In each measurement, the phases of the three signals can have two arrangements:\nphase_0 == +2/3 pi ==> phase_1 == +2/3 pi ==> phase_2\nor\nphase_0 == -2/3 pi ==> phase_1 == -2/3 pi ==> phase_2
\n\nNaively, without any domain knowledge, one would expect this to be distributed about 50-50. But in the training dataset, more than 75% belongs to the first case. Similar distribution in test data. Moreover, all measurements where signals have different target values belong to the first case. Am I just picking up noise or is there some explanation behind this?
\n\n(I was trying to publish a kernel but it keeps getting killed ...)
", "messages": [ { "id": "461357", "postDate": "01/25/2019 20:35:45", "content": "In each measurement, the phases of the three signals can have two arrangements:\nphase_0 == +2/3 pi ==> phase_1 == +2/3 pi ==> phase_2\nor\nphase_0 == -2/3 pi ==> phase_1 == -2/3 pi ==> phase_2
\n\nNaively, without any domain knowledge, one would expect this to be distributed about 50-50. But in the training dataset, more than 75% belongs to the first case. Similar distribution in test data. Moreover, all measurements where signals have different target values belong to the first case. Am I just picking up noise or is there some explanation behind this?
\n\n(I was trying to publish a kernel but it keeps getting killed ...)
", "rawMarkdown": "In each measurement, the phases of the three signals can have two arrangements:\nphase_0 == +2/3 pi ==> phase_1 == +2/3 pi ==> phase_2\nor\nphase_0 == -2/3 pi ==> phase_1 == -2/3 pi ==> phase_2\n\nNaively, without any domain knowledge, one would expect this to be distributed about 50-50. But in the training dataset, more than 75% belongs to the first case. Similar distribution in test data. Moreover, all measurements where signals have different target values belong to the first case. Am I just picking up noise or is there some explanation behind this?\n\n(I was trying to publish a kernel but it keeps getting killed ...)", "votes": null }, { "id": "461388", "postDate": "01/25/2019 22:23:00", "content": "I also stumbled upon this when attempting to synchronise the signals. None of the trios have exactly the same phase (which makes sense since they were measured at different points in time). The drastic phase shift that you mention in your second case might have to do with the (~20 I think it was) different locations where measurements were taken. Perhaps the 25% that belong to your second case come from a location (or locations) where the measurements began at a drastically different point in the cycle than the other 75%.
", "rawMarkdown": "I also stumbled upon this when attempting to synchronise the signals. None of the trios have exactly the same phase (which makes sense since they were measured at different points in time). The drastic phase shift that you mention in your second case might have to do with the (~20 I think it was) different locations where measurements were taken. Perhaps the 25% that belong to your second case come from a location (or locations) where the measurements began at a drastically different point in the cycle than the other 75%.", "votes": null }, { "id": "461557", "postDate": "01/26/2019 10:50:12", "content": "If my knowledge is correct, voltages in three-phase system should be described as follow: U1=Um*sin(ωt) U2=Um*sin(ωt-2/3π) and U3=Um*sin(ωt-4/3π), where Um is the maximum value of voltage. So if I understand you correctly and having regard to what I wrote above, then all of measurements of three phase voltages should be in second arrangement. Can you give an example of the first case?
", "rawMarkdown": "If my knowledge is correct, voltages in three-phase system should be described as follow: U1=Um*sin(ωt) U2=Um*sin(ωt-2/3π) and U3=Um*sin(ωt-4/3π), where Um is the maximum value of voltage. So if I understand you correctly and having regard to what I wrote above, then all of measurements of three phase voltages should be in second arrangement. Can you give an example of the first case?", "votes": null }, { "id": "461560", "postDate": "01/26/2019 11:22:34", "content": "You can compare the signals of measurement 0 and measurement 15.
", "rawMarkdown": "You can compare the signals of measurement 0 and measurement 15.", "votes": null }, { "id": "465317", "postDate": "02/02/2019 19:59:28", "content": "From my understanding as a power system engineer:
\n\nThe phase order is not random. Each electric utility will generally follow a chosen electrical sequence (or \"rotation\") for the three phases, either ABC or ACB.
\n\nPhysical arrangement of the phases is also not random. When looking at a three-phase distribution line physically, there is no way to see what phase is what--it just looks like three wires. For convenience of identifying phases on poles, the utility may standardize on some particular physical arrangement(s) unless voltage imbalance indicates some transposition is needed.
\n\nI can think of a couple reasons for the differences in the phase order ABC vs ACB in the data:\n- Data coming from multiple utilities that use different phasing conventions\n- Sensors are placed monitoring three phases of a line but without regard to the actual phase assignments
\n\nIf you wanted the phase order of all measurements the same, you could swap the phase assignments of two phases of the measurement sets that didn't match your desired phase order.
", "rawMarkdown": "From my understanding as a power system engineer:\n\nThe phase order is not random. Each electric utility will generally follow a chosen electrical sequence (or \"rotation\") for the three phases, either ABC or ACB.\n\nPhysical arrangement of the phases is also not random. When looking at a three-phase distribution line physically, there is no way to see what phase is what--it just looks like three wires. For convenience of identifying phases on poles, the utility may standardize on some particular physical arrangement(s) unless voltage imbalance indicates some transposition is needed.\n\nI can think of a couple reasons for the differences in the phase order ABC vs ACB in the data:\n- Data coming from multiple utilities that use different phasing conventions\n- Sensors are placed monitoring three phases of a line but without regard to the actual phase assignments\n\nIf you wanted the phase order of all measurements the same, you could swap the phase assignments of two phases of the measurement sets that didn't match your desired phase order.", "votes": null } ], "comments": [ { "id": 461388, "author_name": "cchadha", "author_url": "", "post_date": "01/25/2019 22:23:00", "content": "I also stumbled upon this when attempting to synchronise the signals. None of the trios have exactly the same phase (which makes sense since they were measured at different points in time). The drastic phase shift that you mention in your second case might have to do with the (~20 I think it was) different locations where measurements were taken. Perhaps the 25% that belong to your second case come from a location (or locations) where the measurements began at a drastically different point in the cycle than the other 75%.
", "votes": null, "replies": [] }, { "id": 461557, "author_name": "raf123", "author_url": "", "post_date": "01/26/2019 10:50:12", "content": "If my knowledge is correct, voltages in three-phase system should be described as follow: U1=Um*sin(ωt) U2=Um*sin(ωt-2/3π) and U3=Um*sin(ωt-4/3π), where Um is the maximum value of voltage. So if I understand you correctly and having regard to what I wrote above, then all of measurements of three phase voltages should be in second arrangement. Can you give an example of the first case?
", "votes": null, "replies": [ { "id": 461560, "author_name": "sangxia", "author_url": "", "post_date": "01/26/2019 11:22:34", "content": "You can compare the signals of measurement 0 and measurement 15.
", "votes": null, "replies": [] } ] }, { "id": 465317, "author_name": "pdb5627", "author_url": "", "post_date": "02/02/2019 19:59:28", "content": "From my understanding as a power system engineer:
\n\nThe phase order is not random. Each electric utility will generally follow a chosen electrical sequence (or \"rotation\") for the three phases, either ABC or ACB.
\n\nPhysical arrangement of the phases is also not random. When looking at a three-phase distribution line physically, there is no way to see what phase is what--it just looks like three wires. For convenience of identifying phases on poles, the utility may standardize on some particular physical arrangement(s) unless voltage imbalance indicates some transposition is needed.
\n\nI can think of a couple reasons for the differences in the phase order ABC vs ACB in the data:\n- Data coming from multiple utilities that use different phasing conventions\n- Sensors are placed monitoring three phases of a line but without regard to the actual phase assignments
\n\nIf you wanted the phase order of all measurements the same, you could swap the phase assignments of two phases of the measurement sets that didn't match your desired phase order.
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "461357": "In each measurement, the phases of the three signals can have two arrangements:\nphase_0 == +2/3 pi ==> phase_1 == +2/3 pi ==> phase_2\nor\nphase_0 == -2/3 pi ==> phase_1 == -2/3 pi ==> phase_2\n\nNaively, without any domain knowledge, one would expect this to be distributed about 50-50. But in the training dataset, more than 75% belongs to the first case. Similar distribution in test data. Moreover, all measurements where signals have different target values belong to the first case. Am I just picking up noise or is there some explanation behind this?\n\n(I was trying to publish a kernel but it keeps getting killed ...)", "461388": "I also stumbled upon this when attempting to synchronise the signals. None of the trios have exactly the same phase (which makes sense since they were measured at different points in time). The drastic phase shift that you mention in your second case might have to do with the (~20 I think it was) different locations where measurements were taken. Perhaps the 25% that belong to your second case come from a location (or locations) where the measurements began at a drastically different point in the cycle than the other 75%.", "461557": "If my knowledge is correct, voltages in three-phase system should be described as follow: U1=Um*sin(ωt) U2=Um*sin(ωt-2/3π) and U3=Um*sin(ωt-4/3π), where Um is the maximum value of voltage. So if I understand you correctly and having regard to what I wrote above, then all of measurements of three phase voltages should be in second arrangement. Can you give an example of the first case?", "461560": "You can compare the signals of measurement 0 and measurement 15.", "465317": "From my understanding as a power system engineer:\n\nThe phase order is not random. Each electric utility will generally follow a chosen electrical sequence (or \"rotation\") for the three phases, either ABC or ACB.\n\nPhysical arrangement of the phases is also not random. When looking at a three-phase distribution line physically, there is no way to see what phase is what--it just looks like three wires. For convenience of identifying phases on poles, the utility may standardize on some particular physical arrangement(s) unless voltage imbalance indicates some transposition is needed.\n\nI can think of a couple reasons for the differences in the phase order ABC vs ACB in the data:\n- Data coming from multiple utilities that use different phasing conventions\n- Sensors are placed monitoring three phases of a line but without regard to the actual phase assignments\n\nIf you wanted the phase order of all measurements the same, you could swap the phase assignments of two phases of the measurement sets that didn't match your desired phase order." }, "source": "meta" }