Design of Water Pump Station Control in Polder System using IoT and Hybrid Fuzzy Neural Network

This study aims to design a prototype of water pump station control in a polder system using IoT and artificial intelligence at Sringin Polder system. The design includes the hydrology analysis to define how the pump station should be set to control the level of the embankment. The second part is the IoT system that contain the sensor and interface to the water pump in the pump station. The third part is the core of control system using hybrid fuzzy neural network as the heart of the control system. Pump station has important role in the polder system to control the level of water in embankment. Five pumps in the system should operate based on schedule determined by the hydrology modelling. Since it is located in separate area with the control room, MQTT protocol is used to deliver data communication in both ways, to send the water level detected by ultrasonic sensor and also to switch on and off the pump by control system. In other hand, hybrid fuzzy neural network is used to control the pump scheduling based on water level detected.


Introduction
The pump station is a vital component in the management of the polder system for the management of water resources in flood prevention [1].Based on the hydrological analysis that has been made, the calculation results are obtained, how much water discharge must be moved by the pump and how many pumps must work to achieve stable water levels if there is an increase in water discharge [2].The following table is the result of these calculations.Pump operation scheduling are closely related to operational costs.With optimum pump operation schedule, operational cost will reduce and effectiveness will increase.Efficient use of fuel and pump maintenance so that the pump lasts longer [3].In operation, fuel must be regulated to be efficient, namely by activating only the pumps that are really needed, so that it is not wasteful in using fuel or electricity as propulsion.The pump must be active according to the amount needed.In the Long term.It should also be considered that in order for the pump to work better, the load distribution of active pumps is also regulated so that it is evenly distributed.This relates to the scheduling of pump activity.

Experimental model
Nowadays, Artificial Intelligence has been very important to solve complex real-world problem, by combining knowledge, technique and methodology from various source [3].In other hand, communication technology, sensor, and embedded system has enabled the AI system to be developed with more advance functionality.For example, in previous study, Smart IoT can be implemented to monitor Flood [4] [5].A successful model of AI basically imitates human brain in reasoning and making decision.The goal is to convert expert knowledge into an intelligent machine [6].The system should able to sense the environment using sensor (perceive) and process the information to act on its perception [5].
Structured network of an AI system is consisted of some node that has direct connection to each other.Each node has its own functionality in processing input signal and generate output to be sent to another node.It is implemented in an adaptive network in which behaviour of input-output is determined by modifiable parameter.The changing of any parameter in the node, will affect the behaviour of overall adaptive network.There are two categories of adaptive network: feedforward and recurrent.This study will develop a prototype as an adaptive network with feedforward neural network.This study defines the pump station as the controlled entity.To perceive the environment parameter, ultrasonic sensor is used to read the water level in embankment.Another parameter to be calculated is duration of the pumps that determined the operation schedule of all pumps.Pump station consist of five pumps as the actuator to maintain the water level in embankment.A relay is used to switch on and off the pump 2.2.Connectivity.
Interface node communicate to processing node to send the water level data as in put in the controller.It is also received command data to react upon the changing parameter in environment.
• In physical layer, data communication is enabled by establishing TCP connection over Wi-Fi wireless network.This connection is enabled by the ESP32 Module.
• In transport layer, MQTT (Message Queue Telemetry Transport) is used as messaging protocol to publish/subscribe data.ESP32 Module act as an MQTT client, that will send data to a broker by publish command and receive data by subscribe command.

Controller
Controller has responsibility in perceive data, analysed them and send back to corresponding node to react based on the interference engine decision.A conventional fuzzy logic controller, has limitation in handling a more complex model.Fuzzy Neural Network has equivalent functionality to fuzzy interference system [6].Instead of creating fuzzy rule to interfere the system, fuzzy neural network combines two fields: fuzzy logic and neural network [8].In conventional fuzzy inference system, output is determined by a set of rules that is composed by an expert who is familiar with the target system be In contras, neuro-fuzzy will used a neural network model that has been trained with a dataset that reflect the desired behaviour of system [9].Previous study has demonstrated a promising result regarding the effectiveness of the algorithm and better result prediction.
There are two type input to this models, water level and duration of the pumps.Based on the table 1, there should be 5 different level which require different total outlet flow.To accommodate the desired operation of the pump based, the operational schedule should be arranged according to parameter input, i.e., level and duration of each pump.Those parameters are gathered in crisp, then convert to a membership value according to the membership function.The graph is shown in the figure 2   After the input variables are measured, the data is used as input for a hybrid fuzzy-based control system.The process is as follows: input data, consist of water level data and status of the pump is gathered from the field, using a subscribe function.Then the data gathered is pre-processed by normalizing the water level data.Pump duration is calculated.These crisp variables, converted into fuzzy membership sets using scikit-fuzzy library in python [7].Instead of using fuzzy rules to determine input, in hybrid fuzzy artificial neural networks are used to replace complicated fuzzy rules [8].The following is an example of a CNN neural network configuration [9] to determine which pumps are active based on the fuzzy input of the level and duration of each pump: fuzzified input, act as input in the input layer, there will be 5 level categorization and 5 duration categorizations from 5 pumps, totally 30 input parameters.From input layer, next layer is the rules layer and output layer.With 5 pumps condition, on or off, there are 5 nodes in the output layers.Schematic is shown in Figure 4.

Hardware Implementation
Basically, this system is divided into 3 parts.First part is the interface to the Pump, realized by an ESP32 Modul.This module is responsible as interface to the pump as actuator, gather data from sensor, and send/receive data to the cloud.Second part is a mini computer that will host the neuro fuzzy controller.It is implemented using a raspberry pi.All the control process is done in this node, including collect data from the cloud and also process and analyse the data to inference.The third part is a Wi-Fi wireless network to connect all the node.

Human interface
After all the hardware and software is set up, the system works as expected.To ensure that the system is going well, a monitoring system is required.Thanks to MQTT, it is easy to porting data to everywhere, without neglected data security [10].By using Smartphone, we can easily monitor the data, using an MQTT client software, available in android or PC. Figure 6 below is displaying the result of pump station monitoring using MQTT client application.

Conclusions
Realizing that pump station management is very important in a polder system, it is necessary to develop a pump operational management system.This study aims to implement IoT, Artificial Intelligence and cloud computing technology on a prototype pump station management system in the Sringin polder system.The prototype that has been developed shows that pump management can be done automatically according to the expected functionality.The pumps work according to a schedule that is based on the water level and the working duration of each pump.Besides being functionally able to work well, the system can be monitored properly thanks to the Smart IoT that connects every existing node.For the next study, it is necessary to optimize the control part, by applying an advance algorithm in optimizing the desired behaviour of the system

Figure 2 Figure 3
Figure 2 Water Level Membership function graph

Figure 4
Figure 4 Neural network architecture

Figure 5
Figure 5 Hardware configuration

Figure 6
Figure 6 Monitoring Display (on android phone)

Table 1 .
Pump outlet flow requirement based on water level