Method of artificial nueral network loadflow computation for electrical power system

摘要

ANN Loadflow Computation Method (LCM) is invented involving input vector composed of net nodal injected real power (P), reactive power (Q), and diagonal elements of conductance and susceptance matrices multiplied by squared initial guess node voltage magnitude (V) or sum of its squared real and imaginary components. Training, and testing and validating input data sets/vectors are generated by applying uniform and non-uniform scaling factors applied to base case loads P and Q at PQ-nodes, and resistance and reactance of network branches. These scale factors are increased until Loadflow Solution by a conventional LCM diverges due to V or angle (θ) instability. V and θ values in the solution just before divergence are respective stability limits, and corresponding values in Loadflow Solution provide direct measure of respective stability margins. Suresh's diakoptics based feature selection technique is presented for calculating one node variable with one neuron in each of ANNs output layers.

主权项

1. A method of forming and solving an Artificial Neural Network Loadflow (ANNL) computation model of a power network to affect control of voltages and power flows in a power system, comprising the steps of:
obtaining on-line or simulated data of open or close status of all switches and circuit breakers in the power network, and reading data of operating limits of components of the power network including maximum Voltage×Ampere (VA or MVA) carrying capability limits of transmission lines, transformers, and PV-node, a generator-node where Real-Power-P and Voltage-Magnitude-V are specified, maximum and minimum reactive power generation capability limits of generators, and transformers tap position limits, obtaining on-line readings of specified Real-Power-P and Reactive-Power-Q at PQ-nodes, Real-Power-P and voltage-magnitude-V at PV-nodes, voltage magnitude and angle at a slack node, and transformer turns ratios, wherein said on-line readings are the controlled variables, generating input and output data sets comprising simulation of feasible and continuous nonlinear operating region of the power system, and using said input and output data sets in obtaining trained, tested and validated, formed, and stored ANNL computation model by using general purpose computing apparatus, wherein different input data sets each of dimension 2n is generated as modified specified real, RPp and reactive, RQp power injections at node-p given by equations (3) to (10), by changing values of operational parameters Ppg, Ppl, Gpp, gp, and Qpg, Qpl, Bpp, bp corresponding to different operating condition of the power system,
RPp=(Ppg−Ppl)−Vpo2(Gpp+gp)  (3)RQp=(Qpg−Qpl)+Vpo2(Bpp+bp)  (4)ORRPp=(Ppg−Ppl)−(epo2+fpo2)(Gpp+gp)  (5)RQp=(Qpg−Qpl)+(epo2+fpo2)(Bpp+bp)  (6)ORRPp=(Ppg−Ppl)+Vpo2(Gpp+gp)  (7)RQp=(Qpg−Qpl)−Vpo2(Bpp+bp)  (8)ORRPp=(Ppg−Ppl)+(epo2+fpo2)(Gpp+gp)  (9)RQp=(Qpg−Qpl)−(epo2+fpo2)(Bpp+bp)  (10)Where, p=1, 2, . . . , n in the power network of the total number of n-nodes, and modified specified real, RPp and reactive, RQp power injections at node-p calculated by equations (3) to (10) are normalized, Gpp and Bpp are real and imaginary components of diagonal elements of admittance matrix without shunts, gp and bp are real and imaginary components of total shunt admittance at any node-p, Ppg and Qpg are specified real and reactive power generation at node-p the values of which would be zero in case of no generation at any of the nodes, Ppl and Qpl are specified real and reactive power load at node-p the values of which would be zero in case of no load at any of the nodes, flat-start being the same voltage angle of zero degree at all nodes, and specified voltage magnitude at respective generation node and slack node and voltage magnitude of 1.0 pu at all load nodes, which is conventionally used as initial starting solution guess for classical Newton-Raphson Loadflow (NRL) and Super Super Decoupled Loadflow (SSDL) computation methods, and Vpo, and epo, fpo are the flat-start voltage magnitude and its real and imaginary components respectively at node-p, performing loadflow computation by solving said trained, tested and validated, formed, and stored ANNL computation model to calculate, complex voltages or their real and imaginary components or voltage magnitude and voltage angle at nodes of the power network providing for calculation of power flow through different components of the power network, and to calculate reactive power generations at PV-nodes and slack node, real power generation at the slack node and transformer tap-position indications, evaluating loadflow computation for any over loaded components of the power network and for under or over voltage at any of the nodes of the power network, correcting one or more controlled variables and repeating the performing loadflow computation, evaluating, and correcting steps until evaluating step finds no over loaded components and no under or over voltages in the power network, and affecting a change in power flow through components of the power network and voltage magnitudes and angles at the nodes of the power network by actually implementing the finally obtained values of controlled variables after evaluating step finds a good power system or stated alternatively the power network without any overloaded components and under or over voltages, which finally obtained controlled variables however are stored for acting upon fast in case a simulated event actually occurs.