Interpretable neural network for non-linear control

A controller circuit implements an interpretable neural network based proportional integral derivative (PID) control function. The controller circuit includes a controller output signal for input to the non-linear device, a controller input signal representative of an error in an output of the non-linear device, and a neural network, the neural network is configured to generate, for a first signal dependent on a current value of the controller input signal, a second signal at least partially generated by the first neural network that estimates a difference of the controller input signal, and summing at least in part from a third signal generated by the second neural network that estimates an integration of a controller input signal over time, calculating a controller output signal from the controller input signal. 一种控制器电路实现可解释的基于神经网络的比例积分微分(PID)控制函数。该控制器电路包括用于输入到非线性设备的控制器输出信号、表示非线性设备的输出中的误差的控制器输入信号、以及神经网络，该神经网络被配置成：通过对取决于控制器输入信号的当前值的第一信号、至少部分地由估计控制器输入信号的差分的第一神经网络生成的第二信号、以及至少部分地由估计控制器输入信号的随时间的积分的第二神经网络生成的第三信号求