What Is an Adaptive Model?

Adaptive model refers to a model with anthropomorphic adaptive function. The adaptive model can modify the structure or parameters of the model under the condition that the characteristics of the simulated object or the operating environment of the model change, and keep the fit of the model and the simulated object normal. . ^[1]

The concept of "self-adaptation" comes from biological science, which mainly refers to the ability of biology, especially humans, to maintain the stability of the internal environment of the organism and the normal physiological state when the external environment and conditions change, which is one of the advantages of biological control systems.

In cybernetics and biological cybernetics, the adaptability of machines and biological control systems is studied. in

The adaptive model consists of 3 parts:

(1) Model ontology

The model ontology can be a mathematical model, a knowledge model, or a network model, all of which can be modified. For example, the parameters or expression structure of the mathematical model can be corrected; the knowledge base of the knowledge model can be added, deleted, and modified; the connection coefficient of the network model Or the topology can be adjusted.

(2) Adaptor

The adaptor accepts feedback from the degree of fit and information about changes in environmental conditions, formulates adaptation strategies and methods, and modifies the model ontology to maintain the best or satisfactory fit between the model and the simulated object, that is, the least fitting error or Within the scope of the permit, and meet the new requirements of environmental conditions.

3) Fitness evaluation

According to the purpose and application of the model, the accuracy and granularity of the model are required, and the accepted or user-accepted evaluation criteria are used to evaluate the output of the model and the simulated object.

In order to make the model adaptive, the following three methods need to be studied:

(1) Method of obtaining simulated object and environmental information.

When the simulated object and environment are passive systems and machines and equipment, the identification and observation methods are used to obtain information. Automatic or manual observation of the output or state of the simulated object and the influencing factors of the environmental conditions can be used online and in real time. Observation and identification can also use offline, non-real-time observation and identification. After filtering, shaping and transforming, useful information can be obtained.

When the simulated object and environment is an active system, such as an expert or an expert group, human-computer interaction can be adopted. When the simulated object and environment is an active system, which is a person or a crowd, such as an expert or expert group, then Information can be obtained using human-machine interactive or expert survey methods, such as the Delphi method.

2) Evaluation method of the fit between the model and the simulated object

When the model body is a mathematical model or a network model, and its output is quantitative data, the fitting error can be used as the basis for the evaluation of the degree of fit, and the relevant methods in Systems Identification are used for evaluation, such as the least square method.

When the model ontology is a knowledge model or a network model, its output is a qualitative conclusion. For example, conclusions and prescribing opinions on disease diagnosis. At this time, the "fuzzy evaluation" method or the "expert rating" method can be used to evaluate the degree of fit.

3) Modification or correction method of the model

When the model body is a mathematical model, such as a transfer function, a state equation (differential equation or difference equation), an algebraic equation, etc., a method similar to a self-tuning line adjuster or controller can be used to correct the parameters or structure of the model. Such as changing the transfer coefficient, time constant or integration, differentiation, and time delay in the transfer function, changing the numerical values of the A, B, C, and D matrix elements in the state equation or the distribution characteristics of the order of the matrix and non-zero elements.

When the model ontology is a knowledge model, such as a knowledge base of a production system, a knowledge base management system can be used to add, delete, and modify the rule base and the fact base through knowledge assimilation and knowledge adaptation technologies. If necessary, the control strategy of the cattle-producing system can also be modified.

When the model body is a neural network model, such as a multilayer perceptron model, the weight coefficients of synaptic connections, the action threshold of the neuron model, or the topological structure of synaptic connections, and the number of perceptron layers can be changed. ^[1]