48 results about "Sum squared error" patented technology

The invention relates to a ship autopilot composite neural network PID control method. The method comprises the following steps: providing a composite neural network four-layer structure which comprises an error sequence input layer, a network algorithm hidden layer, a network output layer and a network identification layer; determining the number of neurons and parameter values of each layer through open-loop test training learning, wherein the output layer is composed of three adjustable PID control parameters with non-negative values larger than zero; taking a performance index function ofthe error quadratic sum to set a gradient steepest descent method of a weighting coefficient, and adding an inertia term to prevent local convergence; increasing jacobian information of the output course of the ship to the input steering angle, improving the learning capacity and sensitivity to control characteristics, and achieving PID control parameter online self-adaptive adjustment. Accordingto the control method disclosed in the invention, the course keeping control performance of the ship uncertain motion can be improved; a neural network algorithm improved by Jacobian information identification based on a gradient steepest descent method and ship course keeping and embedding of a radial basis Gaussian function into a back propagation hyperbolic function can be used to solve the problems of large course deviation amplitude and high course reciprocating crossing frequency, and energy conservation and consumption reduction are realized.

The invention discloses a K-means three-dimensional clustering algorithm-based wind pressure coefficient rapid partitioning method and system, and a storage medium. The method comprises the steps of firstly obtaining wind pressure coefficient data of building surface points; establishing a K-means clustering algorithm model, and dividing the K-means clustering algorithm model into K clusters; respectively calculating the distance between each cluster center and the cluster center; minimizing an error quadratic sum of the clusters; calculating a clustering number K value range; finally, calculating a unified index parameter value and determining an optimal K value; and outputting cluster results. According to the wind pressure coefficient rapid partitioning method provided by the invention,on the basis of one-dimensional clustering of wind pressure extreme value gradient information, a certain weight is given to each parameter in a K-means clustering algorithm, the influence of spatialposition information is considered to assist wind pressure partitioning, and a k value selection range is reduced according to a method based on error sum of squares and contour coefficients. And anoptimal k value is determined by adopting a series of clustering indexes and engineering indexes. By means of the method, the wind pressure coefficient partitioning work can be well completed.

The invention discloses a state estimation method based on combination of a BP neural network and a D-S evidence theory. The state estimation method comprises the steps of: (1) constructing evidence of the D-S evidence theory by using an output result of the BP neural network; (2) proposing a way of using the output result of the BP neural network after normalization as evidence for determining basic credibility assignment in the D-S evidence theory since the D-S evidence theory is a powerful method for processing uncertain information; (3), and carrying out evidence combination calculation toobtain an estimated state. The state estimation method is simple, has good stability of estimation results, uses a value of a sum of square error function (SSE) of the BP neural network to representuncertainty evidence, and forms a plurality of BPA by adopting a plurality of BP networks.

The invention discloses a combined prediction method based on a double-period Holt-Winters model and a SARIMA model, and belongs to the field of wireless network flow prediction and network optimization. The method comprises the following steps: firstly, extracting m pieces of wireless network flow data of a base station, carrying out double-period Holt-Winters model prediction and SARIMA model prediction with a prediction step length of k by utilizing first m-k items of data, respectively storing the data into an array, respectively comparing original flow data at m-k-m moments with prediction result sums, and calculating error quadratic sums EDHW and ESA; determining weight coefficients of the DHW model and the SARIMA model by using an error quadratic sum reciprocal method; and similarly, carrying out double-cycle Holt-Winters model prediction and SARIMA model prediction on the first m items of data, storing prediction results at m + 1 to m + k moments into arrays yDHW and ySA, and carrying out weighted combination on the prediction results at the corresponding moments by utilizing weight coefficients wDHW and wSA to obtain prediction results ycombine [i] of subsequent k time granularities. And finally, the change condition of the flow data in the next k hours is observed by utilizing a prediction result. The method is high in calculation efficiency, and improves the stability and accuracy.

The invention discloses a modeling method for a support vector machine for a bridge crane through bulged ring crossover operation RNA-GA, and belongs to the field of intelligent modeling. The modelingmethod comprises the following steps: 1) acquiring the sampling data input and output by a bridge crane system through an experiment or field acquisition, and taking the error square sum of the expected output and the actual output as the target function of the RNA-GA; 2) abstracting out the modeling method for the support vector machine for the bridge crane through the bulged ring crossover operation RNA-GA through the enlightening of the molecular bulged ring structure of RNA; 3) setting the running parameters of an algorithm; and 4) running the bulged ring crossover RNA-GA to acquire the optimal values of the parameter estimation of a support vector machine model with the least square estimation model for the position of the bridge crane and a support vector machine model with the least square estimation model for the oscillating angle of the bridge crane separately, and substituting the optimal values into the support vector machine models with the least square estimation models to form a bridge crane model. The modeling method for the support vector machine has the advantages of being high in convergence rate, high in accuracy, and the like, and is also applicable to modelingfor other complex systems.

The invention discloses an IM/DD-OFDM/OQAM-PON system channel estimation method based on ANN_LS, and the method comprises the steps: firstly collecting time domain data, and carrying out the forward propagation through the time domain data; carrying out reverse propagation by using a gradient descent algorithm to continuously adjust W and b of the algorithm so as to minimize the sum of squares of errors of the whole algorithm and further obtain a transfer function (TF) of the system; and finally, carrying out channel estimation by using LS in a frequency domain. Simulation results show that when the optical fiber transmission distance is long, compared with an algorithm based on ANN channel estimation, higher CE precision can be obtained based on the ANN_LS algorithm, and better system optimization performance is achieved.

The invention discloses an automatic route grid evaluation method, electronic equipment and a storage medium, and the method comprises the steps: reading and processing data source information and travel data, and storing the data source information and travel data to a big data platform; for the travel data, selecting indexes capable of describing the aviation market, analyzing the aviation market in a clustering mode, dividing the aviation market, and calculating parameters of an air route network evaluation market by combining Logistic regression and a deep neural network method in each obtained classification; calling the parameters and the flight plan, and calculating the market share of the journey and the passenger rate of the flight; if the information of a certain flight is manually adjusted, recalculating the index of the flight route aviation market where the flight is located; and judging the category of the cluster of the route network evaluation market where the route islocated in an error quadratic sum mode and storing the category. From the perspective of automation, an original mode based on subjective experience is converted into a calculation mode based on a model, manual intervention is reduced, the transport capacity arrangement of airlines is better adjusted, and high fault tolerance is achieved.

An electric steering engine simulation modeling method based on parameter identification aims at solving the problems that accurate simulation data is not easy to obtain due to the fact that friction link analysis is complex and rotational inertia influence factors are many in actual modeling; instead of analysis step by step, an overall equivalent mode is adopted, wherein the rotational inertia and the friction coefficient of the multi-stage links are concentrated on two parameters for overall conversion. In this way, the rising time Tr and the overshoot sigma of a typical input signal are used as target values, and a Matlab M script file is used for programming simulation. Then, according to mechanism analysis of the electric steering engine, through Laplace transformation derivation, a preliminary simulation model is built, an iterative calculation program is written, and the rise time Tr and the overshoot sigma which are calculated through simulation each time are compared with target values; and the deviation degree is calculated by adopting an error sum of squares minimum principle, and the rotational inertia and the friction coefficient meeting the error range requirement are searched, so that the purpose of establishing an accurate model is achieved.

The invention discloses a differential privacy protection clustering method based on Kmeans. The method comprises the steps of data preprocessing, clustering iteration and differential privacy; wherein each iteration needs to be carried out as follows: setting a privacy budget, calculating noise, calculating the sum of data points of each cluster and the number of points, and then adding the calculated noise to the data points; for each iteration, maintaining the upper bound and the lower bound of one privacy budget, selecting adaptive privacy budget allocation between the upper bound and thelower bound according to the availability and the privacy protection degree, and repeating the steps until the sum of squares of errors converges or reaches the upper limit of the number of iterations. According to the method, in the iteration process of the Kmeans clustering algorithm, through adaptive privacy budget allocation, data privacy is protected on the basis of data distortion, and dataavailability is guaranteed.

According to the combined evaluation method based on the feedback adjustment weight, the difference characteristics and the similarity characteristics between different objects to be evaluated can be fully reflected, and scheme decision making and sorting are facilitated. The evaluation method comprises the following steps: obtaining an initial weight value based on a deviation maximization combined evaluation method; adjusting the initial weight by taking the weighted error sum of squares as an optimization target to obtain a final weight combination; using the final weight combination for combined evaluation, obtaining a comprehensive evaluation result, so that the difference characteristics between different objects to be evaluated are fully reflected; and meanwhile, enabling the sum of squares of weighted errors to be minimum, thus, the similarity characteristics between different objects to be evaluated are fully reflected. According to the evaluation method, the combination result consistency is higher, and the combination effect is better. The whole evaluation process is clear in concept and definition, and the calculation method is simple and convenient to apply.

The invention discloses an earth pressure balance shield tunneling machine propelling speed control method, a modeling method and a modeling device. Aiming at a propelling speed subsystem of an earth pressure balance shield tunneling machine, with a propelling speed control quantity used as a control input variable, a soil bin pressure, a total propelling force and a cutter head rotating speed used as measurable interference variables, and the actual propelling speed of the shield tunneling machine used as an output variable, an RBF-ARX model of the propelling speed subsystem is constructed. The optimal RBF-ARX model parameters and structure for describing the dynamic characteristics of the propelling speed of the shield tunneling machine are obtained by minimizing the sum of squares of errors between predicted output of the RBF-ARX model and the actual value of the propelling speed under the constraint condition of a propelling speed dynamic response mode through utilizing actual sampling data of the propelling system of the earth pressure balance shield tunneling machine. A propelling speed prediction controller is designed based on the RBF-ARX model, and technical support is provided for achieving automatic control over the propelling speed of the earth pressure balance shield tunneling machine.

The invention relates to a turnover equipment consumption optimal combination prediction method based on an uncertain weight. The method comprises the following steps: respectively constructing single prediction models by adopting a three-term moving average method, a four-term moving average method, a primary exponential smoothing method, a secondary exponential smoothing method, a cubic exponential smoothing method, a grey system prediction method and a unary regression analysis method; constructing a combined prediction model by using different numbers of single prediction models; performing optimization combination on the seven different single prediction models, comparing weighting coefficients which are respectively solved by adopting an arithmetic average method, a simple weighted average method and a quadratic programming method, and selecting an optimal weighting coefficient as a weighting coefficient of the combined model; and comprehensively evaluating the combined prediction model by adopting the error sum of squares, the mean square error and the Hille unequal coefficient, selecting one group with the minimum prediction error as an optimal combined prediction model, and predicting the consumption of the turnover equipment according to the optimal combined prediction model. The method is high in precision of predicting the consumption quantity of the turnover equipment.

The invention provides a wind direction clustering method and device and electronic equipment, and the method comprises the steps: obtaining the periodic wind direction data of each fan of a wind power plant; determining a similarity measurement index suitable for the periodic wind direction data according to the periodic wind direction data; based on a similarity measurement index, selecting k clustering centers, and dividing the periodic wind direction data into k clusters; determining an error sum-of-squares criterion function based on the periodic wind direction data; and evaluating the clustering effect of the k clusters based on an error sum of squares criterion function, and determining a final clustering effect. The similarity measurement index and the error sum of squares criterion function suitable for the periodic wind direction data are determined through the periodic wind direction data, so that the defect that a traditional K-means algorithm is not suitable for clustering the periodic wind direction data is overcome, and the reliability and authenticity of a clustering result are ensured; and reliable and reasonable fan division can be obtained according to a clustering result, so that the subjectivity of manual sector division is avoided.

The invention discloses a method for processing measured data of a multispectral radiation thermometer, solves the problem that an existing data processing method of the multispectral radiation thermometer is only suitable for a few materials, and belongs to the technical field of spectral data processing. The invention relates to a method for processing measured data of a multispectral radiation thermometer. The method comprises the following steps: S1, acquiring measured data xk of the multispectral radiation thermometer and a corresponding real temperature yk; s2, according to xk and yk, fitting by using y = a0 + a1x +... + amxm to obtain a residual error sum of squares Q; s3, removing the mth term, and fitting by using y = a0 + a1x +... + am-1xm-1 to obtain a residual error sum of squares Q '; and S4, acquiring a variance ratio Fm according to the Q and the Q ', judging whether the Fm is greater than a critical value Falpha, if so, determining that m is equal to m + 1, and turning to S2, otherwise, according to the value of the current m, adopting y = a0 + a1x +... + am-1xm-1 as a fitting curve model of the measurement data of the multispectral radiation thermometer.

The invention provides a transformer area line loss data processing method and device based on a Canopy-Kmedoids algorithm, and a medium, and the method comprises the steps: collecting basic data information of a power distribution network transformer area, wherein the basic data information comprises static parameters of the transformer area, and screening a transformer area sample set with a stable line loss rate based on the static parameters; and analyzing transformer area line loss rate influence factors based on the basic data information, selecting transformer area static parameters with a large influence degree, adopting a Canpoy algorithm to carry out coarse clustering on a data set, taking a central point of a Canpoy subset as an initial central point of a K-medoids algorithm to carry out fine clustering, calculating the sum of squared errors (SSE) of each clustering scheme, making an elbow diagram according to the sum of squares of errors of each clustering scheme, selecting a K value at an inflection point of the elbow diagram as an optimal K value as an optimal result, performing line loss calculation of the transformer area by using the optimal result, and controlling a power grid according to a line loss calculation result. The method improves the clustering effect, and enables the line loss calculation to be accurate and objective.