Title of Invention

"AN INTEGRATED CONTROL DEVICE OF ARTIFICIAL NEURAL NETWORK WITH AN ELECTROSTATIC PRECIPITATOR"

Abstract

According to the present invention there is provided an integrated control device of artificial neural network with an electrostatic precipitor to control collection of fly ash of the flue gas by the said electrostatic precipitator comprising an artificial neural based controller adapted with opacity set point (2) and an input of opacity feedback for measurement of desired opacity (9) from a capacity monitor (5) mounted at the Electrostatic Precipitator duct (6), the output signals (11) of the said artificial neural network based controller being connected to plurality of fields (13) through twisted pair communication line (10) providing charge ratio signals (11) to the said plurality of fields (13) and the outputs of control signals (12) from each said plurality of fields (13) being connected to the electrostatic precipitators (3) having corresponding number of plurality of fields, the said electrostatic precipitator (3) being provided with an inlet (4) for flue gases and EP dust (6) connected to a chimney (7) having an outlet (8) for release of ash free flue gas to the atmosphere.

Full Text

The invention relates to a device of artificial neural network in the control of electrostatic precipitator to collect flyash from the flue gas. Coal fired boilers in Thermal Power Plants cause substantial air pollution through emission of fly-ash particles. Similarly process plants such as Cement mills and steel plants also, cause air pollution through their exhause gases. Electrostatic precipitator are widely used in many industries for fly-ash particle collection from flue gases. In a thermal power plant, electrostatic precipitator serve to limit flu ash pollution of atmosphere, in addition to reducing, erosion of the induced draft fan blades. 3 Fly ash concentration of about 30 - 60 gm/nm in 3 the flue gas is brought down, often to below 150 mg/nm y using Electrostatic precipitator which are widely used in thermal Power Plant's due to their low pressure drop, high efficiency and reliability, low operation and maintenance cost, and suitability for high temperature fluctuations in flue gases. Particles in the flue gases entering the Electrostatic precipitator are charged by Corona ions from the discharge electrode. These charged particles get deposited on collecting plates due to the electrostatic forces of the electric field near the plates. These ^ollected particle form into a layer on these plates, which rs then dislodged by rappers and collected in the hoppers below. There are disadvantages associated with the efficient operation of the present system of an electrostatic precipitator. One of the main disadvantages of the present system of operation of an electrostatic precipitator is that their efficiency depends upon many variables like charging method, particle size, flue gas temperature and dust resistivity. Another disadvantage with the present system of operation of an electrostatic precipitator is that for optimum efficiency it is essential to closely monitor and accurately control the key parameters of the control system. The precipitators exhibits a large number of non-linearities and time lag. The performance estimates and the precipitator design one based on empirical relations and assumptions which do not really hold good in actual operation. Electrostatic precipitators are physicall divided into several passes. The inlet dust concentrations through these passes are expected to be same theoretically. But due to geography of mechanical structure, the flow rate and dust concentrations are not same through these passes. Hence the load distribution is not equal in all passes.. The collecting efficiency suffers due to these uneven flue gas velocities. It is difficult to vary the control parameters following the fas velocity variations a there is no fixed relation that exists or that can be established between these process parameters and. control parameters. Similarly another process parameter which influences the efficiency is flue gas temperature. A narrow band of flue gas temperature will give optimum performance of electrostatic precipitator. Good quantitative models which incorporate the effects of these variables are not available for full scale industrial precipitators. Therefore the main object of the present invention is efficient functioning of the electrostatic precipitator by minimizing the power consumption and maximizing dust collection. Another object of the present invention is to provide a good quantitive model that can be applied to full scale industrial precipitators. Further object of the present invention is to develop several control strategories to adopt to meet the broad requirement and overcome problems in performance optimization of varying the control parameters following gas velocity variations and the flue gas temperature variations. According to the present invention there is provided a device of artificial neural network in the control of electrostatic precipitator to collect fly ash from the flue gas comprising an artificial neural network based control having a opacity set point and an input of opacity feedback for measurement from a opacity monitor mounted at the EP duct, the output signals of the said atificial neural network based control connected to plurality of fields through twisted pain communication line providing charge ratio signals to the said plurality of fields and the outputs of control signals from each of said plurality of fields connected to the electrostatic precipitator having corresponding number of plurality of fields, the said ESP is provided with an inlet for flue gases and an EP duct connected to a chimney having an outlet for ash free flue gas. The nature of the invention, its objective and further advantages residing in the same will be apparent from the following description made with reference to non-limiting exemplary embodiments of the invention represented in the accompanying drawing. Figure 1 shows the schematicall the control system using artificial neural network. Figure 2 shows a flow chart for neural network based control. Figure 3 shows a typical neural network model. An exemplary embodiment of the invention of a device of artificial neural network in the electrostatic precipitator is shown in Figure 1. The heart of the system, the controller is a computer for aritificial neural network based control (1). The ANN based control (1) is provided opacity and an input of opacity feedback for measurement (9) for monitoring the various parameters of the output flue gas of the precipitator. The output of the ANN based control in the form of charge ratio signals (11) is connected to fields (13) by a twisted pair communication line (10). In the preset embodiments there are 12 nos fields (13) designated as ICE-C 1 to 12. The fields (13) provide control signals (12) to the electrostatic precipitator chamber (3) with 12 fields. The ESP (3) is provided with an inlet (4) for flue gases, EP duct (6) and a chimney (7) with an outlet (8) for ash free flue gases. An opacity monitor (5) is provided at the EP duct (6) for providing opacity feedback (9) to the ANN control (1 ). Figure 2 shows the process chart for the ANN based control and numerals 14 to 19 and describes the procedure used for this invention. Figure 3 shows typical neural network model. Input variables ( 20) is shown at the bottom of the model and three layers from bottom are input layer ( 21), hidden layer ( 22) and output layer ( 23). In the present invention a device using neural network method of controlling electrostatic precipitator in a very efficient way has been developed. This device consists of developing the neural network model of the precipitator and using it to determine the control parameters viz current limit and charge ratio to be applied to the precipitator in order to get the desired performance. NEURAL NETWORKS Neural networks are information processing systems made up of an interconnection of simple processing elements and are based on models of how the human brain works. They are capable of learning and appear to be better than digital computers in dealing with systems which have in imprecise data non-leniarities and randomly varying parameters in certain types of applications. Neural networks are capable of recognizing patterns among a seemingly random data base, in a very efficient manner and quickly. They can also tolerate contradictory and imprecise information. During training, they learn the underlying relationship between the input/output data presented to them. Later, when a new input pattern is presented to the input of a trained network, it will produce a most likely output. The neural network model used consists of processing elements organized as layers i.e. the input layer (21); the output layer ( 23) and a hidden layer ( 22). The interconnections between the elements of different layers have " weights" with which the output of a processing element is multiplied before applying as input to the next layer. Altering the weights of these interconnections modifies the network behaviour. The learning procesure involves presentation of a series of input/output vector pairs of process data to the network and calculating the weight matrix using "back propagation learning algorithm". The input layer is passive. It merely receives the import data values and hence the number of processing elements in this layer is equal to the number of input parameters or variables ( 20) to be processed . In the case of electrostatic precipitator, these are the values of various measured and controlling parameters. The hidden ( 22) and output ( 23) layers are active. The processing elements in these layers perform simple mathematical operation on their inputs and produce one output ( 24). All processing elements can have one or more inputs but only one output. The number of processing elements is equal to the number of outputs the network is expected to produce Hence the number of elements in the input and output layers are decided by the application. However, the number of elements in the hidden layer ( 22) is not fixed by the input/outputs. Choosing their number for good network performance, involves experimentation. Each connection between processing elements has a weight factor associated with it. These constitute networks "intelligence" or knowledge. The training algorithm adjusts these weights during a training, such that sum of squares deviation in its output is minimized. The processing element multiplies each of its inputs with the corresponding weight, sums the products and passes it through a sigmoid function to product its output. DETAILED DESCRIPTION OF THE INVENTION In the conventional method, the opacity is controlled only by adjusting the currents into ESP. Adjusting the charge ratios is manually done by operator. The present invention relates to use of neutral network technique to control the ESP (3), taking into account charge ratio ( 11). In the electrostatic precipitator (3) in which this principle was established, there is provided control over the charging currents of the 12 fields, charging ratios ( 11) for 12 fields ( 13) and also the sparking rate. The charge control, input current control as well as parameter monitoring is performed. by a multiple microprocessor based distributed SCADA system capable of being programmed to vary the operating conditions of the precipitator . Input vectors for training the network were obtained at various operating conditions of the precipitator by adjusting the control parameters such as charging duty cycle and current inputs. The outputs for training the network were obtained by measuring the opacity level for each of the input vectors. For making the movel in this particular application the following inputs and outputs are used. Inputs: (20) Field currents of all 12 fields Field voltages of all 12 fields The spark rate ( number of sparks per minute)of all 12 fields Outputs: ( 24) Current reference values of all 12 fields The charge ratio settings of all 12 fields The efficiency of Electro Static POrecipitator (3) is a collective effect of all the fields ( 13). In the present system each field controller (1) works on its own depending on its inputs (2). There is no relation between the controlling parameters of other fields of the same pass. There are number of parameters that are not available for the controller (1). By this optimization of a particular field may deteriorate the performance of other fields. Thus the present type of control system can not co-ordinate the inputs of all fields to optimize the over all performance of ESP (3). Under the above conditions an Artificial Neural Network (ANN) can work can answer the above problems. The ANN can work effectively even in the absence of some of the input parameters. The ANN is a control algorithm that can be implemented for any aize of the ESP/. The ultimate aim of the ESP is to minimize the output emission being measured by an Opacity monitor. The ANN studies the process from the data collected from the process itself. This data is the consists of the effects of all the fields under various measured and unmeasured input conditions. The knowledge gained by the Ann system includes the effects of all the fields over each other. This knowledge is used for controlling the inputs to the ESP for optimizing the process. The invention described hereinabove is in relation to a non - limiting embodiment and as defined by the accompanying claims. WE CLAIM": 1. A device of artificial neural network in the control of electrostatic precipitator to collect fly ash from the flue gas comprising an artificial neural network based control (1) having a opacity set point (2) and an input of opacity feedback for measurement (9) from an opacity minotor (5) mounted at the EP duct (6), the output signals (11) of the said artificial neural network based control connected to plurality of fields ( 13) through twisted pair communication line (10) providing charge ratio signals ( 11) to the said plurality of fields ( 13) and the outputs of control signals ( 12) from each of said plurality of fields (13) connected to the electro static pricipitator (3) having corresponding number of plurality of fields, the said ESP (3) is, provided with an inlet (4) for flue gases and an EP duct (6) connected to a chimney (7) having an outlet (8) for ash free flue gas. 2. A device of artificial neural network in the control of electrostatic precipitator wherein the said ESP (3) are physically divided into several passes corres ponding to plurality of fields ( 13). 3. A device of artificial neural network in the control of electrostatic precipitator wherein the said ANN based control (1) model having input variables (20) with three different layers of input layer (21), hidden layer ( 22) and output layer ( 23). 4. A device of artificial neural network in the control of electrostatic precipitator wherein the said artificial neural network model the inputs variables (20) comprises field currents, voltages and spark rate of all fields ( 13) and outputs ( 24) comprises current reference values and charge ratio settings ( 11) for all the fields ( 13). 5. A device of artificial neural network in the control of electrostatic precipitator wherein said plurality of fields ( 13) comprises twelve in number. 6. A device of artificial neural network in the control of electrostatic precipitator wherein said plurality of control signals (12) comprises twelve channels for twelve chambers of the said ESP (3). 7. A device of artificial neural network in the control of electrostatic precipitator as herein described and illustrated.

Documents:

1487-del-1997-abstract.pdf

1487-del-1997-claims.pdf

1487-del-1997-correspondence-others.pdf

1487-del-1997-correspondence-po.pdf

1487-del-1997-description (complete).pdf

1487-del-1997-drawings.pdf

1487-del-1997-form-1.pdf

1487-del-1997-form-19.pdf

1487-del-1997-form-2.pdf

1487-del-1997-form-3.pdf

1487-del-1997-form-5.pdf

1487-del-1997-form-6.pdf

1487-del-1997-gpa.pdf

abstract.jpg