ESTIMATION OF INCIDENT PROBABILITY

EIP Fuzzy Logic Model

Background

The fuzzy logic approach is based less on data from the HDB, and more on expert judgement with regard to what (traffic) conditions are more likely to produce incidents. It uses fuzzy logic to capture the experience of traffic engineers and traffic police and produce estimates of the incident probability. This approach was initiated because of problems with the data needed for a full statistical model. Prominent problems with the statistical model are the time stamp of incidents and the lack of weather data.

The model is based on expert opinion from sessions with police and traffic engineers. The interview responses and knowledge were then transcribed into fuzzy rules. The steps taken in this process were:

• First, it was established which variables were considered to be the most important for the incident probability.
• Then, which variables were to be taken into account as inputs to the model. These are not necessarily variables that can be derived from data from the PRIME HDB.
• Next, and the most important part of the knowledge we needed from the experts, were the fuzzy rules for the rule base. This information was transformed into the desired form of the fuzzy rules.
• Finally, each of the variables was divided in a number of fuzzy sets, which was done in accordance with the suggestions of the experts.
• In the defuzzyfication process the output of all the fuzzy rules in the rule base was combined into a single output. That output is a representation of the risk incurred by a single driver when travelling the roadway link during the time interval under consideration.

The fuzzy rule base contained 119 rules, of which 14 relate to time variables, 27 to traffic conditions, 27 to weather conditions and 15 to road conditions. The combination of variables between classes resulted in another 36 rules. This version was used to test two examples of conditions that might occur on a motorway.

This fuzzy logic model was developed and tested off-line by TNO. The model has not been specified for a specific test site, but the fuzzy rule base refers to situations as found on Dutch motorways. To apply this model in other circumstances (other regions, other road types), the fuzzy rule base needs to be redefined for local circumstances.

The EIP fuzzy logic model works with input files describing certain conditions over time, e.g. traffic volumes or precipitation intensity per minute. These are, minute by minute, evaluated by the model, which considers the degree of familiarity for each rule in the fuzzy rule base: are traffic volumes high, medium, or low? Combined with the precipitation intensity (high or low?), what is the resulting risk? The model summarises this in the output variable, which is a representation of the risk incurred by a single driver when travelling the roadway link during the time interval under consideration. Multiplied by the traffic volume, this gives the overall risk for all drivers together.

This first version of the fuzzy logic EIP model is constructed with opinions, rather than with quantitative data. Weather variables played an important part in the fuzzy rules, and this type of data was not available for the Barcelona test site. Therefore, it was not possible to develop a Logit and a fuzzy logic model with a similar set of variables. Because of this, it was chosen not to test the fuzzy logic model with data from the Barcelona test site. Instead, two test data sets were constructed for two typical situations frequently occurring on Dutch motorways. These were used to assess whether the fuzzy logic model responded in a logical way.

The two examples that were tested are:

1. Adverse weather conditions (sample of 150 minutes)
2. Shock-wave of a moving jam (sample of 30 minutes)