The Use of Simulation Models and Human Advice to Build an Expert System for the Defense and Control of River Floods

1. Introduction

One practical application of Artificial In­ telligence that has awakened the most interest over the last years is that of expert systems.

The development of technologies for the sys tematic treatment of inexact reasoning (Shortli~ ffe (1), Duda, Hart and Nilsson (2) Shafer (3), Lowrance (4) and Zadeh (5), among others) as well as the construction of systems (Dendral (6), My­ cin (1), Prospector (7) and R1 (8)) applied to the world, has contributed to this. However, on developing an application on these lines, alth­ ough having tools available (Teiresias (9), Ops (10), Rosie (11), etc.), the critical problem is the shaping of knowledge on an issue in a set of rules in such a way that the answers to the different questions are coherent enough and ad­ justed to the actual knowledge of the problems.

The personal experience of the author of this paper in some still unfinished attempts of It is therefore considered, that the cons­ truction of expert sustems may be simpler in areas of knowledge where important theoretical support can be counted on, by building the infe­ rence rules integrating elements of the diffe­ rent theoretical supports or as a synthesis of results obtained through the application of mo­ dels.

The present paper describes the conceptual framework on which the design of an expert sys­ tem is based now under study at the Minist.ry of Public Works, which supports the decisions du­ ring a rise of water in a floodable region. Attention is focused in the definition of rules based on a set of mathematical simulation models of the different aspects of behavior, as an ele­ ment of previous support.

2. The Problem

In autumn of 1982, important floods took place in the Spanish coastal Mediterranean re­ gions, pointing up the needs of technical ins­ trumentation designed to yield real time ans­ wers to a series of questions which arise du­ ring the event . In effect, in the Mediterranean area, the intensive explotation of land by tou­ rism along the coast brings about the existence of substructures and buildings that block drai­ nage as well as agricultural explotations giving rise to towns near the river courses. Furthermo­ re, the climatological conditions bring little precipitation very concentrated in time which produces important flood peaks .

Therefore, the characteristics of the pro­ blem demand a rapid taking of decisions, given the important concentration in time and, further more, the complexity of land use and the system­ of hydraulic substructures, provided that any d~ cision may influence the general behavior of the system, which does away with improvisation since actions which might appear acceptable locally might be un acceptable overall due to the interac tion of factors.

3 . The Syste!'!_

In accordance with the preceeding considera tions, the construction is required of a sys tem­ that gives support to the people responsible for control and civil defense during the flood period in order help them to assess the situation at any moment in the places of interest.

Taking into account the spatial structure of the area under study, its present situation and recent evolution, the clasical answer to this de­ mand would be:

• A data base
• A knowledge base ' 1 able to understand'' the va lues of the variables of the data base in ti me and space and able to give on time ans­ wers to the problematical aspects of the ac­ tual situation, and in the intervals of ti­ me in which it is desired to make forecasts.

4. The Data Base

For the purpose of information in the flood area, two subsystems may be distinguished:

• The hydrographic subsystem, made up of:
• A catching watershed with large slopes that receives the rain, and can be described by :
• a set of catchment areas
• a tree-like network of rapid drainage
• a low zone, whose axis is the river produ­ cing the floods, which can be described as the axis ·of the river together with a se­ ries of floodable pockets conn·ected tb the river through the temporary courses created during the flooding.

The whole is represented in the figure 1.

FIGURE 1

• The social-territorial subsys tem made up of the sys tem of towns, agricultural explotations and road/rail networks connecting it .

Both subsystems occupy the region so that du­ ring a flood event the s tate of the first produ ces effects on the second (flooding of agricul~ tural explotations, towns, blocking roads and railways, etc) .

The information system should present a picture of both subsystems so that questions can be answered on the effects on the social-territo­ rial system. Accordingly, the information sys­ tem should be:

A simplified geographical information system spatially bounded by the contours of the agri­ cultural explotations, urban areas and catch­ ment areas in the watershed and with the spa­ tial definition of the transport network (rail­ ways and highways) and the hydraulic network (steep high watershed network and main river in the low zone).

• A real time information system with:
• Rain intensity in reception areas
• Water levels at control points

5. The Knowledge Base

This is made up of the set of inference ru­ les that, utilize the values of the data bas·e and allow us to make pertinent assertions in or­ der to evaluate the present and predictable situ~ tions.

In principle, these assertions would be:

• Pertinent to technical control:
• ~ecommendable actions on how to operate flood gates in reservoirs, by-pass cha­ nnels, etc, including blasting of ba­ rriers (roads producing backwaters).
• Pertinent to Civil Defense:
• Water levels at different critical points
• Alternative paths to the interrupted connnunications
• Evacuation of urban Communities

The knowledge base must permit inferences on this type of declarations:

• In different spatial locations
• In different intervals of future time

Due to the need of quick answers it is not possible to use a set of complex simulation mo­ dels, on line during the flood, but it is possi­ ble to build an inference engine that synthesize the knowledge obtained from previous model runs. This i nference engine will have three main lines of reasoning:

• Reasoning on the state in every moment T
• Reasoning about the possible future states
• Reasoning about advices of control (propo­ sals of decisions)

• There are two types of rules for reasoning about the state in a time instant:
• Situation rules (SR). They allow the deduction of possible water levels in significant points of the flooded area based on data from control points furnished by the information system.
• Impact rules (IR), for estimating the level of impact in the urban and agricultural areas. They are based on the levels at the signifi­ cant and control points and on their rela­ tionships with the socioeconomic system.
• There are three types for reasoning about the future:
• Rain rules (RR) in arden to estimate rain in­ tensity in the near future based on pluvio­ graphs shapes and their observed evolution.
• Inflow rules (IFR) to compute the flows at every watershed outlet, from the present sta te of levels in the reception area and the es timated rain intensities.
• Flood level rules (FLR) to deduce the expec­ ted water levels at the control points based in the actual state in these points and the possible future evolution in the inflow from the watershed areas.

All these rules will be defined for fixed fu~ ture time intervals (i.e. rules of 1 hr, 2 hr 6 hr ·... )

Once estimated the water levels at the con­ trol points, with different degrees of certainty the expected future levels and impacts at signi­ ficant points, can be infered using the rules SR and IR, defined for the state at time T .

• There are two types of control rules:
• Operation rules (OR) in order to propose varia tions in the control policy of dams, the blas ting of obstacles that close the flows in so~ me areas, etc.
• Civil defense rules (CDR) to propose alterna­ tive travel paths in transport networks, .. etc.

The general structure is summarized in fig. 2 .

6. System Operation

The system operates in the following way:

• A real time information system furnishes the basic data at specified time intervals. From these data the system produces the basic pre dicates that represent the premises for the­ inference engine.
• Based on these premises, real time inferen­ ces can be made on the situation of areas uncontrolled by the information system.
• It is also possible to make inferences on the future situation at the controlled and uncontrolled areas. The time intervals for inference may be chosen among those fixed in the rules (i .e. one cauls use 30 min, 1 hr, etc ... inference rules). The fact of updating the premises at every instant, doesn't imply to repeat all the inferential process. It

will be posible to design a truth maintenance sys tern to modify only those conclusions affected by­ the modified premises.

7. Building the knowledge base

The facts related by every set of rules are complex enough to define the rules only by the criteria of the experts.

Hydraulic engineering offers a set of mathe­ matical methods for modelling the behaviour of water levels at both areas: reception and flooda­ ble. The general methodology to define the rules is:

• Use the existing data to calibrate;
• Statistical models of pluviogram structu­ re
• An analogical model of the catching waters hed areas -
• A numerical integration model of the uns­ teady flow equations for the floodable area
• Development with the advice of the experts, of a plan of model runs oriented to define the rules
• Definition of the rules based on:
• The results of the ar t ificial experience created by the model runs
• The criteria of the experts evaluating the reliability . of the theory of the mo-

dels and the quality of the calibration

• The inclusion of complementary rules, by the experts, on impacts, dams operation and civil defense.

For the definition of rules based on simulation results:

• A frame of discernement with predicates re­ presenting the basic facts must be defined
• Each rule must be described in terms of for­ mulae composed of the basic facts specified for the different objects (rain levels in watershed areas, water levels in control and significant points, points of significant impacts ... )
• The grid of simulated cases of every model must enable to establish the possibility of inference of some basic facts from a composi tion of facts. For that, the inspection of­ the number of simulated cases in which both the antecedent and consequent are satisfied compared with the total cases where some of them is satisfied, will allow to take the de cis ion to include the rule and the es tima-· tion of its strength.

The rules to forecast the rain intensLtLes can be built up by the analysis of a sample of storms time series r.egistered in the data base.

For the representation of the behavior of the watershed area one of two classic simulation systems will be used: Stanford Watershed Simula­ tion Model IV (13) or SWMM (Storm Water Manage­ ment Model) (14).

The inflow rules (IFR) and the operation ru­ les (OR) for the dams in the watershed can be built by analysis of the results of runs with the watershed simulation model.

For the simulation of the unsteady flow in the floodable area, an implicit Scheme (15) will be used for the numerical integration of the Saint Venant equations. This numerical model is able to compute the levels in the river and flood areas at every time instant, based on these re­ sults is possible to build the situation rules (SR) and, by analysis of the relationships bet­ ween the increments of flows from the watershed and the increments of water levels, it is possi­ ble to build the flood level rules (FLR) for pr~ diction.

Finally by expert analysis of the evolution of water levels and the knowledge of the situa­ tion of the social-territorial system, the impact rules (IR) and the Civil defense rules (CDR) can be defined.

8. Conclusions

The specifications and some basic aspects on the design of an expert system have been pre­ sented. The building of the system based on a set of simulation models runs permits the definition or r'.lles with an important level of insight in the system behavior, and allows the experts to work in real expert tasks (Advising on the defi nition of the set of models runs to elucidate - the key aspects of behavior, adjusting the camp~

ted results based on reliability criteria of mo­ dels formulation and calibration,and including final rules on dam operations and civil defense)