hot encoding, for example, there are 8 trigger sources, and the one-hot encoding vector length is 8. The power grid short circuit corresponds to: [1, 0, 0, 0, 0, 0, 0, 0], the transformer fault corresponds to: [0, 1, 0, 0, 0, 0, 0, 0], the inverter fault corresponds to: [0, 0, 1, 0, 0, 0, 0, 0], the bearing overheating corresponds to: [0, 0, 0, 1, 0, 0, 0, 0], the wind speed sudden increase corresponds to: [0, 0, 0, 0, 1, 0, 0, 0], the wind speed sudden decrease corresponds to: [0, 0, 0, 0, 0, 1, 0, 0], the generator fault corresponds to: [0, 0, 0, 0, 0, 0, 1, 0], and the line overload corresponds to: [0, 0, 0, 0, 0, 0, 0, 1]; the voltage sag elements are standardized to a value in the interval [-1, 1], for example, a 30% amplitude drop corresponds to -0.3, and the compensation parameters are kept as original values. Then, the compensator model is trained using the random forest algorithm, taking the trigger source and the voltage sag element as the input feature, and taking the compensation parameter as the output target. The model parameters are optimized through 5-fold cross-validation, such as the number of decision trees is 200 and the maximum depth is 10, so that the prediction error of the model on the validation set is controlled within 5%. At the same time, a real-time communication interface between the compensator and the twin unit model is established to ensure that the parameters output by the compensator can directly drive the twin model to perform simulation verification.

[0080] Finally, after the current trigger source information and the voltage sag element feature are input into the voltage sag compensator, the compensator quickly outputs the initial compensation parameters, such as the reactive power compensation amount and the converter modulation ratio, through the built-in training model. Then, the twin unit model is called to substitute the parameters into the simulation environment to simulate the voltage recovery process. Assuming that the model output shows that the voltage rises to 92% of the rated value within 15 ms after compensation, and the fluctuation range is ±3%, which meets the preset verification standards of recovery time ≤20 ms and fluctuation ≤±5%, the verification is passed, and the compensator formally outputs the compensation decision.

[0081] By constructing a multi-dimensional twin unit model, training the compensator based on historical data, and combining simulation verification, accurate mapping from the trigger source and the sag element to the compensation decision is realized, which ensures that the output compensation decision can effectively respond to different types of voltage sags, and improves the reliability and pertinence of the voltage stability control of the wind turbine generator.

[0082] Further, step A430 in the method provided in the embodiments of the present application further includes:

[0083] A431: If the trigger source is a fault type trigger source, load the historical voltage sag recovery record set to the twin unit model, learn the fault isolation feature and the fault source position relationship, and optimize the voltage sag compensator.

[0084] Specifically, when the trigger source is determined to be a fault type trigger source, such as an internal inverter failure of the wind turbine or a short circuit of the power grid line, first, the records related to the fault type in the historical voltage sag recovery record set are screened. These records contain detailed information of various fault events that have occurred in the past, including the specific location of the fault source, such as the generator side, the transformer end, or a certain section of the power grid line; the fault isolation characteristics at the time of the fault, such as the amplitude and frequency of the current surge, the distortion mode of the voltage waveform, and the imbalance degree of the three-phase voltage; as well as the compensation parameters and operation process for successfully recovering the voltage in the corresponding event.

[0085] Subsequently, the screened historical fault type records are loaded into the twin unit model. The twin unit model simulates the evolution process of fault isolation characteristics under different fault source locations by reproducing historical fault scenarios: for example, when the fault source is an inverter, the model will show a sharp increase in current harmonics in a specific frequency band, and the A-phase voltage distortion is particularly obvious; when the fault source is a short circuit of a certain section of the power grid line, the model will show a simultaneous drop in three-phase voltage, and the sag amplitude decreases with the increase of the distance from the short circuit point. Through simulation learning of a large number of historical scenarios, the model gradually establishes a mapping relationship between fault isolation characteristics and fault source locations, clearly identifying which characteristic combinations correspond to inverter failures and which correspond to power grid line faults, as well as the subtle differences in characteristics under different location faults.

[0086] Next, based on the mapping relationship learned by the twin unit model, the voltage sag compensator is optimized. During the optimization process, the compensator updates its internal decision logic: when a new fault type voltage sag is detected, the compensator can quickly locate the fault source location based on the current fault isolation characteristics, such as real-time collected current harmonic data and voltage distortion conditions, combined with the learned mapping relationship. For different locations of the fault, the compensator will adjust the focus of the compensation parameters, for example, if it is determined to be an inverter failure, it will prioritize the response speed and accuracy of reactive power compensation to suppress harmonic effects; if it is determined to be a power grid line fault, it will strengthen the duration of voltage support to quickly recover the voltage in coordination with the power grid protection mechanism.

[0087] By training the twin unit model with historical fault records, the model learns the correlation between fault isolation characteristics and fault source locations, and then optimizes the voltage sag compensator, enabling the compensator to more accurately match fault types and compensation strategies when facing fault type trigger sources, thereby improving the pertinence and effectiveness of voltage sag compensation under fault conditions.

[0088] Further, the embodiments of the present application also include step A600, which includes:

[0089] A610: Collecting a plurality of other wind turbines connected to the wind turbine.

[0090] A620: Obtain the voltage compensation decisions of the several other wind turbines through a data sharing channel.

[0091] A630: Perform global coordination effect analysis and local compensation optimization on the compensation decision and the several voltage compensation decisions to generate a locally optimized compensation decision.

[0092] A640: Issue the locally optimized compensation decision to the corresponding wind turbine for optimized control.

[0093] In one embodiment, after outputting the compensation decision, the several other wind turbines connected to the current wind turbine are first collected. According to the power grid topology relationship of the wind farm, the associated wind turbines in the same power collection line or sharing the same point of common coupling are identified. The voltage outputs of these wind turbines are correlated, and the compensation behavior of a wind turbine may affect the voltage state of other wind turbines. By including them in the analysis range, the overall power grid fluctuation caused by single wind turbine compensation can be avoided.

[0094] Then, the voltage compensation decisions of these other wind turbines are obtained through a data sharing channel. The data sharing channel needs to have real-time transmission capability to ensure that the compensation schemes of the associated wind turbines, including compensation methods (such as reactive power regulation, converter control, etc.), planned compensation measures and execution time, etc. information, can be collected in time to provide complete decision basis for subsequent global analysis.

[0095] After that, the compensation effect global fusion of the compensation decision and the compensation decisions of other wind turbines is performed to generate global information, the wind turbine with the largest compensation adjustment amount after over-compensation is determined to generate a locally optimized compensation decision, and the specific steps are described in detail in A631-A632.

[0096] Finally, the locally optimized compensation decision is issued to the corresponding wind turbine to ensure that each wind turbine executes the optimized scheme in coordination. Time synchronization needs to be maintained during execution to avoid new voltage instability caused by execution time difference, and to ensure that the compensation behaviors of all wind turbines form a resultant force to effectively stabilize the power grid voltage.

[0097] By collecting the decisions of associated wind turbines, performing global analysis of coordination effect, optimizing the compensation scheme and executing in coordination, the coordination of multi-wind turbine voltage sag compensation in the wind farm is realized, the limitations of single decision are avoided, and the stability and reliability of the overall compensation effect are improved.

[0098] Further, the method provided in the embodiment of the application comprises the following steps:

[0099] A631: Perform global fusion of compensation effect on the compensation decision and the several voltage compensation decisions to generate global fusion compensation effect information.

[0100] A632: judging whether over-compensation occurs based on the global fusion compensation effect information, if yes, extracting a local unit with the largest compensation amount, and making compensation decision adjustment according to over-compensation parameters to generate the local optimization compensation decision.

[0101] Optionally, when the compensation decision is globally fused with the compensation effect of voltage compensation decisions of a plurality of other wind turbines, first, the core content of all decisions is summarized, including the compensation mode, compensation strength and execution rhythm of the reactive power injection and converter output adjustment planned to be taken by each unit.

[0102] Subsequently, through a voltage coordination analysis model at the wind farm level, the comprehensive influence of these decisions on the overall grid voltage when they jointly act is simulated, such as analyzing the voltage recovery curve and fluctuation range of the point of common coupling after the superposition of the compensation behaviors of each unit, and integrating the results of these comprehensive influences into global fusion compensation effect information to comprehensively reflect the overall effect of multi-unit coordinated compensation.

[0103] Among them, the construction of the voltage coordination analysis model needs to take the whole wind farm as the object, first integrate the core parameters of each wind turbine, such as rated voltage, compensation capacity, etc.; the topology structure of the connected grid, such as line connection relationship, impedance parameters; and the influence law of meteorological conditions, to form the basic data layer; then build a digital simulation framework based on these data to simulate the influence of single-unit compensation behavior on local voltage and the coupling transmission process of multi-unit compensation behavior through the grid line; then introduce historical multi-unit coordinated compensation case data to learn the correlation mode of different compensation combinations and the voltage change (such as recovery curve, fluctuation range) of the point of common coupling, and optimize the model parameters; finally, through actual operation data verification, ensure that the model can accurately reproduce the global voltage effect of multi-unit compensation.

[0104] After that, based on the global fusion compensation effect information, whether over-compensation occurs is judged according to the stability standard of the grid voltage, such as the allowed fluctuation range of the rated voltage. If the analysis finds that the comprehensive compensation effect may cause the voltage to exceed this range, it is determined that over-compensation occurs. At this time, from all the units participating in compensation, a local unit with the largest planned compensation amount is extracted, because the compensation behavior of this type of unit usually has the most significant influence on the global voltage change, and is the key object of adjustment.

[0105] Finally, when making compensation decision adjustment according to over-compensation parameters, the degree of over-compensation is combined, such as the amplitude of the predicted voltage exceeding the rated value, and the compensation decision of the local unit is adjusted accordingly, such as reducing its compensation strength or slowing down the compensation rhythm. After adjustment, the global fusion analysis is verified again to ensure that the overall compensation effect is within a reasonable range, and finally the local optimization compensation decision is generated, which takes into account the global stability and the characteristics of the local unit.

[0106] The over-compensation risk is identified by globally fusing the compensation effects of multiple units, and the units with the greatest impact are targeted for adjustment, effectively avoiding voltage overshoot in collaborative compensation, and ensuring the stability and accuracy of overall voltage recovery of the wind farm.

[0107] Further, the step A632 in the method provided by the embodiment of the application includes:

[0108] A632-1: constructing a local-global fusion relationship by analyzing the relationship between historical local compensation data and globally fused compensation data.

[0109] A632-2: making compensation decision adjustment according to the over-compensation parameter based on the local-global fusion relationship.

[0110] In one embodiment, when making compensation decision adjustment according to the over-compensation parameter, first, historical local compensation data and globally fused compensation data are collected. The historical local compensation data includes compensation measures implemented by each wind turbine in the past, such as reactive power adjustment, converter parameter adjustment, etc., and corresponding voltage changes of the unit itself; the globally fused compensation data is the overall voltage state of the entire wind farm or associated power grid after the joint action of these local compensation measures, such as common connection point voltage fluctuation, recovery speed, etc.

[0111] By analyzing the above historical data, the correlation pattern between local compensation behavior and global voltage change is found out, such as how the increase of the compensation strength of a unit will affect the rising amplitude of the global voltage, and then the corresponding relationship between local compensation and global effect, i.e., the local-global fusion relationship, is constructed.

[0112] Based on the constructed local-global fusion relationship, compensation decision adjustment is made in combination with the over-compensation parameter. The over-compensation parameter reflects the degree to which the current global voltage may exceed the stable range, according to the local-global fusion relationship, it is clear which local unit's compensation behavior has the greatest impact on global over-compensation, and how much local compensation needs to be adjusted to pull the global voltage back to the stable interval. For example, if the relationship shows that the compensation amount of a unit is significantly positively correlated with the global voltage rise, and the current over-compensation parameter indicates that the voltage overshoot is significant, then the compensation strength of the unit is targeted to be reduced until the overall voltage is verified to be in the stable range through global analysis, and the compensation decision adjustment is completed.

[0113] By analyzing the historical data to establish the correlation between local and global, and accurately adjusting the local compensation decision according to the over-compensation parameter, effective control in the over-compensation situation is realized, and the stability and safety of the overall voltage of the wind farm are ensured.

[0114] By globally fusing the compensation effects of each unit to identify over-compensation risks, and by adjusting the unit with the largest compensation amount, the voltage overshoot problem in multi-unit coordinated compensation is effectively avoided, ensuring the stability and accuracy of the overall voltage recovery of the wind farm.

[0115] In summary, the wind turbine voltage sag compensation method based on deep learning provided by the embodiments of the application has the following technical effects:

[0116] The application acquires the three-phase voltage output data and unit operating state data of the wind turbine in real time, obtains the trigger source information and voltage sag element characteristics through sag element extraction, twin network similarity identification and other processes, generates compensation decisions through the voltage sag compensator in combination with these information, and adjusts them through global coordinated influence analysis and local compensation optimization, thereby accurately compensating the voltage sag of the wind turbine, making the control and compensation effect of voltage stability in the power supply / distribution circuit system more optimal, meeting the stable operation demand of the wind turbine, and achieving the technical effects of accurately identifying the trigger source and element characteristics of the wind turbine voltage sag and effectively improving the control and compensation of voltage stability.

[0117] Embodiment two, as shown in the same inventive concept as the preceding embodiment one, the embodiment of the application provides a wind turbine voltage sag compensation system based on deep learning, which comprises: Figure 2 A wind turbine data acquisition module 1 is configured to acquire the three-phase voltage output data and unit operating state data of the wind turbine in real time.

[0118] A voltage sag behavior judgment module 2 is configured to judge whether a voltage sag behavior occurs based on the three-phase voltage output data, and if so, to perform sag element extraction and generate voltage sag element characteristics.

[0119] A trigger source information acquisition module 3 is configured to perform fault triggering and fluctuation triggering judgment based on the unit operating state data and the voltage sag element characteristics, and generate trigger source information.

[0120] A compensation decision output module 4 is configured to perform compensation analysis through a voltage sag compensator based on the trigger source information and the voltage sag element characteristics, and output a compensation decision.

[0121] A voltage sag compensation control execution module 5 is configured to perform voltage sag compensation control based on the compensation decision.

[0122] Further, the voltage sag behavior judgment module 2 is configured to perform the following steps:

[0123] ​

[0124] performing voltage sag amplitude calculation based on the three-phase voltage output data to generate voltage sag amplitude information; performing Fourier transform on the three-phase voltage output data, identifying phase characteristics based on the transform result to generate voltage sag phase information; performing duration extraction based on time stamps in the three-phase voltage output data to generate voltage sag duration information; and constructing the voltage sag element features with the voltage sag amplitude information, the voltage sag phase information and the voltage sag duration information.

[0125] Further, the trigger source information acquisition module 3 is configured to perform the following steps:

[0126] Further, the trigger source information acquisition module 3 is configured to perform the following steps:

[0127] The fault trigger refers to the voltage sag caused by internal faults of the wind turbine or power grid faults, and the fluctuation trigger refers to the voltage sag caused by wind speed fluctuation.

[0128] Further, the compensation decision output module 4 is configured to perform the following steps:

[0129] Further, the compensation decision output module 4 is configured to perform the following steps:

[0130] If the trigger source is a fault trigger source, the historical voltage sag recovery record set is loaded into the twin unit model to learn the relationship between the fault isolation features and the fault source position, and the voltage sag compensator is optimized.

[0131] Further, the system further comprises a local optimization compensation decision control module, which is configured to perform the following steps:

[0132] Collect a plurality of other wind turbines connected with the wind turbine; obtain a plurality of voltage compensation decisions of the plurality of other wind turbines through a data sharing channel; perform global collaborative influence analysis and local compensation optimization on the compensation decision and the plurality of voltage compensation decisions to generate a locally optimized compensation decision; and issue the locally optimized compensation decision to the corresponding wind turbine for optimization control.

[0133] Perform global fusion of compensation effects on the compensation decision and the plurality of voltage compensation decisions to generate global fusion compensation effect information; determine whether over-compensation occurs based on the global fusion compensation effect information, and if so, extract a local unit with the largest local compensation amount, and perform compensation decision adjustment according to over-compensation parameters to generate the locally optimized compensation decision.

[0134] Construct a local-global fusion relationship by analyzing the relationship between historical local compensation data and global fusion compensation data; and perform compensation decision adjustment according to the over-compensation parameters based on the local-global fusion relationship.

[0135] The wind turbine voltage sag compensation system based on deep learning provided by the embodiment of the application can execute the wind turbine voltage sag compensation method based on deep learning provided by any embodiment of the application, and has the corresponding function modules and beneficial effects of the execution method.

[0136] Although the present application makes various references to certain modules in the system according to the embodiments of the present application, however, any number of different modules can be used and run on the user terminal and / or server, and each unit and module included is only divided according to the functional logic, but is not limited to the above division, as long as the corresponding functions can be realized; in addition, the specific names of each functional unit are only for easy mutual differentiation, and do not limit the protection scope of the present application.

[0137] The above specific embodiments do not constitute a limitation on the protection scope of the present application. Those skilled in the art should understand that various modifications, combinations and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present application shall be included in the protection scope of the present application. In some cases, the actions or steps described in the present application can be executed in an order different from that in the embodiments and still achieve the desired results. In addition, the processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multi-task processing and parallel processing are possible or can be advantageous.

Claims

1. A wind turbine generator voltage sag compensation method based on deep learning, characterized in that, The method comprises the following steps: Real-time acquisition of three-phase voltage output data and unit operation state data of a wind turbine; Determine whether a voltage sag behavior occurs based on the three-phase voltage output data. If so, extract the voltage sag elements and generate the voltage sag element features; Based on the unit operation state data and the voltage sag element features, make a fault trigger and fluctuation trigger judgment, and generate trigger source information; According to the trigger source information and the voltage sag element features, compensation analysis is carried out through the voltage sag compensator, and compensation decision is output; According to the compensation decision, voltage sag compensation control is carried out; According to the trigger source information and the voltage sag element features, compensation analysis is carried out through the voltage sag compensator, and compensation decision is output, which comprises: Carrying out digital simulation modeling on the wind turbine to construct a twin unit model, wherein the modeling comprises integrated modeling of the unit itself, the connected power grid and the weather conditions; Based on the historical voltage sag record information, the trigger source is counted to generate a plurality of preset trigger sources, and a historical voltage sag recovery record set is collected; Load the historical voltage sag recovery record set into the twin unit model to learn the relationship between the voltage compensation parameters and the voltage sag elements and the trigger sources, and construct the voltage sag compensator, wherein the voltage sag compensator is connected with the twin unit model; The trigger source information and the voltage sag element features are input into the voltage sag compensator, and after compensation parameter analysis, the twin unit model is called for compensation verification. If the verification is passed, the compensation decision is output; If the trigger source is a fault type trigger source, load the historical voltage sag recovery record set into the twin unit model to learn the fault isolation features and the fault source position relationship, and optimize the voltage sag compensator.

2. The deep learning-based wind turbine generator voltage sag compensation method of claim 1, wherein, Extract the voltage sag elements to generate the voltage sag element features, which comprises: Based on the three-phase voltage output data, calculate the voltage drop amplitude to generate voltage sag amplitude information; Perform Fourier transform on the three-phase voltage output data, identify the phase features based on the transform results, and generate voltage sag phase information; Extract the duration information from the timestamp in the three-phase voltage output data to generate the duration information of the voltage sag; Construct the voltage sag element features based on the voltage sag amplitude information, the voltage sag phase information and the duration information of the voltage sag.

3. The deep learning-based wind turbine generator voltage sag compensation method of claim 1, wherein, Based on the unit operation state data and the voltage sag element features, make a fault trigger and fluctuation trigger judgment, and generate trigger source information, which comprises: Construct a voltage sag judgment sample library; Iterate through each voltage sag judgment sample in the voltage sag judgment sample library, input the voltage sag judgment sample together with the unit operation state data and the voltage sag element features into a twin network for similarity identification to obtain similarity output information; If the similarity output information meets the preset threshold, collect the historical voltage sag trigger factors corresponding to the voltage sag judgment sample to complete the fault trigger and fluctuation trigger judgment, and generate the trigger source information.

4. The deep learning-based wind turbine generator voltage sag compensation method of claim 3, wherein, Fault trigger refers to voltage sag caused by internal faults of wind turbines or power grid faults, and fluctuation trigger refers to voltage sag caused by wind speed fluctuations.

5. The deep learning-based wind turbine generator voltage sag compensation method of claim 1, wherein, After outputting the compensation decision, comprising: Collecting a plurality of other wind turbines connected to the wind turbine; Obtaining a plurality of voltage compensation decisions of the plurality of other wind turbines through a data sharing channel; Performing global collaborative influence analysis and local compensation optimization on the compensation decision and the plurality of voltage compensation decisions to generate a locally optimized compensation decision; Downlinking the locally optimized compensation decision to the corresponding wind turbine for optimization control.

6. The deep learning-based wind turbine generator voltage sag compensation method of claim 5, wherein, Performing global collaborative influence analysis and local compensation optimization on the compensation decision and the plurality of voltage compensation decisions to generate a locally optimized compensation decision, comprising: Performing global fusion of compensation effects on the compensation decision and the plurality of voltage compensation decisions to generate global fusion compensation effect information; Based on the global fusion compensation effect information, if over-compensation occurs, extract the local unit with the largest local compensation amount, and adjust the compensation decision according to the over-compensation parameter to generate the locally optimized compensation decision.

7. The deep learning-based wind turbine generator voltage sag compensation method of claim 6, wherein, Adjusting the compensation decision according to the over-compensation parameter, comprising: By analyzing the relationship between historical local compensation data and global fusion compensation data, a local-global fusion relationship is constructed; Based on the local-global fusion relationship, compensation decision adjustment is performed according to the over-compensation parameter.

8. A wind turbine generator voltage sag compensation system based on deep learning, characterized by, The system for implementing the deep learning-based wind turbine voltage sag compensation method of any one of claims 1-7, comprising: A wind turbine data acquisition module for real-time acquisition of three-phase voltage output data and unit operating state data of a wind turbine; A voltage sag behavior judgment module for determining whether a voltage sag behavior occurs based on the three-phase voltage output data, and if so, extracting a voltage sag element to generate a voltage sag element feature; A trigger source information acquisition module for fault triggering and fluctuation triggering based on the unit operating state data and the voltage sag element feature to generate trigger source information; A compensation decision output module for compensation analysis by a voltage sag compensator based on the trigger source information and the voltage sag element feature to output a compensation decision; A voltage sag compensation control execution module for voltage sag compensation control according to the compensation decision.

Images