79results about "Neural network algorithms" patented technology

The invention discloses a variable-speed wind generating set maximum wind energy capturing method based on effective wind speed estimation. The method comprises an effective wind speed estimating model and a maximum wind energy capturing controller. In order to acquire an effective wind speed estimated value, a training set of an SVR model is formed by normalized unit historical output data and historical wind speed measured values, penalty parameters and kernel function parameters are selected by a GA algorithm to obtain the trained SVR model, and the model gives out the wind speed estimatedvalue in an online manner; and when the maximum wind energy capturing controller is design, the real-time optimal wind wheel rotating speed estimated value is obtained according to effective wind speed which is given out by the effective wind speed estimating model, nonlinear characteristics and parameter uncertainty of the system are responded by robust factors and a neural network, and therefore, boundedness of rotating speed tracking errors and stability of a wind generating set system are realized. By the method, a mathematical model and parameters of the unit do not require to be used, adesign process is simple, an implementing cost is low, and the capacity of the unit and the economic benefit of a wind power plant can be improved.

The invention provides a large-size wind turbine feedback linearization power control method based on a radial basis function (RBF) neural network, and relates to the field of wind turbine power control. The control method comprises the steps that 1, on the basis that a wind turbine state space is built, an affine nonlinear model of a variation paddle controller is put forward, and the feedback linearization control law is designed; 2, an affine nonlinear model of a torque controller is built, and the feedback control law is designed; 3, on the basis of analyzing blade force vibration and tower vibration, the threshold value definition of a dual-loop controller is put forward for serving as the basis of controller transformation; and 4, the RBF neural network is designed to serve as a compensation control method for controlling errors. According to the dual-loop controller based on feedback linearization provided by the invention, the constant power control problems that the wind speed change is large, small, and fast after the rated wind speed is exceeded can be solved, the control precision is also improved through error compensation, and the robust performance of the system is improved.

The invention provides a self-adaptive dynamic planning control method and system for a large-scale energy storage power station. The method includes steps of setting a structure and control target parameters of a self-adaptive dynamic planning control system; initiating the parameters and importing an initial state of a controlled object; calculating the original wind electricity power fluctuation rate of a current moment t and performing smoothing treatment on the original wind electricity power by using a change rate control method; calculating the wind storage power fluctuation rate subjected to the smoothing treatment, the energy storage power and the SOC (State Of Charge) of the energy storage system; initiating a training evaluation module and an evaluation module; calculating and saving the control strategy, the wind storage power fluctuation rate, the energy storage power and the SOC of the energy storage system of each moment; outputting the control strategy, the wind storage power fluctuation rate subjected to the smoothing treatment, the energy storage power and the SOC of the energy storage system of each moment. The system includes a parameter initializing module, a data acquisition and calculation module, an execution module, an evaluation module and an output module and the like. By adopting the method and system provided by the invention, optimization control of the large-scale energy storage system is realized and the technical economical efficiency of the energy storage system is improved.

The invention relates to a wind power variable-pitch multi-variable fuzzy neural network PID control method. The control method includes the following steps that a fuzzy parameter setting module is used for presetting the weight of a PID neural network module; the error between a rotating speed reference value and the actual rotating speed output of a wind driven generator is calculated through a PID calculation module to obtain a reference output quantity of torque of the wind driven generator; the error between the power output value and the power reference value of the wind driven generator and the error change rate are set through the fuzzy parameter setting module to obtain a presetting parameter of the weight of the PID neural network module; through a negative gradient algorithm with a momentum factor, the weight of the PID neural network module is trained, and the reference value output of torque and the reference value output of the pitch angle of the wind driven generator are adjusted. The output power of the wind driven generator can be stabilized nearby a rated valve, and safety of a fan is ensured.

A neural network-based system for monitoring a turbine compressor. In various embodiments, the neural network-based system includes: at least one computing device configured to monitor a turbine compressor by performing actions including: comparing a monitoring output from a first artificial neural network (ANN) about the turbine compressor to a monitoring output from a second, distinct ANN about the turbine compressor; and predicting a probability of a malfunction in the turbine compressor based upon the comparison of the monitoring outputs from the first ANN and the second, distinct ANN.

Systems and methods for determining turbine pressure related loads and for condition monitoring are provided. The systems and methods may measure at least one pressure differential on an airfoil. At least one pressure differential may be used to determine a root bending moment associated with the blade. Additionally or alternatively, at least one pressure differential may be used to determine a low-speed shaft moment for a turbine on which the blade is mounted. Still further, at least one pressure differential and/or moment may be used to gauge wear/fatigue and/or damage to one or more wind turbines. Based on this information, a controller may modify various operating characteristics of the turbine or blade to address the fatigue or damage.

The invention discloses an integrated learning based mountain wind generation set behavior predicating model which comprises the following steps: 1, adopting a self-adaptive threshold value method todetermine a wind speed power sub-interval data density threshold value to clear abnormal data; 2, defining a sample matrix, and adopting a self-adaptive comprehensive over-sampling method to generatea new sample matrix for uniformly distributing different wind conditions; 3, performing Hilbert-Huang transform on data generated in the step 2 to obtain decomposition amount of input amount of the mountain wind generator set model; 4, according to the data of the step 4, determining input and output of the mountain wind generator set model, selecting a component learner and adopting a fusion strategy of integrated learning of stacking, and training and integrating to obtain the mountain wind generation set behavior predicating model; 5, adopting a grid search method to determine optimal parameters of the model; and 6, operating and testing the mountain wind generation set behavior predicating model. The integrated learning based mountain wind generation set behavior predicating model canprovide service for wind generation set predicating control, so that maintenance staff can normally operate a maintaining unit more efficiently better.

Provided are a method and an apparatus for monitoring icing of wind turbine blades. The method comprises the following steps of: photographing an image of a blade through a camera; detecting an area where the blade is located from the captured image; removing image information of a background area other than the area where the blade is located from the photographed image to obtain a blade image; the obtained leaf image being input to a leaf icing identification model trained based on a sample set including a plurality of leaf images indicative of the leaf icing to determine whether the photographed leaf is iced. According to the method and the device, whether the blade of the wind turbine freezes or not can be quickly and accurately monitored.

The invention discloses a prediction and tracking control method of a wind driven generator at low wind speed. The method comprises the following steps: 1) the optimal angular speed of a wind wheel of the wind driven generator is predicted; and 2) a controller is designed to perform tracking control on the angular speed of the wind wheel of the wind driven generator. The prediction and tracking control method of the wind driven generator at low wind speed can accurately predict the wind speed of entering a wind field in a period of time in the future to obtain ideal predicting angular speed data near the real angular speed of the wind wheel of the wind driven generator at the future time, can prevent the problem of late control through accurately controlling the angular speed of the wind wheel of the wind driven generator and tracking the ideal predicting angular speed, increases the wind energy utilization rate of the wind driven generator, and enables the wind driven generator to output by the maximum power at low wind speed; and a designed RBF neural network self-adaption controller only depends on the system angular speed errors, and also can control the rotating speed of the wind driven generator under the condition of not knowing part of functions and parameters in a system.

The invention discloses a method, device and system for predicting a fan fault. The method comprises the following steps that sensing data of each component of a fan is acquired through data acquisition sensors mounted on the components of the fan and then transmitted to a data processing platform from a fan field through a wireless network as well as isolation equipment; the sensing data forms time series working data in a preprocessing process through the data processing platform; a well-trained recurrent neural network model is used for extracting implicit feature information in the time series working data, and then future time series prediction data of the fan is predicted to be acquired; and a working state of the fan is judged through the time series prediction data, and early warning is carried out. According to the method, device and system for predicting the fan fault, the technical problems that in the prior art on a safety operation solution of the fan, fault prediction cannot be continuously carried out due to the fact that detecting and alarming is carried out only after the fault occurs or an artificial diagnosis mode is used for predicting the fan fault can be solved.

One embodiment provides a system for facilitating fault diagnosis. During operation, the system collects current signals associated with a physical object which comprises a rotating machine. The system demodulates the collected signals to obtain current envelope signals, which eliminates fundamental frequencies and retains fault-related frequencies. The system resamples the current envelope signals, which converts the fault-related frequencies to constant frequency components. The system enlarges, by a fault-amplifying convolution layer, the resampled envelope signals to obtain fault information. The system provides the fault information as input to a deep convolutional neural network (CNN). The system generates, by the deep CNN, an output which comprises the fault diagnosis for the physical object.

The invention discloses a fan blade detection method and device, a device and a storage medium, and relates to the field of application programs. The method comprises the steps that sound signals generated when fan blades sweep wind are collected through a sound collection device, spectrograms corresponding to the sound signals are generated, image recognition is carried out on the spectrograms through a damage recognition model, and damage recognition results of the fan blades are recognized from the spectrograms. According to the method, the accurate identification of the damage types of thefan blades based on the spectrograms is achieved, the manual inspection is not needed, the manpower resources are saved, and the health state of the fan blades can be monitored in real time.

The invention provides a method for adaptive adjustment of a cut-out strategy. The method includes the following steps: predicting a wind resource parameter of a wind turbine of each of machine sitesby using a wind speed prediction model; predicting a fatigue load and a limit load of the wind turbine by using a load prediction model according to the predicted wind resource parameter and air density; comparing the predicted fatigue load and limit load with a reference load; and determining the cut-out strategy based on a result of the comparison, wherein the step of determining the cut-out strategy includes the step of determining a cut-out wind speed and output power. The method can predict wind resource parameters of wind turbines for different machine sites, and dynamically optimize thecut-out strategy of the wind turbines according to the predicted wind resource parameters to maximize utilization of wind resources, and the utilization efficiency of the wind turbines is further improved.

According to the invention, nonlinear unstable wind turbine cabin suspension system control is divided into a stabilization part and a performance improvement part; the system comprises a cabin suspension stabilization controller, a cabin suspension prediction controller and a model mismatch compensator of an RBF neural network. Cabin suspension current reference setting is cooperatively completed; an unstable cabin suspension system is stabilized by adopting state feedback; the invention discloses a stabilization-based cabin suspension system model. Constructing a discretized model predictionmodel; and comprehensively considering air gap tracking errors, air gap speed and multi-objective optimization state variables input by control variables, implementing prediction of reference current, adopting a model mismatch compensator of an RBF neural network to approach a model mismatch value online, and combining a model mismatch dynamic adjustment coefficient to jointly complete model mismatch compensation. According to the method, the influence of model mismatch on model prediction control performance is effectively solved, Meanwhile, an intelligent compensation way is provided for online optimization and adjustment of model prediction.

The invention relates to a system and a method for auto-tuning a combustion system of a gas turbine. A controller, a gas turbine, and a method for auto-tuning a combustion system of a gas turbine are disclosed. The method includes selecting a first tuning curve from a set of tuning curve for the gas turbine; unbalancing a stable operating point of the gas turbine by modifying one or more operational parameters based on a predefined recipe; determining tuning parameters and storing them while a current operating point of the gas turbine is brought back on the first tuning curve; and generating a backup of tuning parameters to recover the stable operating point.

The invention relates to the technical field of remote operation and maintenance diagnosis of wind power generation equipment, and provides a terminal-cloud collaborative wind power operation and maintenance diagnosis system based on 5G. The system comprises a data acquisition gateway, a 5G-based edge computing network, a wind field control center system, a front-arranged data acquisition serviceand an operation and maintenance diagnosis cloud platform. According to the system, under the synergistic effect of all subsystems, second-level acquisition, remote operation and maintenance and intelligent diagnosis of wind turbine data are achieved, the problems that wind turbine routing inspection is difficult and not easy to monitor are solved, the centralized control level of wind turbines indifferent places is improved, the fault rate of wind turbine operation is reduced, the high-quality power generation efficiency of the wind turbines is guaranteed, and considerable economic efficiency is created for a wind field.

Aiming at the variable propeller pitch control problem of a wind power generation system, the invention provides a variable propeller pitch control method based on proportion-extreme learning machinesteady state estimation. According to the variable propeller pitch control method based on the proportion-extreme learning machine steady state estimation, firstly, steady-state output of a PI controller of a wind generation set under various wind speeds is leaned through an ELM, and then variable propeller control of the wind generation set is performed adopting a method of combining the trainedELM and a proportional controller. By means of the variable propeller pitch control method based on the proportion-extreme learning machine steady state estimation, the defect of lag of traditional PIvariable propeller pitch control can be improved, and the stability of the output power of the wind generation set is facilitated.

The invention provides a wind turbine generator encoder fault tolerance method based on signal reconstruction. The wind turbine generator encoder fault tolerance method comprises the following steps of acquiring data information of a wind turbine generator within a preset time, preprocessing the data information, and identifying a transfer function and a neural network model of a servo mechanism according to the preprocessed data information; and designing a state observer according to the transfer function, setting a gain vector, configuring poles of the state observer in a preset range, verifying the reconstruction performance of the state observer and the accuracy of the neural network model when an encoder is normal, setting a threshold value of a residual error between an encoder measurement value and a state observer reconstruction value, and if the residual error exceeds the threshold value, respectively substituting the encoder measurement value and the state observer reconstruction value into the neural network model at the corresponding wind speed for verification so as to perform corresponding operation on the encoder. According to the method, a neural network model verification link is utilized, and the aim that the pitch angle can still be reconstructed through an algorithm after hardware equipment of the pitch angle encoder of the wind turbine generator breaks down is effectively achieved.

A method for predicting component failure in a drive train assembly of a wind turbine comprises acquiring data from a plurality of wind turbine sensors pertaining to one or more components of the drive train assembly. The data is fed into one or more RETINA remote nodes and is filtering and aggregating into time intervals. The data is archived in a centralized data-warehouse and is used to build a machine learning model configured to determine ideal temperatures of components in the drive train assembly. The ideal temperatures are compared to actual measured temperatures in order to determine one or more temperature deviations. The one or more temperature deviations are used to determine a severity index score. An alert is generated corresponding to a high severity index score, wherein the alert informs of a likely imminent component failure.

A system includes a processor configured to execute an artificial neural network (ANN). The processor is configured to receive one or more operational parameters associated with an operation of a turbine system. The turbine system includes one or more combustors. The processor is further configured to analyze, via the ANN, the one or more operational parameters to determine a characteristic pattern, and to generate, via the ANN, an output based at least in part on the determined characteristic pattern. The output includes an indication of an intensity of a flame of the one or more combustors to determine a presence or an absence of the flame.

The invention presents a middle-sized low-speed permanent magnetic direct drive wind turbine generator and a parameter self-study control method thereof, and belongs to the technical field of wind power. The low-speed permanent magnetic direct drive wind turbine generator structure (shown as Figure 1) comprises a wind wheel-1, a main shaft-2, a main bearing base-3, a main frame-4, a low-speed disk type permanent magnet motor-5, a low-speed shaft brake-6, a variable pitch dragging system-7, a yaw dragging system-8, and a cabin cover-9. The wind turbine generator controls the running by a parameter self-study algorithm, and applies a BP neutral network to predict the control parameter pitch angle beta, rotate speed omega, torque q and power p of the control system, determines the control objective through the prediction value, and completes the speed change and the bending moment control. Compared with the traditional low-speed disk type permanent magnetic wind turbine generator, the middle-sized low-speed permanent magnetic direct drive wind turbine generator reduces the risk that the power generation is slipped down and even cannot generate power for the demagnetization caused by shock of the permanent magnetic motor, and decreases the main shaft and other bearing parts; the main frame can apply a lightweight welding structure, so that the production cost is reduced and the generator is simple to maintain.

The invention discloses an indirect rotating speed control method based on width learning. The indirect rotating speed control method comprises the following steps of acquiring effective air speed information of a unit within a certain period of time; selecting unit output data relative to the effective air speed within a corresponding period of time by using interactive information; removing dependency in the acquired unit output data and carrying out a normalizing operation to construct a training set of a width learning system; determining the structure and parameters of the width learningsystem by using the training set to obtain an air speed estimating model; and giving an effective air speed value online by the model to further calculate a rotating speed tracking error and an indirect rotating speed control expression. The method reserves the advantage that a conventional indirect rotating speed control algorithm is simple in structure, overcomes the defect that the conventionalindirect rotating speed control algorithm is relatively slow in rate of convergence, can accelerate the accelerating and decelerating performance of the unit simultaneously, improves the wind energycapturing efficiency, and is simple and feasible, low in implementation cost and few in parameter needed to be debugged. Compared with the conventional indirect rotating speed control algorithm, the capacity of the unit can be improved and the benefit of an electric field can be increased.