53 results about "Autologistic regression" patented technology

Multi-step statistical modeling in one embodiment of the present disclosure enables anomaly detection, forecasting and / or root cause analysis of the energy consumption for a portfolio of buildings using multi-step statistical modeling. In one aspect, energy consumption data associated with a building, building characteristic data associated with the building, building operation and activities data associated with the building, and weather data are used to generate a variable based degree model. A base load factor, a heating coefficient and a cooling coefficient associated with the building and an error term are determined from the variable based degree model and used to generate a plurality of multivariate regression models. A time series model is generated for the error term to model seasonal factors which reflect monthly dependence on energy use and an auto-regressive integrated moving average model (ARIMA) which reflects temporal dependent patterns of the energy use.

The invention discloses a method and a system for predicting a flow of a self-adaptive differential auto-regression moving average model. The method and the system are used for causing a model to be more fit with a data trend of a present flow. The technical scheme comprises the following steps: utilizing an ARIMA (Autoregressive Integrated Moving Average) model to forecast the flow, and alarming when a practical value is deviated from a predicted confidence interval; while alarming, starting an alternative plan to monitor a flow data, for preventing an abnormal data from entering into ARIMA model prediction; and when the ARIMA model normally runs, judging if a parameter of the ARIMA model is still suitable in real time, and if not, automatically relearning and acquiring a new model parameter by relearning, thereby promoting the accuracy for model prediction.

The present invention proposes an autoregressive model channel prediction method based on quantum computing, including: step 1, obtaining known channel values, setting the order of the AR model, and predicting the channel value at the next moment according to the current known channel value; step 2 : Use the least squares method to solve the estimated value of the AR model coefficient to obtain the calculation formula; Step 3: Use quantum computing to solve the calculation formula to obtain the AR model coefficient |d> of the quantum state; Step 4: Define a set of measurement operators M ,via caculation <d|m|d>Obtain AR model coefficient estimates. The invention combines the least square estimation for solving AR model coefficients with quantum calculation, can effectively reduce the complexity of classical calculation, and the calculation result is equivalent to the existing coefficient estimation effect.< / d|m|d>

The invention discloses a crawler scheduling method, which is applied in the technical field of computers, and includes: obtaining data parameters of webpages to be crawled, and calculating statistics of the data parameters according to time series, and based on the statistics, the statistics include the number of times , mean, variance, covariance, and autoregressive coefficients. Through the logistic regression algorithm and the FTRL algorithm, determine the scheduling time for crawling the data parameters of the web page next time, and update the scheduling task queue according to the scheduling time. The invention also discloses a crawler scheduling device, electronic equipment and a storage medium, which can improve crawling efficiency of the crawler.

The invention discloses a multichannel fusion detection method for distance extension targets under partially uniform environment, and belongs to the field of radar signal processing. According to themethod, a partially uniform multichannel auto-regression signal model is established for the condition of globally uniform auxiliary data missing caused by abnormal value conditions such as clutter edge and the like, so that local uniformity of clutter spaces, time low-order relevancy and multichannel information obtaining ability are sufficiently fused, the utilization rate of clutter prior information is effectively improved and a data model foundation is laid for auxiliary data demand of self-adaptive distance extension target detection; the multichannel fusion detection method for distance extension targets under partially uniform environment is established on the basis of a Rao test criterion, so that the auxiliary data demand is greatly reduced while the characteristic of constant-false alarm rate is ensured; and being compared with the common generalized likelihood ratio criterion method, the method is capable of simplifying the parameter estimation process while remaining thedetection performance unchanged, so that the detector construction efficiency is improved and the method is strong in engineering applicability and has popularization and application values.

The invention discloses a dynamic process monitoring method based on a BP neural network auto-regression model, aims at establishing a nonlinear auto-regression model by using a BP neural network andimplement dynamic process monitoring on this basis. The main core of the method of the invention firstly lies in that the BP neural network is utilized to identify an autocorrelation model of sampleddata of a monitored object, and secondly lies in that an error after filtering by the BP neural network auto-regression model is utilized to implement process monitoring. The method provided by the invention has the main advantages that firstly, a nonlinear auto-regression model is established by utilizing a nonlinear fitting capability of the BP neural network, thereby achieving the purpose of removing nonlinear autocorrelation characteristics in measurement variables; and secondly, the method provided by the invention utilizes a capability of the error of reflecting the abnormal change condition of the nonlinear autocorrelation characteristics, and since autocorrelation no longer exists in error data, convenience is provided for subsequent establishment of a fault detection model. Thus,the method provided by the invention is more suitable for dynamic process monitoring.

The invention discloses a method for quantitatively calculating complex spectral component concentration. The method comprises the following steps of: acquiring an absorption spectrum of a pure substance to obtain different characteristic peaks of the pure substance and correlation functions among the different characteristic peaks; collecting spectral data of a to-be-detected sample to obtain anoriginal spectrum; carrying out qualitative analysis on the to-be-detected sample, and screening out the corresponding characteristic peaks and characteristic peak correlation functions of the pure substance; and importing all the screened pure substance and the characteristic peak correlation functions of the pure substance into the original spectrum, and performing learning and autoregression through adopting a deep learning algorithm to obtain the component concentration. According to the method, different characteristic peaks of the single pure substance are associated, so that the calculation speed is effectively increased; deep learning algorithm regression can be carried out on all characteristic peak concentrations of different components on the spectrum at the same time, so that the calculation speed and accuracy are improved; and a deep learning algorithm is adopted for concentration regression analysis instead of substance identification, the calculation speed is further increased, and calculation data can be output in real time.

The invention belongs to the technical field of computer vision, and provides a fine-grained image classification method based on discriminant learning. We propose a new end-to-end autoregressive localization with discriminative prior network model that learns to explore more accurate discriminative patch sizes and is able to classify images in real-time. Specifically, a multi-task discriminative learning network is designed, including an autoregressive localization sub-network and a discriminative prior sub-network with a bootstrap loss function and a consistency loss function to simultaneously learn the self- Regression coefficients and discriminative prior map. The autoregressive coefficients can reduce the noise information in the discriminative patch, and the discriminative prior map filters thousands of candidate patches into single-digit patches by learning the discriminative probability value. Extensive experiments show that the proposed SDN model achieves the state-of-the-art in both accuracy and efficiency.