Monitoring self-discharge in operating battery cells
Monitoring self-discharge in operating battery cells (i.e., actively charging or discharging) is disclosed. Cell voltage data over time is used to detect cell-to-cell charge voltage imbalance in a battery while battery cells are at their peak charge voltage. Resistors across each cell are switched on or off according to the average charge voltage imbalance. Actual resistor on-times are used as input to train a neural network in real time. The neural network learns the resistor duty cycle needed by each battery cell to precisely balance its self-discharge losses. If the self-discharge changes, the neural network retrains itself as the new data arrives. The neural network may include a Long Short-Term Memory (LSTM) network for each cell in the battery, which can learn the relative self-discharge rate of the respective cell.
1 . A battery system, comprising:
a battery comprising a plurality of battery cells, each cell of the plurality of battery cells having a resistor across the respective cell; and
a computing system operably connected to the plurality of battery cells, the computing system comprising:
memory storing computer program instructions for monitoring self-discharge in the battery cells, and
at least one processor configured to execute the computer program instructions, wherein
the computer program instructions are configured to cause the at least one processor to:
detect cell voltages for the plurality of battery cells over time,
run the detected cell voltages through a neural network to detect cell-to-cell charge voltage imbalances between the plurality of battery cells while the plurality of battery cells are at peak charge voltages over a time window, and
control resistances across the plurality of battery cells to maintain voltage balance between the plurality of battery cells based on output from the neural network by turning respective resistors of the plurality of battery cells on and off based on a difference between a peak voltage of a respective cell and peak voltages of other battery cells of the plurality of battery cells.
2 . The battery system of claim 1 , wherein
the plurality of battery cells are lithium ion cells having self-discharge rates five orders of magnitude or more below normal operating currents of the plurality of battery cells, and
the computer program instructions are further configured to cause the at least one processor to monitor the self-discharge rates of the plurality of battery cells while the plurality of battery cells are actively charging or discharging, via the neural network.
3 . The battery system of claim 1 , wherein the plurality of battery cells are connected in series.
4 . The battery system of claim 1 , wherein the computer program instructions are further configured to cause the at least one processor to:
train the neural network for the plurality of battery cells using actual on times of the plurality of battery cells as input, wherein
the neural network is configured to learn a resistor duty cycle for each respective resistor across the plurality of battery cells to balance self-discharge losses of the respective battery cell.
5 . The battery system of claim 1 , wherein the plurality of battery cells comprise a plurality of series connected strings of battery cells that are connected in parallel to other strings of the series connected strings of battery cells.
6 . The battery system of claim 5 , wherein the computer program instructions comprise a respective neural network for each series connected string of battery cells in the plurality of battery cells.
7 . The battery system of claim 1 , wherein the neural network comprises a Long Short-Term Memory (LSTM) network for each battery cell of the plurality of battery cells, the LSTM network configured to learn a relative self-discharge rate of the respective battery cell.
8 . The battery system of claim 7 , wherein
the LSTM network for each battery cell comprises three LSTM neural network units, and
each of the three LSTM neural network units comprises an LSTM cell configured to remember values over a time interval, an input gate, an output gate, and a forget gate.
9 . The battery system of claim 8 , wherein
a first memory unit of the three LSTM neural network units receives voltages of the respective battery cell and is configured to learn a present deviation of the received voltages from an average battery cell voltage for the plurality of battery cells over the time interval and output the learned present deviation, and
the received voltages are provided from a telemetry stream of battery cell voltages from the plurality of battery cells in the battery system.
10 . The battery system of claim 9 , wherein a second memory unit of the three LSTM neural network units is configured to receive the present deviation of the received voltages from the average battery cell voltage of the plurality of battery cells over the time interval from the first memory unit, learn on and off times for the resistor for the respective battery cell to minimize the present deviation from the average battery cell voltage for the plurality of battery cells, and output the learned resistor on and off times.
11 . The battery system of claim 10 , wherein a third memory unit of the three LSTM neural network units is configured to receive the learned resistor on and off times and learn a long-term duty cycle to match a self-discharge for the plurality of battery cells.
12 . A computer-implemented method, comprising:
detecting cell voltages for a plurality of battery cells over time, by a computing system;
running the detected cell voltages through a neural network to detect cell-to-cell charge voltage imbalances between the plurality of battery cells while the plurality of battery cells are at peak charge voltages over a time window, by the computing system; and
controlling resistances across the plurality of battery cells to maintain voltage balance between the plurality of battery cells based on output from the neural network by turning respective resistors across the plurality of battery cells on and off based on a difference between a peak voltage of a respective cell and peak voltages of other battery cells of the plurality of battery cells, by the computing system.
13 . The computer-implemented method of claim 12 , wherein the plurality of battery cells are lithium ion cells having self-discharge rates five orders of magnitude or more below normal operating currents of the plurality of battery cells and the method further comprises:
monitoring, by the computing system, the self-discharge rates of the plurality of battery cells while the plurality of battery cells are actively charging or discharging, via the neural network.
14 . The computer-implemented method of claim 12 , wherein the plurality of battery cells are connected in series.
15 . The computer-implemented method of claim 12 , further comprising:
training the neural network for the plurality of battery cells using actual on times of the plurality of battery cells as input, by the computing system, wherein
the neural network is configured to learn a resistor duty cycle for each respective resistor across the plurality of battery cells to balance self-discharge losses of the respective battery cell.
16 . The computer-implemented method of claim 12 , wherein
the plurality of battery cells comprise a plurality of series connected strings of battery cells that are connected in parallel to other strings of the series connected strings of battery cells, and
the computing system comprises a respective neural network for each series connected string of battery cells in the plurality of battery cells.
17 . The computer-implemented method of claim 12 , wherein
the neural network comprises a Long Short-Term Memory (LSTM) network for each battery cell of the plurality of battery cells, the LSTM network configured to learn a relative self-discharge rate of the respective battery cell,
the LSTM network for each battery cell comprises three LSTM neural network units, and
each of the three LSTM neural network units comprises an LSTM cell configured to remember values over a time interval, an input gate, an output gate, and a forget gate.
18 . The computer-implemented method of claim 17 , wherein
a first memory unit of the three LSTM neural network units receives voltages of the respective battery cell and is configured to learn a present deviation of the received voltages from an average battery cell voltage for the plurality of battery cells over the time interval and output the learned present deviation, and
the received voltages are provided from a telemetry stream of battery cell voltages from the plurality of battery cells in the battery system.
19 . The computer-implemented method of claim 18 , wherein a second memory unit of the three LSTM neural network units is configured to receive the present deviation of the received voltages from the average battery cell voltage of the plurality of battery cells over the time interval from the first memory unit, learn on and off times for the resistor for the respective battery cell to minimize the present deviation from the average battery cell voltage for the plurality of battery cells, and output the learned resistor on and off times.
20 . The computer-implemented method of claim 19 , wherein a third memory unit of the three LSTM neural network units is configured to receive the learned resistor on and off times and learn a long-term duty cycle to match a self-discharge for the plurality of battery cells.
21 . A non-transitory computer-readable medium storing a computer program, wherein the computer program is configured to cause at least one processor to:
detect cell voltages for a plurality of battery cells over time;
run the detected cell voltages through a neural network comprising a Long Short-Term Memory (LSTM) network for each battery cell of the plurality of battery cells, the LSTM network configured to learn a relative self-discharge rate of the respective battery cell; and
control resistances across the plurality of battery cells to maintain voltage balance between the plurality of battery cells based on output from the neural network by turning respective resistors across the plurality of battery cells on and off based on a difference between a peak voltage of a respective cell and peak voltages of other battery cells of the plurality of battery cells.
22 . The non-transitory computer-readable medium of claim 21 , wherein the plurality of battery cells are lithium ion cells having self-discharge rates five orders of magnitude or more below normal operating currents of the plurality of battery cells and the computer program is further configured to cause the at least one processor to:
monitor the self-discharge rates of the plurality of battery cells while the plurality of battery cells are actively charging or discharging, via the neural network.
23 . The non-transitory computer-readable medium of claim 21 , wherein the computer program is further configured to cause the at least one processor to:
train the neural network for the plurality of battery cells using actual on times of the plurality of battery cells as input, wherein
the neural network is configured to learn a resistor duty cycle for each respective resistor across the plurality of battery cells to balance self-discharge losses of the respective battery cell.
24 . The non-transitory computer-readable medium of claim 21 , wherein
the LSTM network for each battery cell comprises three LSTM neural network units,
each of the three LSTM neural network units comprises an LSTM cell configured to remember values over a time interval, an input gate, an output gate, and a forget gate,
a first memory unit of the three LSTM neural network units receives voltages of the respective battery cell and is configured to learn a present deviation of the received voltages from an average battery cell voltage for the plurality of battery cells over the time interval and output the learned present deviation, and
the received voltages are provided from a telemetry stream of battery cell voltages from the plurality of battery cells in the battery system.
25 . The non-transitory computer-readable medium of claim 24 , wherein a second memory unit of the three LSTM neural network units is configured to receive the present deviation of the received voltages from the average battery cell voltage of the plurality of battery cells over the time interval from the first memory unit, learn on and off times for the resistor for the respective battery cell to minimize the present deviation from the average battery cell voltage for the plurality of battery cells, and output the learned resistor on and off times.
26 . The non-transitory computer-readable medium of claim 25 , wherein a third memory unit of the three LSTM neural network units is configured to receive the learned resistor on and off times and learn a long-term duty cycle to match a self-discharge for the plurality of battery cells.