[go: up one dir, main page]

WO2000063789A2 - Procede permettant de prevoir l'evolution d'une serie de donnees et dispositif de mise en oeuvre dudit procede - Google Patents

Procede permettant de prevoir l'evolution d'une serie de donnees et dispositif de mise en oeuvre dudit procede Download PDF

Info

Publication number
WO2000063789A2
WO2000063789A2 PCT/EP2000/002784 EP0002784W WO0063789A2 WO 2000063789 A2 WO2000063789 A2 WO 2000063789A2 EP 0002784 W EP0002784 W EP 0002784W WO 0063789 A2 WO0063789 A2 WO 0063789A2
Authority
WO
WIPO (PCT)
Prior art keywords
influencing variables
course
neural network
forecast
data
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Ceased
Application number
PCT/EP2000/002784
Other languages
German (de)
English (en)
Other versions
WO2000063789A3 (fr
Inventor
Martin Kuehrer
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Siemens AG Oesterreich
Original Assignee
Siemens AG Oesterreich
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Siemens AG Oesterreich filed Critical Siemens AG Oesterreich
Priority to AU42916/00A priority Critical patent/AU4291600A/en
Publication of WO2000063789A2 publication Critical patent/WO2000063789A2/fr
Publication of WO2000063789A3 publication Critical patent/WO2000063789A3/fr
Anticipated expiration legal-status Critical
Ceased legal-status Critical Current

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q40/00—Finance; Insurance; Tax strategies; Processing of corporate or income taxes
    • G06Q40/08—Insurance
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/049—Temporal neural networks, e.g. delay elements, oscillating neurons or pulsed inputs
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/086—Learning methods using evolutionary algorithms, e.g. genetic algorithms or genetic programming

Definitions

  • the invention relates to a method for predicting the course of a data series, this forecast being determined from the course of suspected influencing variables by means of a neural network.
  • a fundamental problem when using neural networks is that they can only take into account a limited number of influencing variables.
  • the selection of the influencing variables therefore essentially determines the accuracy of an automatic forecasting system.
  • the invention is therefore based on the object of specifying a method of the type mentioned at the outset which leads to forecast values which are as accurate as possible.
  • this is done with a method for forecasting the course of a data series, this forecast being determined by means of a neural network from the course of suspected influencing variables with the following method steps: - Using a data filter, the set of suspected influencing variables recorded in a database is converted into a sentence with a limited number selected by influencing factors;
  • the neural network is trained with the selected influencing variables and part of the known course of the data series; - The forecast accuracy of the trained network is checked on the remaining part of the known course of the data series and assigned a value number;
  • a set of influencing variables is determined in an iterative process for which the value number reaches a minimum
  • the set of influencing variables with the minimum number of values is used to forecast the further course of the data series.
  • the method according to the invention also leads to very good prognosis results for data series with a large number of influencing variables.
  • the average forecast error of the neural network is determined both for the training data and for the check data and the larger value of the two is defined as the number of values.
  • the assessment of the "fitness”, i.e. the quality of a solution, is essential for the accuracy of a genetic algorithm.
  • the assessment of the fitness according to the invention requires comparatively little calculation effort and leads to good results.
  • the neural network is trained and checked for several different sets of influencing variables simultaneously.
  • the processes mentioned take up a lot of computing time. They are therefore executed in parallel, i.e. A neural network is set up on several computers, the selected sentences are distributed to these neural networks for testing, and the results are combined for further evaluation.
  • a device for carrying out the method, in which a plurality of computers are provided which work together using both the TCP and the DCOM protocol in such a way that one of the computers acts as a TCP server and DCOM Client and the other computers work as DCOM servers and TCP clients in accordance with the relevant standards.
  • the course of a data series is to be predicted for the future.
  • the history of the data series in the past and the corresponding values of presumed influencing variables are known. This data is in one
  • N is in the range
  • T in the range of a few 100.
  • Perceptron principle should have this N input neurons and 1 output neuron.
  • the number h e N. the "hidden" neurons of the intermediate layers can be freely selected.
  • G (K + 1) • h.
  • K is the number of input neurons, in this case equal to N.
  • the neural network is trained over the time axis, the maximum number of available training examples T and in the exemplary embodiment is an order of magnitude smaller than G. This network is overdetermined and therefore not realizable.
  • the aim now is to make the number of free parameters G smaller than that To keep the number of training examples 7, if possible by an entire order of magnitude: (K + ⁇ ) - h ⁇ T.
  • a set of K influencing variables x n (t) ... x fl ⁇ _ (t), l ⁇ n ] ... n ⁇ ⁇ N is to be selected on the basis of which the neural network determines an optimal prognosis.
  • this selection is made in an iterative process using a genetic algorithm. It is a search algorithm that mimics the biological processes of mutation, recombination and selection. The calculation of these processes is relatively complex, but leads to very good results.
  • the genetic algorithm and the neural network work together in such a way that a limited number of influencing variables are selected by means of the genetic algorithm and the neural network is trained and tested with these influencing variables and the corresponding values of the data series to be predicted.
  • the results are called the genetic fitness Algorithm fed back and used to determine further sets of influencing variables.
  • the set of influencing variables with the best fitness is then used to forecast the future course of the data series.
  • a filter is defined, the transfer function of which describes the selection process.
  • the filter comprises the database and a vector with K elements (n, ..., n ⁇ ) from which the input vector for the neural network is calculated.
  • the "transfer function" of this filter is therefore
  • the accuracy of the prognosis of the neural network for this input data is determined and defined as fitness.
  • the assessment of the "fitness”, i.e. the quality of a solution, is essential for the accuracy of a genetic algorithm.
  • the assessment of the fitness according to the invention requires comparatively little calculation effort and leads to good results.
  • new sentences are determined by recombination: two sentences are combined by exchanging the parameters, e.g.: from (n ,,, ..., n x ⁇ ) and # ⁇ - p ..., / ⁇ 2je becomes ( n 2l , n 22 , n u , n u , ..., n ⁇ _, n 2 ⁇ ).
  • the newly created sets of influencing variables are changed in a random manner by mutation. Then the fitness values of the neural network are again calculated for these sets.
  • the determination of a fitness value g requires relatively high computing power or computing time, mainly because the training process of the neural network requires a lot of effort. Therefore, according to the invention, several sets of influencing variables are processed in parallel on different computers.
  • the data records to be processed are queued and from there distributed to the individual computers for calculations.
  • This queue is executed on the computer on which the genetic algorithm is also calculated.
  • This computer is set up as a TCP server, in accordance with the “Transmission Control Protocol / Internet Protocol (TCP / IP). "A de facto
  • TCP / IP Transmission Control Protocol / Internet Protocol
  • IP is the agreement on how the individual data packets are formulated and sent.
  • the TCP then takes over the connection establishment and the secure delivery of the , g. / Data packet.
  • the name of TCP / IP is based on the two most used protocols, but they are basically dozens of different protocols. (Log collection). Many of these work invisibly to the user, but address the enormous problems of network access and the cooperation of various networks and routing methods that occur in such a diverse and heterogeneous network as the Internet.
  • the TCP / IP protocols can be roughly divided into four operating levels: network access, network, transport, application.
  • the other computers are set up as TCP clients.
  • conventional TCP servers cannot actively establish a connection, but can only react passively to requests. Therefore, the TCP server is executed using a different protocol (DCOM) as a client, whereby a TCP server / DCOM client TCP client / DCOM server architecture of the computer system.
  • DCOM protocol
  • a connection is set up as follows:
  • - DCOM client (computer 1) starts DCOM server on another computer (computer 2)
  • - DCOM server (computer 2) starts TCP client (computer 2) and

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Business, Economics & Management (AREA)
  • General Physics & Mathematics (AREA)
  • Health & Medical Sciences (AREA)
  • Computing Systems (AREA)
  • Software Systems (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Finance (AREA)
  • General Engineering & Computer Science (AREA)
  • Accounting & Taxation (AREA)
  • Mathematical Physics (AREA)
  • Artificial Intelligence (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Evolutionary Biology (AREA)
  • Physiology (AREA)
  • Development Economics (AREA)
  • Economics (AREA)
  • Marketing (AREA)
  • Strategic Management (AREA)
  • Technology Law (AREA)
  • General Business, Economics & Management (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

Procédé permettant de prévoir l'évolution d'une série de données, cette prévision étant déterminée à l'aide d'un réseau neuronal à partir de l'évolution de grandeurs influentes supposées. Un ensemble de grandeurs influentes est sélectionné parmi ces grandeurs influentes supposées à l'aide d'un algorithme génétique. Ledit ensemble constitue les valeurs d'entrée du réseau neuronal pour la prévision de l'évolution ultérieure de la série de données.
PCT/EP2000/002784 1999-04-16 2000-03-30 Procede permettant de prevoir l'evolution d'une serie de donnees et dispositif de mise en oeuvre dudit procede Ceased WO2000063789A2 (fr)

Priority Applications (1)

Application Number Priority Date Filing Date Title
AU42916/00A AU4291600A (en) 1999-04-16 2000-03-30 Method for prognosing the course of a data sequence and device for carrying out said method

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
AT68499 1999-04-16
ATA684/99 1999-04-16

Publications (2)

Publication Number Publication Date
WO2000063789A2 true WO2000063789A2 (fr) 2000-10-26
WO2000063789A3 WO2000063789A3 (fr) 2001-03-22

Family

ID=3497097

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/EP2000/002784 Ceased WO2000063789A2 (fr) 1999-04-16 2000-03-30 Procede permettant de prevoir l'evolution d'une serie de donnees et dispositif de mise en oeuvre dudit procede

Country Status (2)

Country Link
AU (1) AU4291600A (fr)
WO (1) WO2000063789A2 (fr)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US10242407B1 (en) 2013-09-24 2019-03-26 Innovative Market Analysis, LLC Financial instrument analysis and forecast

Family Cites Families (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
EP0528399A3 (en) * 1991-08-19 1994-12-21 Toyoda Machine Works Ltd Method and apparatus for learning of neural network
JPH10513290A (ja) * 1995-01-31 1998-12-15 松下電器産業株式会社 比率予測システムと混合物生成方法
US5727128A (en) * 1996-05-08 1998-03-10 Fisher-Rosemount Systems, Inc. System and method for automatically determining a set of variables for use in creating a process model

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US10242407B1 (en) 2013-09-24 2019-03-26 Innovative Market Analysis, LLC Financial instrument analysis and forecast

Also Published As

Publication number Publication date
WO2000063789A3 (fr) 2001-03-22
AU4291600A (en) 2000-11-02

Similar Documents

Publication Publication Date Title
DE60301202T2 (de) Verfahren und vorrichtung zur verkehrssteuerung einer web-farm
DE3236832A1 (de) Verfahren und geraet zur sprachanalyse
DE2949582A1 (de) Verfahren und anordnung zur bestimmung der tonhoehe in menschlicher sprache
DE4328752B4 (de) Spracherkennungssystem
DE60124122T2 (de) Kommunikationsnetzentwurf
EP1361737A1 (fr) Méthode et système de traitement du signal de parole et de classification de dialogues
DE1915819A1 (de) Verfahren und Anordnung zur adaptiven Zeichenerkennung
EP1470459B1 (fr) Pronostic precoce de la fiabilite d'un ensemble technique
DE19504664C2 (de) Chaotisches rekurrentes neuronales Netz und Lernverfahren dafür
WO2000063789A2 (fr) Procede permettant de prevoir l'evolution d'une serie de donnees et dispositif de mise en oeuvre dudit procede
WO2025114004A1 (fr) Système et procédé utilisant un grand modèle de langage
EP0643353B1 (fr) Dispositif et méthode de traitement de données
EP1264253A1 (fr) Procede et dispositif pour la modelisation d'un systeme
EP3211499A2 (fr) Procede, produit programme informatique et machine a plier pour le pliage de toles
WO2015078436A1 (fr) Dispositif plate-forme pour la fourniture de savoir collectif quantitatif
WO2005036424A9 (fr) Outil de gestion
DE19811352C2 (de) System und Verfahren zur Suche auf untereinander vernetzten Rechnern mit Informationsbeständen mittels Softwareagenten
EP1363271A1 (fr) Méthode et système pour le traitement et la mémorisation du signal de parole d'un dialogue
WO2021249941A1 (fr) Procédé et dispositif d'analyse automatique d'un comportement d'un dispositif technique
EP0591259B1 (fr) Procede de traitement d'incertitudes dans des donnees d'entree dans des reseaux neuroniques
DE68911719T2 (de) Verfahren und Vorrichtung, die schlechte Leistung eines Rechnersystemes diagnostisiert.
EP4200737B1 (fr) Procédé de détection de manipulation de données relative à des valeurs de données numériques
DE202022101927U1 (de) Ein autorenstammunabhängiges Punktzahlberechnungssystem zur Messung der wissenschaftlichen Unabhängigkeit
EP4506867A1 (fr) Procédé et appareil de traitement assisté par ordinateur d'un circuit quantique sur une unité de traitement quantique
DE102023208154A1 (de) Computerimplementiertes Verfahren zum Trainieren eines neuronalen Netzwerks mittels MTL

Legal Events

Date Code Title Description
AK Designated states

Kind code of ref document: A2

Designated state(s): AU CA JP MX NO RU US

AL Designated countries for regional patents

Kind code of ref document: A2

Designated state(s): AT BE CH CY DE DK ES FI FR GB GR IE IT LU MC NL PT SE

121 Ep: the epo has been informed by wipo that ep was designated in this application
DFPE Request for preliminary examination filed prior to expiration of 19th month from priority date (pct application filed before 20040101)
AK Designated states

Kind code of ref document: A3

Designated state(s): AU CA JP MX NO RU US

AL Designated countries for regional patents

Kind code of ref document: A3

Designated state(s): AT BE CH CY DE DK ES FI FR GB GR IE IT LU MC NL PT SE

122 Ep: pct application non-entry in european phase
NENP Non-entry into the national phase

Ref country code: JP