Projecting interval uncertainty through the discrete Fourier transform

Overview

Projecting interval uncertainty through the discrete Fourier transform

This repository provides a method that can propagate interval uncertainty through the discrete Fourier transform while yielding the exact bounds on the Fourier amplitude and Power Spectral Density function. The algorithm applies to real sequences of intervals. The method allows technical analysts to project interval uncertainty present in the time signals to their Fourier amplitude without making assumptions about the error distribution at each time step. Thus, it is possible to calculate and analyse system responses in the frequency domain without conducting extensive Monte Carlo simulations in the time domain. The applicability of this method in practice is demonstrated by a technical application.

Disclaimer: This code was developed for illustration purposes and for proof-of-concept. Thus this code is not optimized for large-scale applications. An optimized version of the code is currently under development.

References

De Angelis, M.; Behrendt, M.; Comerford, L.; Zhang, Y.; Beer, M. (2021): Forward interval propagation through the discrete Fourier transform, The 9th international workshop on Reliable Engineering Computing, arXiv:2012.09778.

Installation

Clone the git repository on your machine, cd to the repository, open a Python3 interpreter and import the interval Fourier transform ans other useful packages

from fourier.transform import transform as intervalDFT
from fourier.application import application as app
from fourier.number import number as int_num
import numpy
from numpy import (arange, cos, exp, linspace, mean, pi,  sin, zeros) 
from matplotlib import pyplot, cm

Signal generation and interval DFT

At first time and frequency parameters and an analytical PSD function are needed to model a stochastic process.

Define parameters

wu = 2.2975 # upper cut-off frequency
T = 350 # total time length

dt = 2*pi /(2*wu) # timestep size
dw = 2*pi / T # frequency step size

t = numpy.arange(0,T,dt) # time vector
w = numpy.arange(0,wu,dw) # frequency vector

JONSWAP power spectrum

The JONSWAP power spectrum is utilised to generate stochastic processes. The required parameters are:

alpha = 0.0081 # spectral energy parameter
w_p = 0.7 # peak frequency
gamma = 3.3 # peak enhancement factor
sigma1 = 0.07 # spectral width parameter for w <= w_p
sigma2 = 0.09 # spectral width parameter for w > w_p
spectrum = app.jonswap_spectrum(w,alpha,w_p,gamma,sigma1,sigma2)

Plot the JONSWAP power spectrum

ax = app.plot_line(w,spectrum,figsize=(18,6),xlabel=r'#$x$',ylabel='$x$',color=None,lw=1,title='JONSWAP power spectrum',ax=None,label=None)
ax.set_xlabel('Frequency [rad/s]',fontsize=20)
ax.set_ylabel('Power Spectral Density [m$^2$s]',fontsize=20)

fig

Generate time signal and intervalize it

To generate a stochastic process the spectral representation method is utilised. This signal is then intervalized with interval uncertainty ±0.1. Both signals are plotted.

sea_waves = app.stochastic_process(spectrum,w,t) 
pm = 0.1
sea_waves_interval = intervalDFT.intervalize(sea_waves, pm)

ax = app.plot_line(t,sea_waves,figsize=(18,6),xlabel='Time [s]',ylabel='Wave height [m]',color='rebeccapurple',lw=1,title='Signal from stationary power spectrum',ax=None,label=None)
sea_waves_interval.plot(xlabel='Time [s]',ylabel='Wave height [m]',title=r'Signal with $\pm$ '+str(pm)+' information gaps (intervals)')

fig fig

Compute the Fourier transforms

Compute the Fourier transform of the crisp signal and the interval Fourier transform for the interval signal with the selective method and the interval method. Also compute the periodogram of respective (bounded) Fourier amplitudes.

FA = intervalDFT.Fourier_amplitude(sea_waves)
BI,BS = intervalDFT.compute_amplitude_bounds(sea_waves_interval)
BI.insert(0,int_num.Interval(0,0))
BS.insert(0,int_num.Interval(0,0))

FA = app.periodogram(FA, t, dt)
BI = app.periodogram(BI, t, dt)
BS = app.periodogram(BS, t, dt)

Plot the interval Fourier transform

The amplitude of the crisp signal and both bounded Fourier amplituted are plotted.

ax = app.plot_line(w,FA,figsize=(18,6),xlabel=r'#$x$',ylabel=r'$x$',color=None,lw=1,title=None,ax=None,label='Interval uncertainty: $\pm$ '+str(pm)+'')
app.plot_bounds(x=w,bounds=BI,color='cornflowerblue',alpha=0.4,ax=ax)
app.plot_bounds(x=w,bounds=BS,color='orangered',alpha=0.6,ax=ax)
ax.set_xlabel('Frequency [rad/s]',fontsize=20)
ax.set_ylabel('Power Spectral Density [m$^2$s]',fontsize=20)
ax.tick_params(direction='out', length=6, width=2, labelsize=14)

fig

Application to a SDOF system

The system under investigation is a offshore wind turbine simplified to a SDOF system. The parameters are set to

R = 3 # outer radius
r = 2.8 # inner radius
h_pile = 60 # height
rho_steel = 7800 # density of steel
c = 1e5 # stiffness
k = 1e6 # damping coefficient

Get the natural frequency w0 and the damping ratio xi

w0,xi = app.wind_turbine(R,r,h_pile,rho_steel,c,k)

The response can be obtained by pushing the (intervalised) signal through the frequency response function

freq_response_precise = app.frequency_response(w,FA,w0,xi)
freq_response_BI_low,freq_response_BI_high = app.frequency_response_interval(w,BI,w0,xi)
freq_response_BS_low,freq_response_BS_high = app.frequency_response_interval(w,BS,w0,xi)

Those responses can be plotted

ax = app.plot_line(w,freq_response_precise,figsize=(18,6),xlabel=r'#$x$',ylabel=r'$x$',color=None,lw=1,title=None,ax=None,label=None)
ax.fill_between(x=w,y1=freq_response_BI_low,y2=freq_response_BI_high, alpha=0.4, label='Interval', edgecolor='blue', lw=2, color='cornflowerblue')
ax.fill_between(x=w,y1=freq_response_BS_low,y2=freq_response_BS_high, alpha=0.6, label='Selective', edgecolor='red', lw=2, color='orangered')

ax.set_xlabel('Frequency [rad/s]',fontsize=20)
ax.set_ylabel('Power Spectral Density [m$^2$s]',fontsize=20)
ax.set_title(r'Interval uncertainty: $\pm$ '+str(pm)+'', fontsize=20)

ax.tick_params(direction='out', length=6, width=2, labelsize=14)
_=ax.set_xlim([0.5, 1.1])

fig

Comparison with Monte Carlo

In this section it is illustrated how severe interval uncertainty is underestimated by Monte Carlo. To show this, a signal with interval uncertainty ±0.5 is utilised and plotted.

pm = 0.5
sea_waves_interval_05 = intervalDFT.intervalize(sea_waves, pm)
sea_waves_interval_05.plot(xlabel='Time [s]',ylabel='Wave height [m]',title=r'Signal with $\pm$ '+str(pm)+' information gaps (intervals)')

fig

Generate some random signals between the bounds. All signals which are within or on the bounds are possible.

RAND_SIGNALS = sea_waves_interval_05.rand(N=20) # this picks out N (inner) random signals within the bounds

fig,ax = intervalDFT.subplots(figsize=(16,8))
for rs in RAND_SIGNALS:
    intervalDFT.plot_signal(rs,ax=ax)
sea_waves_interval_05.plot(ax=ax)
ax.grid()
_=ax.set_xlim(0,55) # underscore here is used to suppress the output of this line

fig

Computing the Fourier amplitude bounds and the periodogram of the interval signal

BI,BS = intervalDFT.compute_amplitude_bounds(sea_waves_interval_05)
BI.insert(0,int_num.Interval(0,0))
BS.insert(0,int_num.Interval(0,0))

BI = app.periodogram(BI, t, dt)
BS = app.periodogram(BS, t, dt) 

Plotting the bounds of the Fourier amplitude in comparison to the resulting bounds obtained by Monte Carlo

BI_low=[ai.lo() for ai in BI]
BI_high=[ai.hi() for ai in BI]
BS_low=[ai.lo() for ai in BS]
BS_high=[ai.hi() for ai in BS]

fig = pyplot.figure(figsize=(18,6))
ax = fig.subplots()
ax.grid()
ax.fill_between(x=w,y1=BI_low,y2=BI_high, alpha=0.4, label='Interval', edgecolor='blue', lw=2, color='cornflowerblue')
ax.fill_between(x=w,y1=BS_low,y2=BS_high, alpha=0.6, label='Selective', edgecolor='red', lw=2, color='orangered')

n_MC = 10
for x in range(n_MC):
    FX = intervalDFT.Fourier_amplitude(sea_waves_interval_05.rand())
    FX = app.periodogram(FX, t, dt)
    #intervalDFT.plot_line(w,FX,figsize=None,xlabel=r'#$x$',ylabel=r'$x$',color='palegreen',lw=1,title=None,ax=ax,label=None) 
    app.plot_line(w,FX,figsize=None,xlabel=r'#$x$',ylabel=r'$x$',color='#d7f4d7',lw=1,title=None,ax=ax,label=None) 

ax.set_xlabel('Frequency [rad/s]',fontsize=20)
ax.set_ylabel('Power Spectral Density [m$^2$s]',fontsize=20)
ax.set_title(r'Interval uncertainty: $\pm$ '+str(pm)+'', fontsize=20)

ax.tick_params(direction='out', length=6, width=2, labelsize=14)  

fig

Which increasing sample size, the range within the bounds of the interval signal is better covered. However, even a very high sample size is insufficient to get close to the bounds obtained by the interval DFT.

fig = pyplot.figure(figsize=(18,6))
ax = fig.subplots()
ax.grid()

n_MC = 1000
for x in range(n_MC):
    FX = intervalDFT.Fourier_amplitude(sea_waves_interval_05.rand())
    FX = app.periodogram(FX, t, dt)
    app.plot_line(w,FX,figsize=None,xlabel=r'#$x$',ylabel=r'$x$',color='#7cc47c',lw=1,title=None,ax=ax,label=None) 
    
n_MC = 100
for x in range(n_MC):
    FX = intervalDFT.Fourier_amplitude(sea_waves_interval_05.rand())
    FX = app.periodogram(FX, t, dt)
    app.plot_line(w,FX,figsize=None,xlabel=r'#$x$',ylabel=r'$x$',color='#a7d9a7',lw=1,title=None,ax=ax,label=None) 
    
n_MC = 10
for x in range(n_MC):
    FX = intervalDFT.Fourier_amplitude(sea_waves_interval_05.rand())
    FX = app.periodogram(FX, t, dt)
    app.plot_line(w,FX,figsize=None,xlabel=r'#$x$',ylabel=r'$x$',color='#d7f4d7',lw=1,title=None,ax=ax,label=None) 
    
ax.set_xlabel('Frequency [rad/s]',fontsize=20)
ax.set_ylabel('Power Spectral Density [m$^2$s]',fontsize=20)
_=ax.set_title('Bounds estimated by MC', fontsize=20) 

fig

Jupyter notebooks showing best practices for using cx_Oracle, the Python DB API for Oracle Database

Python cx_Oracle Notebooks, 2022 The repository contains Jupyter notebooks showing best practices for using cx_Oracle, the Python DB API for Oracle Da

Christopher Jones 13 Dec 15, 2022
Music Source Separation; Train & Eval & Inference piplines and pretrained models we used for 2021 ISMIR MDX Challenge.

Introduction 1. Usage (For MSS) 1.1 Prepare running environment 1.2 Use pretrained model 1.3 Train new MSS models from scratch 1.3.1 How to train 1.3.

Leo 100 Dec 25, 2022
Minimal implementation and experiments of "No-Transaction Band Network: A Neural Network Architecture for Efficient Deep Hedging".

No-Transaction Band Network: A Neural Network Architecture for Efficient Deep Hedging Minimal implementation and experiments of "No-Transaction Band N

19 Jan 03, 2023
Notification Triggers for Python

Notipyer Notification triggers for Python Send async email notifications via Python. Get updates/crashlogs from your scripts with ease. Installation p

Chirag Jain 17 May 16, 2022
ICON: Implicit Clothed humans Obtained from Normals (CVPR 2022)

ICON: Implicit Clothed humans Obtained from Normals Yuliang Xiu · Jinlong Yang · Dimitrios Tzionas · Michael J. Black CVPR 2022 News 🚩 [2022/04/26] H

Yuliang Xiu 1.1k Jan 04, 2023
Fashion Recommender System With Python

Fashion-Recommender-System Thr growing e-commerce industry presents us with a la

Omkar Gawade 2 Feb 02, 2022
DeepLabv3+:Encoder-Decoder with Atrous Separable Convolution语义分割模型在tensorflow2当中的实现

DeepLabv3+:Encoder-Decoder with Atrous Separable Convolution语义分割模型在tensorflow2当中的实现 目录 性能情况 Performance 所需环境 Environment 注意事项 Attention 文件下载 Download

Bubbliiiing 31 Nov 25, 2022
Code reproduce for paper "Vehicle Re-identification with Viewpoint-aware Metric Learning"

VANET Code reproduce for paper "Vehicle Re-identification with Viewpoint-aware Metric Learning" Introduction This is the implementation of article VAN

EMDATA-AILAB 23 Dec 26, 2022
Code for the TASLP paper "PSLA: Improving Audio Tagging With Pretraining, Sampling, Labeling, and Aggregation".

PSLA: Improving Audio Tagging with Pretraining, Sampling, Labeling, and Aggregation Introduction Getting Started FSD50K Recipe AudioSet Recipe Label E

Yuan Gong 84 Dec 27, 2022
Repository of our paper 'Refer-it-in-RGBD' in CVPR 2021

Refer-it-in-RGBD This is the repository of our paper 'Refer-it-in-RGBD: A Bottom-up Approach for 3D Visual Grounding in RGBD Images' in CVPR 2021 Pape

Haolin Liu 34 Nov 07, 2022
Lyapunov-guided Deep Reinforcement Learning for Stable Online Computation Offloading in Mobile-Edge Computing Networks

PyTorch code to reproduce LyDROO algorithm [1], which is an online computation offloading algorithm to maximize the network data processing capability subject to the long-term data queue stability an

Liang HUANG 87 Dec 28, 2022
Relative Uncertainty Learning for Facial Expression Recognition

Relative Uncertainty Learning for Facial Expression Recognition The official implementation of the following paper at NeurIPS2021: Title: Relative Unc

35 Dec 28, 2022
nanodet_plus,yolov5_v6.0

OAK_Detection OAK设备上适配nanodet_plus,yolov5_v6.0 Environment pytorch = 1.7.0

炼丹去了 1 Feb 18, 2022
An intelligent, flexible grammar of machine learning.

An english representation of machine learning. Modify what you want, let us handle the rest. Overview Nylon is a python library that lets you customiz

Palash Shah 79 Dec 02, 2022
A code repository associated with the paper A Benchmark for Rough Sketch Cleanup by Chuan Yan, David Vanderhaeghe, and Yotam Gingold from SIGGRAPH Asia 2020.

A Benchmark for Rough Sketch Cleanup This is the code repository associated with the paper A Benchmark for Rough Sketch Cleanup by Chuan Yan, David Va

33 Dec 18, 2022
Using NumPy to solve the equations of fluid mechanics together with Finite Differences, explicit time stepping and Chorin's Projection methods

Computational Fluid Dynamics in Python Using NumPy to solve the equations of fluid mechanics 🌊 🌊 🌊 together with Finite Differences, explicit time

Felix Köhler 4 Nov 12, 2022
DeepFaceEditing: Deep Face Generation and Editing with Disentangled Geometry and Appearance Control

DeepFaceEditing: Deep Face Generation and Editing with Disentangled Geometry and Appearance Control One version of our system is implemented using the

260 Nov 28, 2022
Code for Graph-to-Tree Learning for Solving Math Word Problems (ACL 2020)

Graph-to-Tree Learning for Solving Math Word Problems PyTorch implementation of Graph based Math Word Problem solver described in our ACL 2020 paper G

Jipeng Zhang 66 Nov 23, 2022
Cosine Annealing With Warmup

CosineAnnealingWithWarmup Formulation The learning rate is annealed using a cosine schedule over the course of learning of n_total total steps with an

zhuyun 4 Apr 18, 2022
This is the code for the paper "Jinkai Zheng, Xinchen Liu, Wu Liu, Lingxiao He, Chenggang Yan, Tao Mei: Gait Recognition in the Wild with Dense 3D Representations and A Benchmark. (CVPR 2022)"

Gait3D-Benchmark This is the code for the paper "Jinkai Zheng, Xinchen Liu, Wu Liu, Lingxiao He, Chenggang Yan, Tao Mei: Gait Recognition in the Wild

82 Jan 04, 2023