The key to an accurate production picture is knowing how each well is performing in real time. The measurement of oil well output has always been a challenge because the flow of oil is mixed with gas, water, and sand, and a direct measurement is a complex and expensive process. While multiphase flowmetering (MPFM), which was developed during the 1980s, has made real-time measurement practical, it is still not considered economically viable on a per-well basis in most situations.

Well output, however, can be inferred or estimated from other measured parameters. A number of equations have been established that allow liquid (oil and water) and gas flow rates to be calculated using wellhead pressures, fluid properties, temperatures, and other field-available variables. A common measurement point is at the well choke. Measurements of the pressure upstream and downstream of the choke can be used to estimate flow rates.

This figure shows a graphic presentation of well performance. The size of each block corresponds with well output, and color reflects level of performance. (Images courtesy of Matrikon)

A number of reliable well rate estimation techniques have been programmed as part of a commercially available well performance monitoring system, all dependent on different first-principle models and different field measurements.

However, once a number of equations has been introduced, the situation begs the question, “Which estimation is the best to be used in each case, i.e., each well?”

Best real-time estimates

To solve this, a technique known as best real-time estimates (BRTE) is utilized, in which a user-defined rule is set up to determine which one of the rate estimates is the most accurate one to be used in subsequent calculations and aggregations, for example, field and facility level totals.

Once the estimates have been calculated and the BRTEs selected, well-level key performance indicators can be calculated and displayed in a graphic diagram, with individual wells being represented by blocks according to expected output. The color of the display indicates whether the well is performing at expectation (green), above (blue), or below (yellow or red). Blocks can be clicked on, allowing the engineer to investigate individual well conditions.

Benefits associated with using this type of solution include immediate feedback on field optimization actions, a reduction in deferred production from wells, the early detection of abnormal well instability or slugging, and savings on costly electric submersible pump (ESP) repairs due to the detection of wellbore communication and interactions between an ESP well and a gas-lifted well.

In terms of equivalent production increments, this type of solution is consistent with similar technologies in achieving approximately 5% increased throughput. In addition, a better strategy for the shut-down and start-up of the two communicating wells has prevented ESP repair costs in the order of millions of US dollars, paying back for the initial investment many times.

The spikes shown in this figure will be eliminated as noise for flow estimation purposes but may represent useful information for maintenance engineers.

A cure for noisy data

A key challenge with the BRTE method is that signals in the field are prone to noise, and noisy readings play havoc with the calculations. There are two reasons for this. First of all, a false reading can throw a formula wildly off — the error could be greatly enlarged if, for example, the reading is multiplied by a coefficient. Secondly, a false reading won't look false to an equation, and the results may appear perfectly credible to an observer even if completely incorrect.

Consequently, it is absolutely essential to remove all noise and unnatural behavior before there can be any confidence in these methods. The Well Performance Monitor system uses statistical algorithms to accomplish this. Data cleansing is essentially a collection of algorithms that achieve what we call a cleansed tag. The overall principle is that raw tags are retrieved from the field that are full of noise and other complications. From that, users produce a cleansed tag that is guaranteed to either have a real-life value from the field or have a no-bad indicator on it.

Loss of signal during transmission is the most common form of noise. The signal typically originates in an instrument and then passes through a control system, an interface, and a satellite link. Any breakdown in that chain will cause a false reading. Furthermore, the abnormal behavior of the signal can take a variety of forms depending on where the interruption is.

Monitoring systems will often receive a signal that remains frozen over an uncharacteristic period of time. In other situations, a signal might be showing large out-of-tolerance fluctuations or spikes. Sometimes these noisy readings may indicate an equipment malfunction.

However, what is noise for one person may be information for another. For example, a spike of gas may be going through a gas lift choke valve and then going back to zero (Figure 2). For a petroleum engineer interested in using the gas lift rate to estimate a well rate, that spike is noise. But for a maintenance engineer, it is a very valuable piece of information that may indicate a leak in the valve that needs to be sorted out. For that reason, raw signals are never overwritten. The system keeps both raw and cleansed.

The Well Performance Monitor software automatically detects anomalies such as flat lines, spikes, and lost signals by comparing real-time readings against statistical norms. Data is then cleansed by replacing missing values, which eliminates spikes and other out-of-range readings. The software also looks at the whole picture — whether the data being used for a specific calculation is of high enough quality to maintain the appropriate level of confidence. If not, the method is flagged as having insufficient data, and an alternative method is used for well rate estimation.

One of the strengths of the data cleansing algorithms is that they are self-adjusting — they do not need any a priori tuning as they self-adapt to the statistic characteristics of the sample.

As with any model-dependent system, in order to keep the models and estimates current over the long term, the field’s petroleum engineers need to constantly analyze the results, with an eye towards any major change in oilfield conditions that could cause a major shift in the estimated results.

The accuracy achieved by Well Performance Monitor installations have been a revelation to visiting engineers. The results have been impressive; an operator of a small North Sea field is using these techniques on a daily basis to monitor production from all wells. In this operation one can clearly see very tight field estimates and field level MPFMs (see figures 4 and 5), also seen with the facility level fiscal meter. The higher stability of estimates (black series on figures 4 and 5 is noticeable compared to the MPFMs (red series with triangular markers on same figures). See images below.

A number of operators have recognized the importance of well performance visualization using a combination of different techniques such as the ones described above. A number of well performance monitoring systems have been installed and are operating in the Middle East, Brazil, and the North Sea, with new deployments being planned in Norway and the Middle East.

In January 2010, Matrikon announced that Statoil, the Norwegian-based international operator, had selected the well performance monitoring system as the basis for its enterprise-wide visualization system. While the scope of the project goes beyond what has been discussed here, these principles remain at the core of what many oil and gas operators today want to achieve.