What Are the Different Types of Adaptive Devices?
Adaptive selection refers to the selection of conditions according to the different signals adaptively. Adaptive is to automatically adjust the processing method, processing order, processing parameters, boundary conditions or constraints in the processing and analysis process according to the data characteristics of the processed data, so as to adapt to the statistical distribution characteristics and structural characteristics of the processed data. The process of obtaining the best processing results.
- First, obtain all adaptable networks according to the terminal device; then determine whether the network can start adaptation, and record the networks that can start adaptive according to the set priority order of the adaptive network mode; then start the adaptation, then the terminal The device will compare the highest priority wireless network with the currently used network type based on the recorded priority order. If they are the same, do not switch, otherwise switch to the highest priority wireless network; repeat the above steps until the best or Required network environment.
- Adaptive is to automatically adjust the processing method, processing sequence, processing parameters, boundary conditions or constraints according to the data characteristics of the processed data during processing and analysis, so that it is compatible with the processed data.
- There are many simple models of model functions. In fact, some are more complicated. The identification, identification, parameter estimation and selection of these models are very heavy and complicated. Sometimes it can be used
- Through detailed calculation and analysis of economic data, the following characteristics of the data series can be found:
- (1) Data
Adaptive selection background analysis
- Perform background analysis and necessary robustness processing on the data. Grasp two issues through background analysis:
- The first is the background, reliability and availability of data;
- The second is to determine the major categories of prediction methods used in predictive analysis, such as regression models, time series models or econometric models, and input-output models. When abnormal values are found and need to be processed before they can be used, the data needs to be pre-processed robustly.
Adaptive Selection Model Method Library
- Establish a rich model method library and classify these model methods, such as time series models, including a variety of trend models, periodic models, and hybrid models.
- According to the characteristics of data composition, the corresponding modeling method is selected for predictive analysis. However, some predictions can intuitively determine the prediction method according to the prediction target, for example, the prediction of the equipment state, and simple weather prediction can be directly performed using the MARKOV prediction method. Many predictions like this have obvious range characteristics. At this time, a fixed prediction method applicable to such prediction targets can be directly selected.
- The regression prediction method also has obvious structural characteristics. At this time, historical data usually shows that one data (unary regression) or several data (multivariate regression) determines a prediction target, and the historical data is composed of several sequences at the same time. In this case, it is mainly to determine whether it is linear regression or nonlinear regression. The linear regression analysis method uses LS estimation, which is relatively simple and the method is relatively mature.
- For non-linear regression, we can use the neural network prediction method to analyze, use BP three-layer network algorithm to approximate, and then compare and analyze the variance of the prediction results of various methods. Based on the results of comprehensive analysis of various aspects, determine the Types of.
- In many continuous predictions, we should also study an adaptive prediction method, which can timely adjust the prediction method according to the prediction error. This will ensure the optimality of the prediction.
- If the results of the data background analysis show that the changes in historical data show a trend of a certain curve, we can use an approximate and smooth continuous function for analysis. These functions should be as simple as possible, which requires fewer function terms, lower powers, and fewer extremes and inflection points.
- Trend extrapolation function
- According to the specific background analysis, the commonly used trend extrapolation functions are as follows:
- (1) Linear type;
- (2) Quadratic type;
- (3) index type;
- (4) Hyperbolic type;
- (5) Growth curve type;
- (6) Period curve type, etc .:
- After the specific prediction objects of these models are clear, that is to say, after knowing the major categories of prediction methods, they should find as many available method models as possible. Create an optional model group.
Adaptive selection establishes the basis of discriminant theory
- In order to compare and select a relatively suitable model, criteria for discrimination and comparison must be established. Generally, people choose the variance of the fit (methods such as least square estimation and statistical test are based on the minimum variance) as a criterion for judgment. When selecting a forecasting method, because forecasting is a prediction of future economic events, errors will inevitably occur. The size of the error indicates the accuracy of the forecast and is an important criterion for evaluating the quality of the forecast. Therefore, when selecting the forecasting method, the error should be made to the best Possible small. The smaller the variance, the higher the prediction accuracy. This is a general principle.
- In addition to using the mean square error to measure the quality of prediction, in practice, we should also combine other discriminative methods, such as the information entropy H (r) of random errors and other empirical formulas. Use these theoretical methods to judge the availability of prediction methods.
- We can decompose the mean square error MSE (f) into three parts, which can further explain the source of the systematic error, but also help forecasting engineers and technicians get a more detailed basis, and identify the different sources of the systematic error.
- The mean square error can be divided into three terms:
- The first term is the squared deviation, which reflects the prediction error due to the true deviation;
- The second term is the prediction error caused by the square sum of the remainder of the slope as the weight, which reflects the error of the slope;
- The third term is the actual variance with weights of the remainder of the actual value and the prediction discrimination coefficient, which reflects the residual part of the error.
- Understanding the composition of errors can help analyze the sources of errors in the economic forecast analysis process and try to reduce them. In some cases, we not only use the mean square error to judge, but also measure the degree of dispersion of the error sequence. We see the error sequence as a random variable
- Assume
- We can also consider introducing other standards to reflect the predictability of different methods on an economic data, which is also a technical issue worthy of study. Because forecasting technicians may face columns of economic data from all walks of life, these data are both linear, non-linear, and even chaotic. Therefore, the issue of predictability must also be considered. For a poorly predictable information data column, even if people temporarily find a good prediction method and the prediction accuracy is very high, its long-term applicability may be problematic.
- We can consider defining a predictability parameter, which is set to P 0. The larger the P of a method, the higher the predictability. Therefore, we can reflect the predictability of the predictability of the model. At the same time, it also reflects the forecasting model's ability to control the economic forecasting system under study and the uncertainty of economic system development.
- Simple discrimination is not comprehensive and scientific. When predicting and analyzing specific economic problems, it is necessary to establish a comprehensive and scientific basis for discrimination. This is what we call the unified discrimination theory. Key issues in predictive research.
- Concerning the unified judgment theory standard is a subject that needs further research. Of course, currently in our method. The minimum variance method is used, that is, the variance of the fit is compared among a group of model groups, and the model with the smallest fit method is selected.