Long term verses short term production forecast

Many reservoir engineers spend most of their time and effort forecasting for the long-term to meet business objectives where estimates of life-of-field production are required (reserves estimates, valuations of projects etc). However, a number of other business objectives require estimates of production to be made over shorter timescales (nominations under gas contracts, quarterly cash flow estimates and so on). Such a short-term production forecast requires no difference in approach, in principle, to a longer term forecast. However, the important factors affecting the forecast may be different to those affecting a long-term forecast. Many such factors are related to the performance of production and processing facilities or commercial considerations (such as seasonal variations in gas prices). Forecasting methodologies and workflows have developed to integrate models of these constraints with reservoir models. From a practical point-of-view, this often requires use of reservoir models other than 3D simulation in order for the forecast to be completed on a reasonable timescale. There is, of course, no quantitative definition of “short term” or “long term” that is universally accepted, nor is a distinction between the two always helpful or useful. There is certainly considerable overlap, especially as techniques for creating forecasts are not time-bound in any way. For the purposes these guidelines, short-term forecasts are considered to be of a duration of up to one year, but this is somewhat arbitrary. Note that for the very short-term forecast to answer questions such “what happens when I open this choke?” there is little documented either within companies or the open literature. Probably this is due to the highly operational context in which such questions are usually asked. Furthermore, the short feedback times with respect to the quality of decision making are likely to make the need to document in the literature seem irrelevant to most. Nevertheless, techniques are employed whether they are purely based on experience ie, an extrapolation from what happened the last time a similar operation was performed or more quantitatively with single well models – and often in combination with each other. This may be the extent of the analysis required in order to forecast in a situation where a well’s rate needs to be increased to maintain overall field production.

Methodology and Workflows

Beyond this very short-term requirement, a common workflow to create a forecast (or, indeed, a range of forecasts) is to use a long-term forecast such as that associated with a formal reserves estimate, in which understanding of reservoir dynamics is incorporated, and then to vary this forecast over the short-term period. The variations are typically based on one of the long-term forecasts (for example, the forecast associated with the 2P reserves estimate) and different uptime assumptions made. The resulting forecasts are illustrated schematically in Fig 1.

(a) Uptime Variation on plateau

(b) Uptime variation on “central case” long-term forecast, early to mid-decline

(c) Uptime variation on “central case” long-term forecast, tail end production

Figure 1.1 Short-term forecasts based on pre-existing long-term forecasts

Note that implicit in this technique is the assumption that variations in production performance arising from reservoir effects are negligible over the duration of the short-term forecast, compared to the variation in uptime. This is probably a reasonable assumption for times shown in Figures 1a and 1b, but towards the end of producing life (Figure 1c), reservoir and uptime effects may be of at least the same order. Another implicit assumption is that the long-term forecast incorporates constraints placed on production by the processing capacities and pressures, or that the variation of these over the short-term forecast is negligible compared to the uptime variation.

Many forecasters use integrated reservoir and surface facilities models to overcome any limitations imposed by the latter constraint. The goal of these methods is to account for the coupling between reservoir, well and facilities constraints where these vary (pressure at various points in the system being the most obvious). Typically, this involves coupling relatively sophisticated models of the surface network and vertical well performance with rather simpler models of the reservoir (certainly compared with 3D reservoir simulation). The main barrier to producing forecasts with fully coupled models is that of model runtime. Frequently, “look up” tables generated from independently run simulation models are incorporated. Typically, forecasts are required to be generated on short timescales and complex coupled models can take several days to run (depending on the size of the problem). Much current research is, therefore, focused on improving runtimes and methodologies so that forecasts can be generated more quickly (see “The State of the Art” below). However, consideration should always be given to the objectives of the forecast and complex, coupled models may not be required. In other words, the significance of any differences between forecasts generated from coupled models or, say, from those generated from existing long-term profiles or look up tables, in which the implicit assumption is that constraints in the total system are independent of each other, is problem dependent.

A review of the literature (Reprint 2010) suggests that the most successful forecasts have the following features:

• Involve all relevant disciplines in construction of the forecast
• Data availability for the method applied
• Supporting documentation states assumptions clearly
• Explicit inclusion of known events (eg shut ins)
• Use of models that are as transparent and simple as possible
• Quantitative assessments of forecast uncertainty
• Provision of a statement of the economic risk associated with the forecast
• Checking the forecast against actual performance.

As mentioned earlier, the forecasting methodology used is likely to be problem dependent and chosen based in time available, corporate requirements and other constraints, as much as technical merit. Table 1 indicates some of the important features for the forecaster to consider for each of the broad categories of modelling approach discussed here.