641results about How to "Improve forecast accuracy" patented technology

The invention discloses a machine learning model training method and device. The method comprises the following steps: on the basis of initialized first weight and second weight of each sample in a training set, and with features of each sample as granularity, training a machine learning model; on the basis of prediction loss of each sample in the training set, determining a first sample set, where corresponding target variables are predicated inaccurately, and a second sample set, where corresponding target variables are predicated accurately; on the basis of the prediction loss of each sample in the first sample set and the corresponding first weight, determining overall prediction loss of the first sample set; on the basis of the overall prediction loss of the first sample set, improving the first weight and the second weight of each sample in the first sample set; and inputting the updated second weight of each sample in the training set and features of each sample and the target variables into the machine learning model, and with the features of each sample as granularity, training the machine learning model. Through the machine learning model training method and device, prediction accuracy and training efficiency of the machine learning model can be improved.

The invention discloses a short-time traffic flow prediction method based on a long-time and short-time memory recurrent neural network. The method comprises the following steps: aggregating inputted historical traffic flow data according to a prediction time interval of a short-time traffic flow; carrying out pretreatment on the aggregated historical traffic flow; setting a reasonable parameter for a long-time and short-time memory recurrent neural network; training a neural network prediction model by using data after pretreatment; and invoking a traffic flow, predicted by the prediction model, of a designated time interval and evaluating a prediction error. According to the invention, because the long-time and short-time memory recurrent neural network has the advantage of being capable of memorizing inputted historical data for long time, the high prediction precision is realized; and the expansibility for different prediction intervals is good.

The invention relates to a lithium ion battery residual life forecasting method of a dynamic gray related vector machine. The lithium ion battery residual life forecasting method of the dynamic gray related vector machine solves the problem of an existing lithium ion battery residual life forecasting method. The problem of poor accuracy of a state transition equation forecasting is solved only by giving single-point forecasting value and an empirical model. Firstly, a gray model is adopted to conduct trend forecasting on small sample data, then regression forecasting is conducted by a related vector machine, and finally the forecasting model is updated dynamically by related analysis, on the basis of a combined model, a short-term forecasting result is continuously updated into a training data series, the related analysis is conducted, and training is conducted again according to correlation, so that accuracy of multistep iteration forecasting is further improved. The lithium ion battery residual life forecasting method of the dynamic gray related vector machine is applied to the field of a lithium ion battery.

The invention discloses a thunder and lightning approach forecasting method based on a particle swarm support vector machine, relates to the technical field of thunder and lightning forecasting and aims at applying a particle swarm support vector machine method in the thunder and lightning approach forecasting. The method comprises the following steps: carrying out relevance analysis and selecting related factors which influence the occurrence of thunder and lightning from the overhead and ground historical information of an MICAPS (Meteorological Information Comprehensive Analysis and Processing System) and the actual thunder and lightning data of a ground station; preprocessing the data and reasonably interpolating missing data aiming at the characteristics that data which prove whether thunder and lightning occur or not in the thunder and lightning data are imbalanced; optimizing the parameters of the support vector machine by a particle swarm optimizing algorithm; establishing a training sample set, training the support vector machine and establishing a thunder and lightning approach forecasting model; inputting a test data set into the trained forecasting model, so as to judge whether thunder and lightning occur or not. The method has the advantages of high precision and strong generalization capability.

The invention discloses an enhancement type bi-directional motion vector predicting method aiming at interframe predictive coding in a mixed video coding framework, comprising the following steps of: (1) solving an alternative set containing a plurality of bi-directional motion vector prediction values through an prediction operator by comprehensively utilizing the time domain correlation and the space domain correlation of motion information for a current coding module; and (2) selecting an optimal motion vector prediction value from the alternative set by adopting a self-adaptation motion information prediction selection method to carry out subsequent coding computation and decoding computation as the motion vector prediction of the current coding module. The invention enhances the prediction accuracy of a final motion vector by comprehensively utilizing the correlated characteristics of the motion information on time domains and space domains and can independently complete on coding and decoding ends without detuning by utilizing a self-adaptation selection technology based on a prediction coding block SAD standard, thereby enhancing the compressive property of the bi-directional interframe predictive coding and being widely used for the field of mixed video communication.

The invention discloses a graph convolutional neural network model and a vehicle trajectory prediction method using the same. The model is composed of an encoder module, a spatial information extraction layer module and a decoder module. The method comprises the following steps: firstly, sampling a predicted vehicle and surrounding vehicles in a traffic scene at a frequency of 5Hz, and collectingposition coordinates and kinetic parameters of each vehicle sampling point, including horizontal and longitudinal coordinates, horizontal and longitudinal vehicle speeds and accelerations; calculatingcollision time TTC between the predicted vehicle and surrounding vehicles according to the coordinates and speeds of the predicted vehicle and the surrounding vehicles, and judging vehicle behaviors;inputting each historical track of the vehicle containing the information into the model, encoding time sequence interaction features in the track, extracting spatial features, summarizing the features into context vectors, and inputting the context vectors into an LSTM decoder to generate future track coordinates of the vehicle. According to the method, the problem that feature information generated by vehicle interaction cannot be obtained by using a traditional recurrent neural network is solved, and the prediction precision of the vehicle trajectory is greatly improved.

The invention discloses a human body posture estimation method and system based on structure guidance deep learning, and belongs to the field of computer vision and machine learning. A problem to be solved by the invention is to provide a method for guiding learning by utilizing human skeleton structure knowledge by aiming at shortages of low accuracy and bad robustness caused by the fact that a neural network generally directly predicting human body articulation points does not contain human body structure information. The method mainly comprises input data preprocessing, structure-guidance convolutional neural network predicting, converged convolutional neural network predicting and final output processing. According to the human body posture estimation method and system based on the structure guidance deep learning, the learning is guided by importing priori knowledge of a human body structure, and a human model is implicitly learned, so that human body postures can be accurately recognized, a certain complex postures and an occlusion problem can be accurately recognized, and high robustness is achieved.

The invention provides a method for constructing a photovoltaic power station generation capacity short-term prediction model based on multiple neural network combinational algorithms and belongs to the technical field of photovoltaic power generation, power grid connection technology and solar energy photovoltaic forecasting. The method overcomes the problem that a usually-used algorithm for constructing the photovoltaic power station generation capacity short-term prediction model is single and is likely to fall into local optimization, further resulting in big measurement error of the prediction model. The technical construction method of the invention is realized as follows: firstly using four different neural network algorithms to construct sub-models for neural network prediction; secondly screening and classifying weather information and analyzing the suitability of the various sub-models for neural network prediction; giving weighted parameter values of the sub-models in a combined model according to the suitability to further make the combined neural network model for prediction suitable for different weather conditions and then completing the construction of the photovoltaic power station generation capacity short-term prediction model. The method is mainly used for photovoltaic power station grid connection short-term prediction.

The invention discloses a short-time traffic flow prediction method based on space-time correlation and a convolutional neural network. The method comprises: mining space-time correlation in the traffic flow data; traffic flow data is converted into a two-dimensional matrix with space-time traffic flow information, the space-time characteristic information of traffic flow in the matrix is extracted, processed and learned by utilizing the convolutional neural network, a prediction result is finally obtained, and in addition, optimal input data is screened out by adopting a space-time characteristic selection algorithm. The randomness and uncertainty of the traffic flow data are deeply mined, and the space-time correlation in the traffic flow data is fully considered, so that the predictionprecision of the traffic flow data is effectively improved.

The invention discloses an image question-answering method based on multi-scale deep learning. The method is enlightened by the cognitive behaviors of people during the image question-answering process. The method copmrises the following steps of 1) setting the same picture as three pictures different in scale size according to requirements, adopting a pre-trained convolution neural network for extracting picture features and obtaining multi-scale picture features; 2) obtaining the feature expression of an interrogative sentence by adopting a recursion neural network, and acquiring n-element interrogative sentence feature expression by connecting a convolution layer with different convolution kernel sizes; 3) exploring the built-in incidence relation between picture features of different scales and n-element interrogative sentence feature expression, namely the similarity measurement; 4) finally, fusing picture features of different scales and n-element interrogative sentence features,and deducing the answer of a predicted problem from the large scale to the middle scale and the small scale by adopting a hierarchical network structure. According to the invention, the cognitive behaviors of people during the image question-answering process are simulated. Meanwhile, the high precision is obtained on the basis of a reference data set.

The invention discloses an intelligentized operation fault prediction method of an elevator control system. The method comprises the following steps: (1) establishing a typical fault prediction expert knowledge base according to correlation analysis of elevator operation monitoring signals and sensor data, and various faults, and carrying out fault prediction through knowledge base rule reasoning; (2) monitoring relevant signals of a controller and the sensor data, collecting the monitoring data, and standardizing the monitoring data to use as sample data of neural network learning; (3) establishing a model of a hierarchical successively-connected neural network system for fault prediction, and utilizing the collected sample data for neural network training and learning; and (4) inputting real-time signals into the fault prediction system, and then fusing a fault prediction result obtained by expert knowledge reasoning and a fault prediction result of hierarchical successively-connected neural networks. According to the method, the accuracy of prediction is improved, the operation stability and safety of an elevator system are greatly improved, a technical threshold of elevator control system repair and maintenance is reduced, and elevator maintenance and repair are enabled to be more accurate, simpler and faster,

The invention discloses a short-term power load forecasting method based on a fuzzy clustering similar day. The method includes that firstly meteorological factors are divided into temperature, pressure, wind speed, rain and other occasions and then constitute influence factors of the similar day with week styles and date styles, a fuzzy coefficient characteristic mapping table is built through fuzzy rules, clarification is carried out by the method of fuzzy clustering based on the preceding steps, the similar day is chosen according to clustering levels, according to obtained load data of the similar day, load sequences are projected to different scales and low frequency components are obtained by means of wavelet transformation, a support vector machine is optimized by means of a particle swarm optimization algorithm to achieve forecasting for a short-term power load low frequency portion, and forecasting for a high frequency portion is achieved by the method of weighted average. Eventually, application researches are carried out by means of the load data of a power grid in Shanghai city, and good forecasting effects can be achieved in weekdays, at weekends and in holidays.

The invention discloses a neural network method for performing short-term earthquake prediction by using earthquake parameters, which is a method for predicting earthquakes by using regression analysis and neural network combined technology and comprises: (1) determining a region to be studied; 2) collecting and preprocessing earthquake catalogues; 3) acquiring the earthquake parameters and time varying parameters by using the earthquake catalogues; 4) sorting information; 5) performing regression analysis; 6) performing neural network simulation; and 7) computing earthquake magnitudes of predicted earthquakes. The method has the advantages that: regional short-term earthquake prediction can be realized and the prediction time is no more than 6 months; earthquake magnitude quantization prediction can be realized, the computing result is stable and the prediction precision is high; the maneuverability and practicality are high; and the regional short-term earthquake prediction result can be used to provide service for 'disaster prevention and reduction', and the social and economic benefits are obvious.

The invention discloses a wind power combined predication method based on ensemble average empirical mode decomposition and an improved Elman neural network, and belongs to the technical field of wind power prediction. The method comprises the following steps: step 1, extracting wind speed sequence historical data and normalizing the data; step 2, making sequence analysis of the extracted wind speed sequence historical data through empirical mode decomposition; step 3, reconstructing a phase space for the sequences obtained through empirical mode decomposition; step 4, cyclically selecting the number of nodes in a hidden layer to train an Elman neural network, and superposing the prediction results of the sequences to obtain a wind speed prediction result; and step 5, making error analysis of the wind speed prediction result. The modeling process of the method is simple and practical. By adopting the method, the wind power can be predicted quickly and effectively. The method is of great significance to the security and stability and dispatching operation of a power system under the condition of wind power grid-connection.

The invention provides an inertia-assisted multi-frequency multi-mode GNSS cycle slip repair method and system. The method comprises the steps that cycle-slip observation is conducted on all satellites to judge the satellites with the existence of cycle slip; after cycle-slip parameters are determined, a pseudo space and phase position non-difference uncombined epoch difference observation equation on frequencies and systems is formed, and a clock difference variation parameter is introduced; according to the activity degree of an ionosphere layer, fitting functions with different orders in a sliding window are used to model and forecast the time-varying ionosphere layer; inertia-assisted cycle slip calculation is conducted, and posteriori residuals of the cycle slip repair equation are tested; a three-step method is used to fix cycle slip values and carry out the cycle slip repair, and finally phase position observation values are repaired. The inertia-assisted multi-frequency multi-mode GNSS cycle slip repair method and system can accurately repair cycle slip values on different frequencies of different GNSS systems in a dynamic and complicated environment, and provide clean and pollution-free observation data for subsequent positioning and calculating treatments.

The invention provides a stratum aperture pressure prediction method based on variety earthquake attributes, which indirectly constructs the non-linear relation between the stratum aperture pressure and the variety earthquake attributes and realizes the prediction of the stratum aperture pressure. The advantages of the invention reside in two aspects. On one hand, the invention comprehensively uses variety earthquake attributes, and deducts other influence factors, like porosity, shale content, etc, while utilizing the inversed longitudinal wave speed to carry out stratum aperture pressure prediction so as to improve the prediction accuracy; on the other hand, the invention creatively uses two calculation methods to obtain the longitudinal wave speed, obtain the target sand-mud stratum longitudinal wave speed through high precision wave impedance inversion and the obtains the background longitudinal wave speed by using the DIX formula inversion. The former can relatively accurately inverts the longitudinal wave speed inside the target stratum so as to relatively calculate the effective stress Pe inside the target layer; and the later can reflect the relation between the longitudinal wave speed and the density as a whole so as to accurately calculate the overlying strata pressure Pov.

The invention discloses a mechanical equipment fault intelligent early warning method based on multivariate estimation and prediction. The method comprises the following steps: step 1, modeling a datasubset; step 2, preprocessing the modeling data subset; step 3, constructing a prediction model of the mechanical equipment state parameters and the working condition parameters; step 4, estimating and predicting the state parameters of the mechanical equipment corresponding to the actually measured state parameters in a normal operation state; step 5, subtracting the actual measurement result ofthe state parameter from the prediction result to obtain a residual value of the state parameter, judging whether the absolute value and the growth trend of the residual value exceed corresponding threshold values or not, further detecting the fault abnormality of the equipment and giving an alarm; the intelligent early warning model of the mechanical equipment is established based on the multivariate estimation prediction method, and then intelligent early warning of the variable working condition mechanical equipment fault is achieved. Compared with a traditional mechanical equipment faultearly warning method, the method has the advantages of being high in prediction precision, high in early warning accuracy and more timely in early warning.

The invention provides a method for controlling fine rolling thickness of hot continuous rolling. The method comprises the steps of acquiring rolling machine parameters and steel strip specification parameters; performing unit step response test for the tail rack rolling machine; determining the unit step response cycle which is the time parameter of hydraulic cylinder transmission function; monitoring the control cycle of an AGC (Automatic Gain Control) system and the quantity of sampling dispersing points after unit step response lag; controlling the tail track rolling machine through the Smith pre-estimation control strategy of a proportional plus integral controller with an inertial element; monitoring an AGC system control model by fine rolling of the hot continuous rolling; controlling the thickness of the next cycle by adjusting a hydraulic system. According to the method, the AGC monitoring control process is treated as a control object with pure lag; the Smith pre-estimation compensation is introduced into the monitoring of the AGC control system; the GM method is carried out to directly perform soft measurement for roller gaps of the rolling machine; therefore, the possible calculation error caused by inaccurate HGC (hydraulic gap control) transmission function can be avoided; the response speed, stability and control precision of the control system can be obviously improved.

The invention discloses a dense clastic rock natural gas well productivity prediction method based on reservoir classification. The method includes the steps that the effective thickness of each reservoir and the effective thickness of each thin interbed are acquired according to logging information; the shale content of each reservoir is calculated according to a curve of natural gamma and three porosity; the porosity of each reservoir is calculated according to a curve of sound waves, neutrons and density, and the reservoirs are classified; the permeability and gas saturation of the reservoirs are calculated according to core analysis pore permeability and an Archie equation; reservoir classification calculation and summing are conducted through thickness weighting, and then a comprehensive gas index is acquired; the conception of the thin interbeds is introduced according to the differences of reservoir depositional microfacies, and then the improved comprehensive gas index is acquired; the relation between the improved comprehensive gas index and test productivity is built, and then the productivity of other reservoirs is predicted with the relation as a model. On the basis of ensuring simple operation, the method has high reservoir productivity prediction accuracy.

The invention discloses a data prediction method and device. The method comprises the steps of collecting data traffic information; analyzing and clustering the data traffic information through a preset protocol, thereby obtaining stability sequences; determining orders of the stability sequences by selecting corresponding quantization values, thereby obtaining a traffic prediction model; and predicting data traffic according to the traffic prediction model. According to the method and the device, the technical problem that the prediction precision of computing data is influenced due to the defect in the prior art is solved.

The invention discloses a digital twin driven numerically-controlled machine tool cutter monitoring system. The system comprises a cutter digital twin model construction module, a cutter stress simulation module, a cutter real-time state data acquisition and analysis module, a data preprocessing module and a residual life prediction module; the cutter digital twin model construction module constructs the virtual model of a data machine tool cutter; the cutter stress simulation module can simulate stress borne by the interior of the cutter in the cutting process of the cutter; the cutter real-time state data acquisition and analysis module can acquire environmental data and machining data in real time by configuring a data source and perform multi-dimensional analysis on historical data; and the data preprocessing module is mainly used for carrying out data cleaning, duplicate removal, dimension reduction, time domain feature extraction and frequency domain feature extraction on the acquired data; and the residual life prediction module is used for predicting and pre-warning against the wear state of the cutter based on the fused data of information physics. With the system disclosed by the invention adopted, the problem of the failure to grasp the wear state of the cutter and perform early warning in time in the milling process of the numerical control cutter can be solved, andthe management and control efficiency of the numerical control machine tool cutter is improved.

The embodiment of the invention discloses a network quality prediction method, a service adjustment method, related equipment and a storage medium, the network quality prediction method comprises thefollowing steps: obtaining the historical network quality of a to-be-predicted target geographic area in a historical time period, and the historical network quality is obtained based on report information submitted by a plurality of terminals; and calling a network quality prediction model to perform prediction processing according to the historical network quality to obtain the prediction network quality of the target geographic area in a prediction time period, the starting time of the prediction time period being later than the end time of the historical time period. According to the embodiment of the invention, the network quality prediction can be better performed, and the prediction accuracy of the network quality is improved.

The invention provides a user loss prediction method and system based on a cellphone game store, and the method comprises the following steps: collecting the basic information, behavior information and game information of users from a log of a server, and dividing the users into a training set user and a prediction set user; building a loss user tag for the training set user, and carrying out thepreprocessing of original data; carrying out the feature extraction, selection and standardization of the basic information, behavior information and game information of the training set user and theprediction set user; training a gradient increasing decision-making algorithm to obtain a user loss prediction model according to the features and loss user tag of the training set user; and recognizing a loss user of the cellphone game store through the user loss prediction model according to the features of the prediction set user. According to the invention, the method can quickly and accurately recognize the potential loss users based on the business scene of the cellphone game store, and provides a decision making support for the cellphone game store for the timely recall of the loss users.

The invention discloses a flood forecasting service construction method based on a WebGIS (Geographical Information System). The powerful space analysis technology of the GIS is fully utilized to be complementary with a traditional flood forecasting model to efficiently extract the natural converge information of the flood model so as to improve model forecasting accuracy. Meanwhile, WebGIS spaceinformation is used for expressing advantages so as to enable the flood forecasting to be more accurate and more visual in expression. The method comprises the following main steps that: (1) on the basis of terrain data, constructing a digital elevation model; (2) adopting a GIS space analysis method to preprocess a DEM (Dynamic Effect Model); (3) adopting a D8 algorithm to construct a running water network; (4) on the basis of a water flow direction and a soil water content, constructing a natural converge matrix; (5) utilizing a water flow accumulation value to generate a water channel in adrainage basin boundary, and carrying out hierarchical coding on the water channel on the basis of a Strahler method; (6) on the basis of a three-water-source XinAn River model, initially deciding andcalibrating parameters; and (7) on the basis of the WeBGIS, simulating a flood forecasting result.

The invention discloses a method for predicting fractal traffic flow combining weekly similarity characteristic. The invention comprises the following steps: 1) traffic flow data of different working days takes one week as a period, the traffic flow data are grouped to form traffic flow sequences with different directions at the same intersection in a scheduled period of time; 2) the scheduled time before the current time is extracted to the traffic flow sequences {Ni} of the current time, initialized n is equal to one, and {Si} is obtained through performing n-order cumulative calculation, {Sni}(i=1, ..., n)=N(A, epsilon) I, the obtained value is N(A, epsilon) i+1; 3) according to the traffic flow sequences in the same period of time a week ago, the traffic flow sequences in the same period of time two weeks ago, the traffic flow sequences in the same period of time three weeks ago, ... the traffic flow sequences in the same period of time m weeks ago, the calculations of the step 2) are respectively performed to obtain each predicted data, and the predicted data undergoes error correction to obtain the predicted result data. The invention provides a method for predicting fractal traffic flow combining weekly similarity characteristic with good real-time and high prediction precision.

The invention provides a time sequence anomaly detection method and device, electronic equipment and a storage medium. The time sequence anomaly detection method comprises the steps: carrying out themulti-window sampling of a current time sequence, and generating a corresponding current sequence matrix; encoding the current sequence matrix to obtain corresponding current encoding information; inputting the current coded information into attention processing to obtain the current coded information carrying the attention information; performing sequence reconstruction on the current coded information carrying the attention information to obtain prediction information of a next time sequence; obtaining a residual error sequence according to the actual measurement information and the prediction information of the next time sequence; and traversing each residual value in the residual sequence, and when the residual value is out of the reasonable interval, determining that the measured value in the next time sequence corresponding to the residual value is abnormal data. According to the time sequence anomaly detection method, the prediction accuracy of the regression model can be improved, and the problem of strong hypothesis of data distribution in an anomaly detection strategy is avoided.

The invention discloses an active tracking method, and belongs to the field of machine learning. The method comprises that an ith picture frame shot by a camera is obtained; the ith picture frame is input to a deep network which comprises a feature extraction network and a strategy function; a picture feature is extracted from the picture frame via the feature extraction network; output probabilities of n output motions are calculated respectively according to the picture feature via the strategy function, and n>=2; the maximal probability is selected from the output probabilities of the n output motions; and the camera is controlled to execute the output motion corresponding to the maximal probability, and it is supposed that i=i+1. A 3D virtual environment is used to train parameters inthe deep network, and the deep network is used to control motion of the camera according to the picture frames of the camera. Only one deep network is trained, linked parameter adjustment for a motionobject tracking module and a camera control module is not needed, and the workload of parameter adjustment is reduced.

The embodiment of the invention, which relates to the field of the internet technology, provides a method and apparatus for predicting a re-purchasing probability of a user and electronic equipment. The method for predicting a re-purchasing probability of a user comprises: carrying out learning on a training sample set to obtain a prediction model for a re-purchasing probability of a user; obtaining a feature data set of a to-be-predicted user; and using the feature data set of the to-be-predicted user as an input of the prediction mode and carrying out processing to obtain a re-purchasing probability prediction value of the to-be-predicted user by the prediction model. Therefore, on the basis of multi-feature-dimension data of a user, the re-purchasing probability prediction value of the to-be-predicted user can be obtained automatically by the prediction model, so that the prediction accuracy is improved and the prediction efficiency is enhanced.

The invention relates to a space-time related air quality prediction method. The method comprises: dividing a city into grids with the same size, each grid being influenced by grids in adjacent areas,and dividing the grids into a prediction area and an estimation area according to whether air monitoring stations exist in the grids or not; obtaining related time series data influencing the air quality data, conducting feature extraction on the time series data through a recurrent neural network and spatial deep learning, and conducting time series model training; obtaining related non-time-series data influencing the air quality data, conducting feature extraction on the non-time-series data through a convolutional neural network, and conducting space model training; performing cooperative training on the time sequence model training and the space training model to obtain a prediction model; training a prediction area by using the trained collaborative training model to obtain air quality data of the prediction area; and training an estimation grid region by using the trained collaborative training model to obtain air quality data of the estimation region.

The present invention relates to a reliability prediction method for the power supply of a power distribution network. The method comprises the following steps: selecting a variety of original index factors which influence the power supply reliability; subjecting the original index factors to standardization treatment to obtain standardized index factors; calculating a correlation coefficient between each standardized index factor and a preset reliability evaluation index, and screening out a core index factor which obviously influences the power supply reliability from all index factors according to the correlation coefficient; according to the core index factor, constructing a radial primary function neural network; obtaining the optimal center vector, the optimal base width vector and the optimal output weight of the hidden layer of the radial primary function neural network; according to the optimal center vector, the optimal base width vector and the optimal output weight, obtaining a power supply reliability prediction model; and predicting the power supply reliability of the power distribution network according to the power supply reliability prediction model.