State Estimation modifications for WLS #2414

marcopau · 2024-10-07T13:08:24Z

Fixed some bugs to state estimation code, like:

power injection results
shunts handling
current magnitude measurements handling

50 hz developments

Heiner92 · 2024-10-13T18:38:36Z

Great initiative! I hope these changes will improve the results I'm getting in distribution system state estimation where current measurements are usually only available as magnitudes.

AnkurArohi · 2024-10-14T06:46:03Z

pandapower/build_bus.py

    root1 = ds_find(ar, bus1)
    root2 = ds_find(ar, bus2)
    if root1 == root2:
        return
-    if bus_is_pv[root2]:
+    if bus_is_active[root2] and ~bus_is_pv[root1]:      # if root2 is an active bus (load or gen) and root1 is not a PV bus, it will merge root1 into root2


Here we should actually also make sure that root2 is not external grid or reference bus as well

If root2 is an external grid it is still fine to merge root1 into root2. You want to keep the PV bus, and in this case root2 (which is the PV bus) would be kept.

AnkurArohi · 2024-10-14T06:47:11Z

pandapower/build_bus.py

    if net._fused_bb_switches.any():
        # Note: this might seem a little odd - first constructing a pp to ppc mapping without
        # fused busses and then update the entries. The alternative (to construct the final
        # mapping at once) would require to determine how to fuse busses and which busses
        # are not part of any opened bus-bus switches first. It turns out, that the latter takes
        # quite some time in the average usecase, where #busses >> #bus-bus switches.

+        # create mask arrays to distinguish different grid elements


Redundant code

AnkurArohi · 2024-10-14T06:49:10Z

pandapower/build_bus.py

                    # one pv bus. Get bus from pv_buses_in_set
-                    map_to = bus_lookup[pv_buses_in_set.pop()]
+                    ref_bus = pv_buses_in_set.pop()


not a very robust way since if pandapower decides to not have the first element as REF, this wont work and will be difficult to debug

Explain what "if pandapower decides to not have the first element as REF" means. Feel free also to provide an example where this does not work, so I can look at it.

if the pv_buses_in_set first element which you are popping is not REF bus because of breaking change from pandapower

In pv_buses_in_set you have the list of pv buses for the considered set. What I am popping is a PV bus. That bus is then taken as bus to be kept later when doing the merging among multiple buses.

AnkurArohi · 2024-10-14T06:51:46Z

pandapower/build_gen.py

@@ -222,11 +220,12 @@ def _build_pp_gen(net, ppc, f, t):

    # set bus values for generator buses
    ppc["bus"][gen_buses[ppc["bus"][gen_buses, BUS_TYPE] != REF], BUS_TYPE] = PV


why here all other buses are set to PV

Here there is no change. Gen buses are always modelled as PV nodes in pandapower, sgens as PQ.

yes i am aware of that, the question was why here are the PQ buses exclusively excluded

This function only deals with gen_buses, and it assigns the label "PV" to them. Other buses are handled somewhere else.

AnkurArohi · 2024-10-14T07:10:57Z

pandapower/pypower/dIbr_dV.py

+
+    vb = range(len(V))
+    diagV = sparse((V, (vb, vb)))
+    diagVnorm = sparse((V / abs(V), (vb, vb)))


Here there should be try except block because division with zero can occur due to abs(V), it will help enormously if throughindices one could also find the pandapower element (in dataframe) which leads to this exception , for example NaNor zero g_us_per_km attribute for lines, etc

A division by 0 when dividing by the voltage magnitude? The second part of the comment (looking at the pandapower elements that can bring a division by zero) is not something that has to be addressed in a function to calculate the Jacobian. There are other functions and routines in pandapower that do that.

yes we have this many times for networks which have problems in attribute and is very central as then the algorithm does not work. I am unaware of other functions that does that

I see, but then the problem in that case is with the network data and with its processing, not with this pull request or with this specific function, so I guess we should shift the discussion somewhere else.

AnkurArohi · 2024-10-14T07:11:22Z

pandapower/pypower/dIbr_dV.py

+    ib = range(len(If))
+    idxf = abs(If) == 0
+    idxt = abs(It) == 0
+    diagIfnorm = sparse((conj(If) / abs(If), (ib, ib)))


same here with the try and except block

There is a routine in the WLS to avoid using the current magnitude measurements at the first iteration, when currents are initialized as zero. In the other iterations it is quite unlikely to have branch currents exactly equal to 0. If that case happens, I believe we should have some more elaborated routine to deal with it, not just a try and except. It is something that requires some thinking and that would come in one of the next round of updates. For the moment, the priority was to provide a Jacobian for current magnitude measurements that works, at least in most of the cases.

AnkurArohi · 2024-10-14T07:13:06Z

pandapower/pypower/dIbr_dV.py

+    idxt = abs(It) == 0
+    diagIfnorm = sparse((conj(If) / abs(If), (ib, ib)))
+    diagItnorm = sparse((conj(It) / abs(It), (ib, ib)))
+    diagIfnorm[idxf,idxf] = 0


why this is equated to zero?, please comment

AnkurArohi · 2024-10-14T07:20:07Z

pandapower/estimation/algorithm/matrix_base.py

-        dita_dv = (np.angle(1e-5 * dit_dv + It.reshape((-1, 1))) - np.angle(It.reshape((-1, 1))))/1e-5
+        difm_dth, difm_dv, ditm_dth, ditm_dv = dIbr_dV_new(self.eppci['branch'], self.Yf, self.Yt, V)
+        difm_dth, difm_dv, ditm_dth, ditm_dv = map(lambda m: m.toarray(), (difm_dth, difm_dv, ditm_dth, ditm_dv))
+        difa_dth, difa_dv, dita_dth, dita_dv = 0*difm_dth, 0*difm_dv, 0*ditm_dth, 0*ditm_dv


why are all these values zero?

At the moment, we are not yet supporting current phase angles for state estimation. I will add a line of comment there.

AnkurArohi · 2024-10-14T07:20:40Z

pandapower/estimation/algorithm/base.py

@@ -83,6 +86,7 @@ def estimate(self, eppci: ExtendedPPCI, **kwargs):

        current_error, cur_it = 100., 0
        # invert covariance matrix
+        eppci.r_cov[eppci.r_cov<(10**(-6))] = 10**(-6)


This value of 10-6 should ideally be parameterizable

It is an option I am considering, but I am not yet sure if to put it as an optional parameter for the user or not (I see some pros and cons). In any case, this will be definitely improved in the next round of updates, for the moment it was urgent to avoid having 0 or close to 0 values in the covariance matrix.

This is central to the feasibility of the solution and to evaluate how good the estimate is and where are problems in the network, so one could change it and then simulate multiple usecases

marcopau and others added 6 commits October 7, 2024 11:35

Fix pwr injection results for merged buses in SE

8b19466

Added shunt results in SE

0bbd43d

Fix initialization bugs in SE

26ac3c7

Enhancements and logs to WLS estimator

abc4619

FIx bugs with current mag meas in SE

e4b6fbf

Merge pull request #7 from marcopau/50Hz_developments

d4342f0

50 hz developments

vogt31337 marked this pull request as draft October 7, 2024 13:11

AnkurArohi reviewed Oct 14, 2024

View reviewed changes

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

State Estimation modifications for WLS #2414

State Estimation modifications for WLS #2414

marcopau commented Oct 7, 2024

Heiner92 commented Oct 13, 2024

AnkurArohi Oct 14, 2024

marcopau Oct 14, 2024

AnkurArohi Oct 14, 2024

AnkurArohi Oct 14, 2024

marcopau Oct 14, 2024

AnkurArohi Oct 14, 2024

marcopau Oct 14, 2024

AnkurArohi Oct 14, 2024

marcopau Oct 14, 2024

AnkurArohi Oct 14, 2024

marcopau Oct 14, 2024

AnkurArohi Oct 14, 2024

marcopau Oct 14, 2024

AnkurArohi Oct 14, 2024 •

edited

Loading

marcopau Oct 14, 2024

AnkurArohi Oct 14, 2024

marcopau Oct 14, 2024 •

edited

Loading

AnkurArohi Oct 14, 2024

AnkurArohi Oct 14, 2024

marcopau Oct 14, 2024

AnkurArohi Oct 14, 2024

marcopau Oct 14, 2024

AnkurArohi Oct 14, 2024

		@@ -222,11 +220,12 @@ def _build_pp_gen(net, ppc, f, t):

		# set bus values for generator buses
		ppc["bus"][gen_buses[ppc["bus"][gen_buses, BUS_TYPE] != REF], BUS_TYPE] = PV

State Estimation modifications for WLS #2414

Are you sure you want to change the base?

State Estimation modifications for WLS #2414

Conversation

marcopau commented Oct 7, 2024

Heiner92 commented Oct 13, 2024

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

AnkurArohi Oct 14, 2024 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

marcopau Oct 14, 2024 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

AnkurArohi Oct 14, 2024 •

edited

Loading

marcopau Oct 14, 2024 •

edited

Loading