使用模糊逻辑分析模糊系统设计师- MATLAB & Simulink - MathWorks España

使用模糊逻辑分析模糊系统设计师

Once you define the membership functions and rule base of your fuzzy inference system (FIS), whether by manual construction or tuning, you can analyze your design using the following documents.

Control Surface— Plot the output values for all combinations of two input variables.
Rule Inference— Specify input values and view the inference diagram and resulting output value.
System Validation— Compare FIS output values to corresponding reference validation data.
Error Distribution— View the distribution of FIS output errors for different input combinations based on reference validation data.

To open these documents, on theDesigntab, select the corresponding icon in theSimulationgallery.

App toolstrip highlighting the Simulation gallery.

You can export your analysis results to the MATLAB^®workspace for further analysis. For more information, seeExport FIS and Simulation Data from Fuzzy Logic Designer．

When analyzing your FIS, you can configure the resolution of the output variable universe of discourse. To do so, set theNumber of Samplesparameter. This parameter is not supported for Sugeno systems.

Control Surface

TheControl Surfacedocument shows the output values from the active FIS design for all possible combinations of two input variables.

Control Surface document showing a nonlinear surface plot of tip values. At the top, you can select the input and output MFs for the X, Y, Z axes. Below that, you can select the number of mesh points to use along each of the X and Y axes. Below that, you can specify reference values for unused input variables.

In theAxessection, select the input and output variables using the corresponding drop-down menus.

In theMesh Pointssection, you can specify the plot resolution along theXandYaxes. For a smoother plot, specify larger resolution values.

If your system has more than two inputs, you must specify a reference value for each input variable not shown in the plot. To do so, specify theReference Inputsparameter. For example, the following figure shows a reference input for a three-input system for which the first and second variables are used in the control surface plot.

Control Surface document settings showing the Reference Inputs parameter for a three-input system. The first two inputs are selected for the X and Y axes of the plot. The reference values are [NaN NaN 0.5].

For each variable that is used in the control surface plot, specify a reference value ofNaN．

You can also view a control surface for a single input variable. In this case, in theAxessection, in theXdrop-down list, select the input variable and, in theYdrop-down list, selectnone．然后,指定所有unuse引用值d input variables using theReference Inputsparameter.

For example, the following figure shows a control surface for a two-input system where the output is plotted against the first input variable.

Control Surface Document showing a nonlinear plot of the output value against a single input variable

Rule Inference

In theRule Inferencedocument, you specify values for your input variables, inspect the inference process, and view the resulting output value produced by the active FIS design.

For more information on the fuzzy inference process for different FIS types, see:

For type-1 Mamdani systems, seeFuzzy Inference Process．
For type-1 Sugeno systems, seeMamdani and Sugeno Fuzzy Inference Systems．
For type-2 systems, seeType-2 Fuzzy Inference Systems．

To select input values, in theRule Inferencedocument, you can specify theInput valuesparameter. Alternatively, you can drag the input bars

Rule inference document showing the inference process for a type-1 system. On the left above the plots, the Input values text box is highlighted. Also, for the first input variable on the left, the figure shows arrows indicating that you can drag the input value bar left and right.

The preceding figure shows theRule Inferencedocument for a type-1 Mamdani system.

The fuzzified inputs activate the second and third rules. The firing strengths of these rules are applied to the output variable. When the resulting output membership functions are aggregated and defuzzified, the final output value is 21.7.

The following figure shows the same inference process for a type-2 Mamdani system. In this case, the fuzzified inputs generate firing strengths for both the upper and lower membership functions.

Rule inference document showing the inference process for a type-1 system.

System Validation

To validate your FIS designs, you can compare output values to corresponding reference validation data.

TheSystem Validationdocument shows simulation results for all designs that you select in theDesign Browser．To select a FIS design, select the corresponding entry in theComparecolumn.

To perform system validation for your FIS, you must first specify reference input and output data. For more information, seeSpecify Validation Data．

Using theSystem Validationdocument, you can view plots of:

Reference input values. Select the inputs to plot in theReference Inputstable.
FIS output values and the corresponding reference output values. Select the outputs to plot in theOutputstable.
Output prediction errors, which are the differences between the output and reference output values. To view the prediction errors, select thePrediction errorsparameter. You can view the root mean square error (RMSE) for each output in the legend of the prediction error plot.

To view corresponding numerical values from across all the plots, move your mouse over any of the plot lines.

Error Distribution

TheError Distributiondocument is a graphical representation of the output error for the active FIS design as compared to reference output data.

For a multi-output FIS, in theCalculatedrop-down list, select the output for which you want to analyze the error distribution.

To analyze the error distribution for your FIS, you must first specify reference input and output data. For more information, seeSpecify Validation Data．

Using theError Distributiondocument, you can:

Find input regions for which your FIS produces a less-accurate model of the data, which results in larger output errors.
Find input regions where you have fewer data points

For example, the following figure shows the error distribution for a tipping system with two inputs, service quality and food quality. The FIS performs well for most of the input space. However, there are significant errors in the upper-right region of the plot, which corresponds to situations where both the service and food have a high quality rating. In this case:

If you manually created the FIS, you should add or update rules for this input condition.
If you tuned the FIS from data, you might need more training data that covers this input region.

Specify Validation Data

TheSystem ValidationandError Distributionanalysis documents require reference validation data.

To select validation data, on theDesigntab:

In theInput Datadrop-down list, select the reference input data.
In theOutput Datadrop-down list, select the reference output data.

For both input and output data, you can select previously imported data, such as training data from tuning, or data from the MATLAB workspace.

TheInput DataandOutput Datadrop-down lists show only data that is compatible with the input and output configurations of your FIS.

The input validation data must be a numerical array where the number of columns is equal to the number of input variables in your FIS.
The output validation data must be a numerical array where the number of columns is equal to the number of output variables in your FIS.
The input data and output data arrays must have the same number of rows.