Improving customer demand and cost forecasting methods |
About the project
This project will address the complex issue of hierarchical forecasting of the inventory and capacity requirements at the individual nodes of supply chains, see SECTION 1, ANNEX 3. This issue is an instance of an applied scientific problem, which is universally applicable to both manufacturing and service supply chains. It is of immense value to UK businesses since it determines the levels of non-added value working capital needed to finance operations and hence their ability to sustain and grow their businesses. The AI forecasting tools to be developed will be fully ALIGNED with the Data and Content Storage, Management, Retrieval and Analysis Technology Priorities since they will need to provide highly efficient content mining techniques capable of the search, retrieval, aggregation and interpretation of high volumes of data from large numbers of disparate multimedia data sources and for the discovery of the inventory and capacity demand relationships from within the collected data. PROJECT CHANGES have been made to the proposal since the Outline stage which involve
- increasing the number of end user partners in response to DTI feedback;
- allocating the Project Leadership to Dr Karl Seare of STSL who has significant project management experience and expertise in both inventory and capacity forecasting.
The new partners have strengthened the project, ie Unipart Logistics Ltd (ULL) provide supply chain products and services and have significant expertise in both inventory and capacity forecasting and Preactor International (PI), who have replaced Greycon, as software vendors of capacity scheduling tools.
From an industrial perspective the focus on competing through supply chains, the increasing levels of off-shore manufacturing, and the rapid adoption of new materials, processes and product technology are all tending to rapidly increase the levels of both complexity and variability inherent in the forecasting process. Greater demands will be placed on the forecasting function to provide significantly improved forecasting techniques able to provide more accurate forecasts in more complex environments, which will help reduce non-added value inventory and capacity and thereby decrease the levels of operating capital required. Industrial and service organisations are well aware of the existing and future needs for greatly improved forecasting capabilities. The proposed work is particularly TIMELY since the AI forecasting technologies that will satisfy these needs will be developed through this project and will lead to marketable products and services that will help to meet the objective identified by the Technology Strategy Board of helping “leading industrial sectors to remain prosperous in the face of global competition”. The project will require radically different INNOVATIVE approaches consisting of the development and exploitation of totally new forecasting concepts that will push the boundries over and beyond current leading edge world science. Innovations will occur both in the core concepts associated with individual AI components, (eg Type II Fuzzy Logic), the way these components are combined into the overall system and the way in which both system and components are self-optimised. This project will catalyse a new endeavour to revitalise a potentially stagnant research area, (see SECTION 2, ANNEX 3), that remains of critical importance to industry. The work proposed is NOVEL in that the degree of differentiation from existing forecasting techniques will be considerable and will involve self-optimisation and autonomous decision making.
Four main AI paradigms will be employed, ie fuzzy logic, artificial neural networks, genetic algorithms and ontologies combined with semantic networks. These basic AI components will be selected and combined to provide the functionality required and in doing so where appropriate, a key role will be played by the process of ‘hybridisation’. The AI components will provide the primary technologies used to develop three knowledge systems, ie (1) Knowledge Information Systems that will support knowledge collection by identifying potential data types and data sources and providing navigation paths during data collection; (2) Knowledge Collection Systems that will undertake the actual search, navigation and aggregation processes required to find appropriate data sources and extract from these sources the individual items of data that may represent predictor variables and/or dependent variables within the resulting forecasting models; (3) Knowledge Analysis Systems that will analyse the collected data in order to discover appropriate relationships from which forecasting models can be developed. A genetic algorithm optimiser, using error value feedback from models, will be used to improve the structure and content of each knowledge system. See ANNEX 1 for further details.
Without government funding this project would not go ahead in any shape or form since it would not be able to overcome the significant combination of market barriers, technological barriers, prevailing practice barriers and investment barriers that project initiation requires. MARKET BARRIERS, in both industrial and service sectors, exist due to a lack of market familiarity with the use of AI based methods, ie there is no known industrial planning and control application existing. TECHNOLOGICAL BARRIERS exist due to a lack of expertise in applying AI within the forecasting application domain, the wide variety of AI components that will need to be employed and the complexity involved in combining AI components. PREVAILING PRACTICE BARRIERS exist since when compared with the use of AI methods, current time series forecasting practices are relatively easy to understand and apply. Hence, the ease of use of these prevailing practices also forms a barrier to change. This general inertia for change is compounded by the fact that many of the existing forecasting tools within the industrial market are provided as part of wider ERP/MRP computerised planning and control systems. There are a wide variety of such systems commercially available and it is generally accepted within industry that such is the nature of forecasting and its associated uncertainties that little can be done to improve forecasting accuracy. Finally INVESTMENT BARRIERS arise because for the project to reach a critical mass a considerable fixed amount of cost and time resources are required to bring to fruition a prototype system. However, the prototype system once developed will provide a means of overcoming the reluctance of the market to investment in this type of technology. Other differences Technology Programme support makes are firstly that the project requires a wide range of technological, scientific and forecasting practical expertise which can only be gained through a collaborative project of this nature. Such support also has significant effects on the quality of outputs by enabling the development activity to examine a range of industry sectors and hence the generic cross-sector aspects of the use of AI forecasting techniques can be better understood and applied.