35 results about "Radial basis function network" patented technology

The invention provides a coordination control method for a photovoltaic off-grid hydrogen production system, which comprises the following steps of: acquiring voltage and current time domain signals of a photovoltaic array, constructing an impedance frequency characteristic mapping model through tensor decomposition and a radial basis function network, and calculating converter control parameters; based on converter control parameters and environment data, temperature-sensing-free self-correction photovoltaic characteristic prediction is achieved through electrical parameter indirect temperature estimation, and an MPPT reference track and a power prediction track are generated; multi-time-scale data coordination processing of millisecond-level electrical control, second-level pressure regulation and minute-level temperature optimization is executed, potential conflicts are solved, and an execution instruction set is generated; and generating and issuing a system comprehensive control instruction by adopting a double-layer deep reinforcement learning network. According to the method, the system efficiency and stability are improved, and the performance is excellent especially under the illumination fluctuation working condition.

The invention belongs to the technical field of optical detection, and particularly relates to a global power wave aberration prediction-based Ritchey-Common detection method and system, and the method comprises the steps: firstly building a three-dimensional optical model through the parameters of a standard spherical mirror, a to-be-detected plane mirror and an interferometer, and combining a preset Ritchey angle and an out-of-focus / astigmatism coefficient, injecting vibration and temperature noise by using Monte Carlo simulation to generate an enhanced training data space; secondly, constructing a phase error adjustment model based on an adversarial network, and decoupling a mapping relation between an adjustment error and wave aberration distortion through a multi-scale attention mechanism and a radial basis function network to realize adaptive correction of sub-aperture phase data; and finally, splicing sub-aperture phases by using a graph neural network and fusing multi-angle full-aperture measurement data to accurately reconstruct the surface shape of the plane mirror to be measured. According to the invention, through confrontation training and dynamic error compensation of physical constraints, the detection efficiency and reliability of the large-aperture optical element are significantly improved.

The invention relates to the technical field of medical image analysis, in particular to a left ventricular compliance assessment method and system based on medical image.The method comprises the following steps that a planar three-dimensional coordinate from a left ventricular apex to a mitral annulus is obtained through a CT image, a thin-plate spline interpolation algorithm is called to establish a left ventricular long-axis model, the ultrasonic end-stage internal diameter is obtained synchronously, and a left ventricular compliance assessment result is obtained. And inputting the left greenhouse long axis model and the ultrasonic diastole end-stage inner diameter into an Euclidean distance calculation function to generate a left greenhouse length-diameter ratio. According to the method, a left greenhouse long axis model is constructed by integrating CT three-dimensional coordinates and ultrasonic inner diameter, the length-diameter ratio is calculated by combining Euclidean distance, point cloud registration is carried out, a myocardial wall area is segmented, local deformation is quantized by utilizing Gaussian curvature gradient, a correction factor is generated by inputting a radial basis function network, and a parameter coupling mechanism is established by fusing an aorta root angle. And optimizing the weight of the pressure-volume equation, establishing a dynamic model, and realizing a quantitative evaluation system of spatial heterogeneity and time continuity.

The invention discloses a time-varying state constraint terminal sliding mode tracking control method based on an adaptive network, and relates to the technical field of adaptive control, and the method comprises the steps: 1, constructing a constraint potential energy error for each actual state of an engineering controlled object; 2, constructing a composite sliding mode variable with a fixed rational index; all sliding mode variables of all actual states of the project controlled object form an integral sliding mode surface on a vector level; and 3, carrying out online approximation on the actual dynamic state of the engineering controlled object by utilizing the single-hidden-layer radial basis network, and introducing network output as a dynamic compensation item into control input, thereby ensuring that the whole sliding mode surface is driven to zero within finite time and synchronously completing real-time compensation on the actual state of the engineering controlled object. According to the invention, the finite time high-precision tracking control of the controlled object of the controlled project with unknown nonlinearity and double-side time-varying boundaries is realized.

The invention relates to the technical field of industrial automation control, and discloses a mixed PID temperature control system and method based on GDNN and multi-modal prediction. The system comprises a data acquisition module, an enhanced GDNN core controller, a prediction optimization module and an actuator module. And real-time adaptive tuning of PID parameters (Kp, Ki and Kd) is realized by fusing a general dynamic neural network, a genetic algorithm optimization initial weight, a variable learning rate and an additional momentum method, and a prediction model module of a multi-layer perceptron (MLP) and a radial basis function (RBF) network. According to the system, nonlinear, time-varying and interference factors can be processed for temperature control scenes such as a high-temperature furnace and a drying system, the mean absolute error, overshoot and stabilization time are remarkably reduced, and the energy efficiency is improved. A multi-modal sensor fusion and edge computing framework is innovatively introduced, and distributed industrial application is supported. The system is suitable for the fields of high-temperature furnaces, drying systems, intelligent buildings and the like, and provides high robustness and interpretability.

The invention belongs to the field of converter transformer temperature monitoring, and relates to a converter transformer winding temperature monitoring method based on digital twinning, and the method comprises the steps: building a converter transformer electromagnetic thermal coupling two-dimensional simulation model, and inputting actual operation current data as an excitation source into the two-dimensional simulation model for simulation calculation; performing electromagnetic simulation according to ampere-turn balance to obtain a winding temperature rise distribution cloud picture under different working conditions; inputting the data under the operation condition into a machine learning model for training to obtain a rapid calculation model based on a radial basis function network; a real-time current value of each winding during operation is collected by a sensor, the real-time current values are imported into a rapid calculation model, second-level real-time display of winding temperature distribution is realized, and the load is fed back and adjusted according to a set alarm threshold value. According to the method, a multi-working-condition electromagnetic thermal simulation model is established, temperature field simulation data are extracted and input into a machine learning model for data analysis, and the converter transformer winding temperature field digital twinborn body is achieved.

Embodiments of the invention provide a control method and apparatus, and an electronic device. The method comprises the steps of determining current information of a Stewart platform; the current information comprises a current pose and the rod length of each rod; inputting the current information into a prediction model to obtain adjustment information of the Stewart platform; wherein the prediction model is a network model obtained by optimizing a radial basis function network through a particle swarm optimization algorithm; control information of the Stewart platform is determined according to the adjustment information; and adjusting the current information according to the control information.

The invention discloses an SAR (Synthetic Aperture Radar) image geometric correction method and system for optimizing a radial basis function neural network based on a genetic algorithm. The method comprises the following steps: normalizing three-dimensional geographic coordinates of ground feature points, and inputting the normalized three-dimensional geographic coordinates to a full connection layer for feature extraction; constructing a ReRBF network, wherein the ReRBF network is formed by stacking a plurality of layers of RBF sub-modules through a residual connection structure; and the output of the previous module is used as a residual error, is spliced with the original input or the intermediate feature, and is jointly used as the input of the next module to form a progressive feature enhancement and residual error learning mechanism, so that the network can refine the coordinate mapping relationship layer by layer. And a DBSCAN algorithm is adopted to perform automatic clustering on training data, so that the generalization ability of the model in multiple scenes is improved. And optimizing the ReRBF network by using a genetic algorithm, and adaptively determining the optimal width parameter of the radial basis function network. According to the method, the geometric positioning precision and the processing efficiency of the SAR image under the complex terrain condition are improved.

The application discloses a mechanical arm iterative learning control method based on a radial basis function network, which solves the limitations of traditional mechanical arm control methods in dealing with nonlinear characteristics and external disturbance problems in a complex dynamic environment. The method comprises the following steps: firstly, constructing a mechanical arm system dynamics model, and adopting a dynamic correction strategy to dynamically correct and optimize a reference trajectory; then, designing a radial basis function neural network to construct a nonlinear compensation term, and designing dynamic weight parameters and dynamic learning gains for optimizing the performance of a controller; finally, designing an iterative learning controller, and verifying the stability and error convergence of the control algorithm. Through the radial basis function network, the dynamic adjustment strategy and the iterative learning strategy, the control precision, adaptability and error convergence speed of the mechanical arm system are improved.

A method for controlling a torque of at least one wheel of a mobile platform. The method includes: providing at least one current slip value of the wheel and at least one current wheel acceleration of the wheel as input values; providing a trained radial basis function network designed to determine, by means of the input values, at least one torque change as an output value for control of the at least one wheel; and determining a current torque change, by means of the trained radial basis function network and the provided input values, for control of the torque.

The invention discloses an aeromagnetic interference compensation method based on a radial basis function network. The method comprises the following steps: taking a 16-term coefficient matrix as input data of a deep neural network; a hidden space is constructed for data in hidden nodes in a hidden layer by using a radial basis function, the hidden layer performs certain conversion on input data, and low-dimensional mode input data is converted into a high-dimensional space, so that a linear inseparable problem in the low-dimensional space becomes linearly separable in the high-dimensional space; training a sample set of aeromagnetic data to obtain a network center and other weight parameters meeting supervision requirements; fitting the aeromagnetic interference by using the obtained weight parameter information to obtain predicted magnetic interference; and calculating the compensated standard deviation according to the predicted magnetic interference. The anti-interference capability of the model can be effectively improved.

The present invention discloses a rapid measurement system for a rectangular workpiece under three-coordinate structured light, which relates to the field of rapid measurement technology. The system comprises: an acquisition module, which is used to project four-directional orthogonal structured light through a dual projector, and cooperate with a high-speed industrial camera and a six-degree-of-freedom robotic arm to acquire a coarse positioning point cloud of the rectangular workpiece; an encoding module, which is used to generate a combined encoding sequence of Gray code and phase-shift stripes, obtain the wrapped phase through phase solution, calculate the heterodyne absolute phase using dual-frequency stripes, perform phase expansion in combination with a confidence weight map, and compensate for projection distortion through a radial basis function network; a point cloud processing module, which marks the point cloud as a plane, edge or corner area through semantic segmentation, imposes a normal vector verticality constraint on the plane area, imposes a straightness constraint on the edge area, and imposes a curvature constraint on the corner area, and determines whether the measurement is qualified by calculating a comprehensive error, thereby ensuring the accuracy and reliability of the measurement result.

The application discloses a kind of intelligent sports movement state analysis method, system, equipment and medium based on multi-module collaborative optimization.Method includes: using lightweight real-time model to video stream is human detection and initial key point positioning;Through radial basis function network, initial key point is nonlinear fine-tuning to improve accuracy;From the key point sequence after fine-tuning, extract high-dimensional kinematic feature, and utilize principal component analysis to reduce dimension, obtain core low-dimensional time sequence feature;Finally, through time sequence neural network, feature sequence is modeled, realizes action recognition and posture evaluation.The application passes through the collaborative optimization link of "perception-fine-purification-understanding", effectively solves the problems, such as imbalance of accuracy and speed, feature redundancy, time sequence modeling deficiency and error accumulation, in prior art, provides efficient, accurate, automated analysis solution for wisdom sports training.

The application discloses a kind of cable fault prediction processing method, device and electronic equipment.Therein, the method comprises: obtaining the partial discharge characteristic data of high-voltage cable;First training set data in partial discharge characteristic data is respectively input to initial radial basis function network model and initial long short-term memory network model for training;First test set data in partial discharge characteristic data is respectively input to the radial basis function network model after training and the long short-term memory network model after training, obtain the first output result output by the radial basis function network model after training, and the second output result output by the long short-term memory network model after training;Based on first output result and second output result, determine cable fault prediction model.The present application solves the technical problems of low cable fault prediction efficiency and poor prediction accuracy caused by poor performance of cable fault prediction model in related art.

The embodiment of the application provides a three-dimensional induction welding coil optimization design method and related equipment, and belongs to the field of advanced manufacturing and electromagnetic heating technology. The method comprises the following steps: discretizing a three-dimensional design space, defining geometric and topological constraints, and constructing a scalarization objective function; according to the obstacle layout, the homotopy class is divided, the representative curve is generated as an initial topological template; a differential homeomorphism mapping parameterized by a radial basis function network is constructed to drive the representative curve to deform geometrically; a serialization decision strategy is adopted, a fast proxy model based on the Biot-Savart law and a local gradient adaptive step are combined to iteratively optimize the deformation parameters at low cost; the candidate path is verified by high-fidelity finite element simulation, and the design meeting the engineering index and having the maximum and minimum performance margin is selected as the final result. The application significantly reduces the calculation cost while strictly guaranteeing the physical feasibility of the coil, and improves the efficiency and yield of complex structure design.

The invention discloses a chip electromagnetic interference prediction system and method based on adaptive scanning, and relates to the technical field of intelligent chips, the system comprises the following components: a two-dimensional automatic scanning platform, a low-frequency magnetic field probe, a signal acquisition module and a data processing and analysis module; the two-dimensional automatic scanning platform comprises an X-axis moving mechanism, a Y-axis moving mechanism, a stepping motor and an Arduino microcontroller, the X-axis moving mechanism and the Y-axis moving mechanism are both provided with a sliding rail system, the stepping motor is in driving connection with the corresponding axis moving mechanism, the Arduino microcontroller is electrically connected with the stepping motor, the Arduino microcontroller outputs a control signal to drive the stepping motor to operate, and the sliding rail system is electrically connected with the Arduino microcontroller. The stepping motor drives the low-frequency magnetic field probe to move in a plane in a chip radiation field; according to the method, electromagnetic interference space distribution reconstruction and prediction are realized based on the radial basis function network (RBFN) and singular value decomposition (SVD), the defect in the aspect of data modeling stability in the prior art is overcome, and the prediction precision is improved.

A method for reconstructing images, an electronic device and a storage medium is provided. In the method, the electronic device obtains an image to be reconstructed of an object captured by a target camera device, and obtains a sample image of the object. A radial basis function network is trained based on the image to be reconstructed and the sample image. An image reconstruction model is obtained by converting the radial basis function network based on a deconvolution algorithm, the image reconstruction model is configured for reconstructing images captured by the target camera device. By performing the method, a clarity of the image to be reconstructed can be improved.

The application discloses a kind of servo motor system control parameter intelligent optimization method, this method is by constructing radial basis function optimization network, with the characteristic parameter of given position trajectory as the input of optimization network, control parameter as output, directly to servo motor system multiple control parameters are optimized;Optimization process needs to train radial basis function network, with the mean square error of given position trajectory and actual position trajectory as loss function, the undetermined parameter of radial basis function network is updated using gradient direction propagation algorithm, when network parameter converges, the output control parameter is the optimal control parameter of optimization completion.Therefore, the feature of the present application is that network structure is simple, and the amount of calculation is small, without understanding servo system control structure, can efficiently complete the simultaneous optimization of multiple parameters.

This invention relates to the field of partial discharge pattern recognition technology, specifically to a partial discharge pattern recognition method based on adaptive mathematical morphology. The method first acquires partial discharge signal data from defective equipment and plots PRPD (partial discharge pattern recognition) maps. After grayscale and binarization processing, binary discharge images are obtained, and a dataset is constructed. Then, adaptive mathematical morphology parameter optimization is performed on the binary discharge images in the training set. The optimal results are applied to extract morphological change features and multi-dimensional morphological structure features from the binary discharge images in the entire dataset. Finally, the extracted features are weighted and fused, input into a radial basis function network for pattern recognition, and the recognition result is output. This invention effectively improves the accuracy of partial discharge pattern recognition through adaptive morphological parameter optimization and multi-dimensional feature fusion, and is suitable for insulation condition monitoring and fault diagnosis of power equipment.

A Yellow River Ningxia section ice flood prediction method comprises the following steps: S0, collecting daily weather, highest temperature, lowest temperature and daily average runoff information in a monitored river reach in a specific historical period, and constructing a data set by taking the weather, the highest temperature and the lowest temperature as input and the daily average runoff as output; s1, constructing an ice flood prediction model by adopting the data set and taking a support vector regression model and a radial basis function network model as basic models; s2, on the basis of the ice flood prediction model, a corresponding change rule between the daily average runoff volume and the weather, the highest air temperature and the lowest air temperature is obtained; s3, collecting daily weather, highest temperature and lowest temperature information in a future prediction period in the monitored river reach; s4, inputting the weather and temperature information in the S3 into an ice flood prediction model, and obtaining daily average runoff information of the monitored river reach in a future prediction time period; and S5, predicting the ice flood level of the monitored river reach in the future prediction time period.

The invention relates to the field of automobile transmission, in particular to a locking control method and system of a hydraulic torque converter based on a radial basis function. Acquiring working condition data; constructing a radial basis function network decision model according to the working condition data, and outputting a blocking coefficient based on the radial basis function network decision model; and the locking coefficient is compared with a locking threshold value or an unlocking threshold value according to the vehicle state, and a locking instruction or an unlocking instruction is sent to the hydraulic torque converter. According to the method, the radial basis function network decision model is constructed by obtaining the vehicle working condition data, the nonlinear fitting capability and the local response characteristic of the radial basis function network decision model are utilized, complex working condition characteristics are converted into locking coefficients to be output, and the energy loss in the hydraulic transmission stage can be reduced to the maximum extent according to the locking coefficient generation instruction; in addition, the pause feeling in the locking process is effectively restrained through the radial basis function network decision model, and particularly under the complex road condition that loads change frequently, the driving comfort is improved.

This invention relates to the field of materials data modeling technology, specifically to a method for constructing a thermal processing map for metallic materials. The method includes the following steps: calculating the true stress-strain sequence; fitting the true stress-strain sequence to a continuously differentiable three-dimensional rheological response surface based on a radial basis function network; performing analytical differentiation on the three-dimensional rheological response surface to extract global strain rate sensitivity field data; calculating the power dissipation efficiency matrix and rheological instability judgment parameters based on the strain rate sensitivity field data; and then generating a thermal processing map. In this invention, by establishing a continuously responding surface model based on global optimization and using analytical derivative solutions to replace the traditional differential calculation of discrete data, the numerical oscillation phenomenon caused by experimental noise is effectively eliminated, ensuring thermodynamic consistency in the constitutive relation calculation process, significantly improving the geometric accuracy of the power dissipation efficiency contour lines, and achieving accurate prediction of the safe processing window for metallic materials under complex thermal deformation conditions.

The invention provides a gas extraction evaluation unit multi-parameter collaborative automatic metering method and system. The method comprises the steps that gas concentration, multi-path mixed gas flow velocity, pipeline pressure and multi-component gas data are collected, a radial basis function network is input, gas density-viscosity correction parameters and cross interference correction parameters are output, and a flow velocity compensation coefficient is calculated; calculating a first-order difference norm and fluctuation entropy of a gas concentration sequence in a rolling time window, and judging a signal state according to a joint criterion: correcting the concentration by using a correction parameter when the signal is stable, decomposing and reducing noise by using a wavelet packet when the signal fluctuates, and uniformly outputting and correcting the gas concentration; constructing a flow velocity-pressure covariance matrix, calculating a characteristic value time change rate, generating a confidence factor in combination with a fluctuation entropy, carrying out weighted fusion on each path of flow velocity to obtain a reference mixed flow velocity, and carrying out correction by using a compensation coefficient to obtain a compensated mixed flow velocity; and aligning the time sequences of the corrected gas concentration and the compensated mixed flow velocity, and calculating the gas extraction pure quantity based on the two time sequences.

The disclosure relates to a multi-source target identification method based on single-channel nonlinear blind source separation. The method comprises the following steps: establishing a single-channel nonlinear blind source separation model by using multi-interference source information collected by a receiving device at the same time; converting the multi-interference source information into multi-source nonlinear aliasing signals; performing blind source separation processing on the multi-source nonlinear aliasing signals to obtain separated signals; and performing signal identification processing on the separated signals to obtain multi-source target identification results; wherein the blind source separation comprises a self-organizing feature mapping network and a radial basis function network. The multi-source target identification method based on single-channel nonlinear blind source separation can effectively solve the mode aliasing problem and the end effect problem, thereby improving the similarity between the separated signals and the multi-interference source information.

The application belongs to the field of robot control, and discloses a kind of polishing robot end variable stiffness actuator force planning and active stiffness matching anti-disturbance control method and system, the method comprises: stable contact before force planning method, stable contact after force planning method, active compliance anti-disturbance force control method;The stable contact before force planning method is by controlling the total impulse of input system to thereby constrain the maximum collision force size generated collision;The stable contact after force planning method is by considering environmental stiffness constraint, actuator stiffness constraint, robot dynamics and kinematics constraint, actuator drive error modeling, material removal constraint, optimization is carried out with all constraints to optimal efficiency T as target, obtains optimal force, position, stiffness control target value;The active compliance anti-disturbance force control method includes a kind of noise and disturbance suppression environmental stiffness estimator, stiffness controller, sliding film controller, radial basis function (RBF) network, extended state observer (ESO), switching controller.The application shows excellent noise suppression and stiffness estimation accuracy.In robot polishing, it shows excellent force control accuracy.

The invention discloses an online kernel learning target positioning method based on data driving, which comprises the following steps of: firstly, projecting a TDOA (Time Difference of Arrival) measurement vector into a regeneration kernel Hilbert space by utilizing TDOA radar measurement, and converting a target positioning problem into a function approximation problem in a feature space; a KLMS-TDOAL algorithm based on kernel least mean square TDOA positioning is provided to solve the problem, an online vector quantization method is introduced to obtain the KLMS-TDOAL algorithm so as to control the size increase of a radial basis function network, and finally the effectiveness and superiority of the algorithm are evaluated through numerical simulation. The method enriches the technical framework of distributed MIMO radar target positioning, provides a new effective technical means for coping with positioning challenges in a low signal-to-noise ratio environment, not only improves the reliability and accuracy of a positioning system, but also provides a new thought and method for the development of the distributed MIMO radar technology, and has wide application prospects.

The present application belongs to the technical field of production scheduling optimization, and discloses a mixed flow assembly processing system batch plan optimization method based on material matching, which comprises the following steps: establishing a mixed flow assembly processing plan integrated optimization model with the minimum finished product inventory and work-in-process inventory cost as the target; setting total cost confidence and external demand satisfaction confidence constraints to re-describe the mixed flow assembly processing plan integrated optimization model as a two-level fuzzy chance-constrained programming model; rewriting the two-level fuzzy chance-constrained programming model into a multi-level programming model; and solving the multi-level programming model based on a hybrid algorithm of a radial basis function network, a particle swarm algorithm and a cross-entropy method. The present application can maintain the lowest work-in-process inventory and finished product inventory cost, and maximally reduce product delivery delay cost.

The application belongs to the technical field of optical detection, and particularly relates to a Ritchey-Common detection method and system based on global power wave aberration prediction. The method first establishes a three-dimensional optical model through a standard spherical mirror, a to-be-detected plane mirror and interferometer parameters, combines a preset Ritchey angle and a defocus / astigmatism coefficient, generates an enhanced training data space by using Monte Carlo simulation to inject vibration and temperature noise; secondly, a phase error adjustment model is constructed based on a generative adversarial network, a multi-scale attention mechanism and a radial basis function network are used to decouple and adjust the mapping relationship between the error and the wave aberration distortion, so that the adaptive correction of the sub-aperture phase data is realized; finally, a graph neural network is used to splice the sub-aperture phase and fuse the multi-angle full-aperture measurement data, and the to-be-detected plane mirror surface is accurately reconstructed; through the physical constraint adversarial training and dynamic error compensation, the detection efficiency and reliability of the large-aperture optical element are significantly improved.

The invention relates to the technical field of calibration, in particular to a dynamic intelligent optical sighting telescope calibration method and system, and the method comprises the following steps: sorting initial data of a sighting telescope based on collected environment parameters, obtaining environment state information, carrying out the data filling of an input layer of a radial basis function network according to the environment state information, carrying out the preliminary training of a model, and carrying out the calibration of the optical sighting telescope. The weight and deviation of the model are adjusted to obtain a preliminary adjustment model; according to the method, environmental parameter analysis and real-time data adjustment are integrated, the adaptability and accuracy of the calibration process are enhanced, dynamic adjustment is performed by using environmental state information, a calibration model is allowed to be automatically optimized when external conditions change, the accuracy is ensured to be closely matched with an operation environment, and the calibration setting is automatically updated, so that the calibration efficiency is improved. The operation complexity is reduced, the automation level of the calibration process is improved, the calibration of the sighting telescope is quicker and more accurate, and the use effect of the sighting telescope in various environments is enhanced through an automatic data feedback and performance adjustment mechanism.

The invention belongs to the technical field of electro-hydraulic servo control, and provides a backstepping control method and system for an electro-hydraulic servo system, and the method comprises the steps: building a mathematical model of the electro-hydraulic servo system; controlling the electro-hydraulic servo system by using a preset backstepping controller according to the established mathematical model of the electro-hydraulic servo system; according to the method, a command filter is utilized to obtain a derivative of a virtual input variable and a virtual control law differential signal, so that the problem of'complex explosion 'in standard backstepping control is avoided, and estimation of non-matching uncertainty is realized through reconstruction; meanwhile, compared with the traditional back-stepping control based on the neural network which needs a plurality of neural networks, the method only utilizes one neural network of the radial basis function network to approach the matching uncertainty, so that the complexity and vulnerability of the controller caused by the plurality of neural networks are avoided.