Capacity measurement compensation method under influence of output ripples in formation and capacity grading process

Abstract

The invention discloses a capacity measurement compensation method under the influence of output ripples in formation and capacity grading processes. The method comprises the following steps: acquiring an output current sequence of a target channel in a capacity statistics window; identifying a ripple period boundary corresponding to the output current sequence, and dividing the capacity statistical window into a front-end incomplete period, a middle complete period section and a rear-end incomplete period according to the ripple period boundary; integrating the output current sequence in the middle complete period section to obtain a main body capacity value; constructing a front-end local ripple template and a rear-end local ripple template, performing integration on the front-end local ripple template corresponding to the front-end incomplete period to obtain a front-end compensation amount, and performing integration on the rear-end local ripple template corresponding to the rear-end incomplete period to obtain a rear-end compensation amount; and adding the main body capacity value, the front-end compensation amount and the rear-end compensation amount to obtain a final compensation capacity value. According to the invention, the capacity measurement precision under the condition that output ripples exist in the formation and capacity grading process is improved.

Description

Technical Field

[0001] This invention relates to the field of battery formation and capacity testing, and more specifically, to a capacity measurement compensation method for the influence of output ripple during the formation and capacity testing process. Background Technology

[0002] After manufacturing, lithium-ion battery cells typically undergo formation and capacity testing. Formation is used to gradually bring the cell to a stable operating state during controlled charging and discharging, while capacity testing is used to test and screen the cell's capacity and consistency. In formation and capacity testing equipment, capacity measurement is usually based on the integration of the output current over time. Therefore, the sampling accuracy of the output current, the sampling timing, and the method of determining the integration interval all directly affect the capacity measurement results.

[0003] In the prior art, Chinese patent CN102645636B, entitled "A Battery Capacity Detection Method," discloses a scheme for capacity detection based on parameters such as open-circuit voltage, state-of-charge curve, and DC internal resistance. Its focus is on shortening capacity testing time and reducing testing energy consumption by utilizing battery parameter models. This type of scheme primarily focuses on the capacity detection mechanism and detection efficiency. In the prior art, Chinese patent CN107450024A, entitled "Battery Charge / Discharge Tester Ripple Absorption Device and Method," discloses a ripple absorption scheme for the DC side of charge / discharge testing equipment. This scheme reduces the DC ripple content at the output of the testing equipment by setting up a ripple absorption circuit composed of passive components, thereby reducing the impact of ripple on the accuracy of test parameters. This type of scheme primarily suppresses output ripple from the hardware side.

[0004] However, in the actual capacity measurement process, it is not as simple as integrating the current over the entire measurement interval. When there is ripple at the output, if the middle of the capacity statistical interval contains many complete ripple cycles, the current fluctuations within these complete ripple cycles usually revolve around a certain average output level. From the perspective of capacity integration, the high and low fluctuations in the middle section have a relatively dispersed impact on the results over multiple complete cycles. Therefore, the complete ripple cycles in the middle are usually not the main source of capacity measurement deviation. Conversely, the start and end times of the capacity statistical interval are usually triggered by step control events and are not specifically aligned with the ripple cycle boundaries. This makes it easier for the beginning and end to form boundary sections that only capture a portion of the ripple waveform.

[0005] This boundary section contains only incomplete waveforms, and is also prone to superimposed factors such as asynchronous sampling phases, limited number of sampling points, and transients during step switching. Therefore, the integration results of the boundary section are more likely to fluctuate, which in turn amplifies the impact on the final capacity measurement results.

[0006] In other words, existing technologies primarily address two issues: one focuses on how to perform capacity testing based on battery parameters, while the other focuses on reducing output ripple from the hardware side. However, for the formation and capacity testing process, the impact of a complete ripple cycle in the middle is relatively minor, while the beginning and end boundary sections are more likely to become the main source of error in capacity measurement due to ripple truncation, asynchronous sampling, and step switching. Therefore, a specific solution is still lacking for this problem. Thus, how to perform more targeted capacity measurement processing on the beginning and end boundary sections of the capacity statistical interval has become a technical problem that needs to be solved. Summary of the Invention

[0007] The technical problem to be solved by the present invention is to provide a capacity measurement compensation method under the influence of output ripple during the formation and capacity testing process, so as to solve the problems mentioned in the background art.

[0008] To achieve the above objectives, the present invention adopts the following technical solution:

[0009] A method for compensating for capacity measurement under the influence of output ripple during the formation and capacity testing process, comprising:

[0010] Obtain the output current sequence of the target channel within the capacity statistics window;

[0011] Identify the ripple period boundary corresponding to the output current sequence, and divide the capacity statistics window into a front-end incomplete period, a middle complete period segment, and a rear-end incomplete period based on the ripple period boundary.

[0012] Integrate the output current sequence within the complete central period segment to obtain the main capacity value;

[0013] A front-end local ripple template and a back-end local ripple template are constructed based on the set of complete ripple cycles adjacent to the front-end incomplete cycle and the set of complete ripple cycles adjacent to the back-end incomplete cycle, respectively. The front-end local ripple template is integrated with respect to the front-end incomplete cycle to obtain the front-end compensation amount, and the back-end local ripple template is integrated with respect to the back-end incomplete cycle to obtain the back-end compensation amount.

[0014] The final compensation capacity value is obtained by adding the main capacity value, the front-end compensation amount, and the back-end compensation amount.

[0015] Preferably, the output current sequence includes current sample values ​​collected according to a uniform sampling period and sampling timestamps corresponding to each current sample value; the capacity statistics window is defined by the start time and end time of the capacity statistics window; integrating the output current sequence within the middle complete period segment includes performing integration calculations on each current sample value within the middle complete period segment according to the sampling timestamps.

[0016] Preferably, identifying the ripple period boundary corresponding to the output current sequence includes:

[0017] The output current sequence is input into the ripple period identification model;

[0018] The ripple period recognition model outputs boundary labels corresponding to each sampling timestamp;

[0019] The boundary labels determine the ripple period boundary, the front-end incomplete period, the middle complete period segment, and the rear-end incomplete period.

[0020] Preferably, the ripple period recognition model is a temporal segmentation neural network, which includes an input layer, a temporal convolutional layer, a bidirectional recurrent layer, and a sequence labeling layer;

[0021] The training process of the ripple period recognition model includes: acquiring an output current sequence with ripple period boundary labels as a training sample set; inputting the output current sequence in the training sample set into the time-series segmentation neural network; using the corresponding ripple period boundary labels as supervision labels; and performing parameter training on the time-series segmentation neural network based on the supervision labels to obtain the ripple period recognition model.

[0022] Preferably, before constructing the front-end local ripple template and the rear-end local ripple template, the method further includes:

[0023] Obtain the output voltage sequence and process instruction sequence corresponding to the target channel;

[0024] The output current sequence, the output voltage sequence, and the step instruction sequence are input into the stable segment identification model;

[0025] The unstable interval is output by the stable segment identification model;

[0026] Complete ripple cycles located within the unstable interval are removed from the set of complete ripple cycles adjacent to the front-end incomplete cycle and the set of complete ripple cycles adjacent to the rear-end incomplete cycle.

[0027] Preferably, the stable segment identification model is a temporal classification neural network, which includes a feature encoding layer, a temporal fusion layer, and an interval output layer;

[0028] The training process of the stable segment identification model includes: acquiring output current sequences, output voltage sequences, and process instruction sequences labeled with stable and unstable intervals as training sample sets; inputting the output current sequences, output voltage sequences, and process instruction sequences in the training sample sets into the time-series classification neural network; using the corresponding stable and unstable interval labels as supervision labels; and performing parameter training on the time-series classification neural network based on the supervision labels to obtain the stable segment identification model.

[0029] Preferably, the steps of constructing the front-end local ripple template and the back-end local ripple template include:

[0030] According to the order of time distance from the start time of the capacity statistics window from small to large, select the complete ripple cycles indicated by the template quantity configuration value in the template configuration table to form a front-end template cycle set.

[0031] Phase alignment is performed on each complete ripple cycle in the front-end template cycle set, and the current values ​​at the same phase positions are averaged to obtain the front-end local ripple template.

[0032] According to the order of time distance from the end time of the capacity statistics window from small to large, select the complete ripple cycles indicated by the template quantity configuration value in the template configuration table to form a backend template cycle set.

[0033] Phase alignment is performed on each complete ripple cycle in the back-end template period set, and the current values ​​at the same phase position are averaged to obtain the back-end local ripple template.

[0034] Preferably, when the number of complete ripple cycles is less than the template quantity configuration value in the template configuration table, the method further includes:

[0035] Obtain the output current sequence of the reference channel set that belongs to the same power supply unit channel group as the target channel and whose step identifier is consistent with the step identifier corresponding to the capacity statistics window;

[0036] Extract the synchronization ripple segment from the output current sequence of the reference channel set;

[0037] Construct a common ripple auxiliary template based on the synchronous ripple segment;

[0038] The common ripple auxiliary template is used to determine the front-end compensation amount and the back-end compensation amount.

[0039] Preferably, the same power supply unit channel group is a channel set consisting of the target channel and the reference channel set powered by the same power supply unit;

[0040] The step of constructing a common ripple auxiliary template based on the synchronous ripple segment includes:

[0041] Perform time alignment on each of the aforementioned synchronous ripple segments according to the sampling timestamp;

[0042] Normalization is performed on each synchronous ripple segment according to the average current of the complete ripple cycle in which each synchronous ripple segment is located.

[0043] Based on the channel weights in the auxiliary template configuration table, the normalized synchronization ripple segments are aggregated to obtain the common ripple auxiliary template.

[0044] Preferably, the step of using the common ripple auxiliary template to participate in determining the front-end compensation amount and the back-end compensation amount includes:

[0045] Query the channel mapping template library to obtain the mapping record corresponding to the channel identifier of the target channel and the step identifier corresponding to the capacity statistics window;

[0046] Amplitude mapping is performed on the common ripple auxiliary template according to the mapping record to obtain the target auxiliary template;

[0047] When the number of complete ripple cycles used to construct the front-end local ripple template is less than the template quantity configuration value in the template configuration table, the target auxiliary template is integrated corresponding to the front-end incomplete cycle to obtain the front-end compensation amount.

[0048] When the number of complete ripple cycles used to construct the back-end local ripple template is less than the template quantity configuration value in the template configuration table, the target auxiliary template is integrated corresponding to the back-end incomplete cycle to obtain the back-end compensation amount.

[0049] The advantage of this invention over existing technologies lies in that it does not uniformly process the output ripple throughout the entire capacity statistics interval, but instead focuses on the boundary sections at the beginning and end of the capacity statistics interval. For the complete ripple cycle contained in the middle of the capacity statistics interval, since the output current fluctuations corresponding to this part usually revolve around the average output level under the working step, and the influence of the high and low fluctuations within each cycle on the integration result is relatively dispersed when there are many complete cycles, the middle section still uses the measured output current sequence for direct integration. For the boundary sections at the beginning and end of the capacity statistics interval, this invention first identifies the ripple cycle boundaries, then distinguishes between the front-end incomplete cycle, the middle complete cycle section, and the rear-end incomplete cycle, and constructs the front-end local ripple template and the rear-end local ripple template using the complete ripple cycle sets adjacent to the front-end incomplete cycle and the complete ripple cycle sets adjacent to the rear-end incomplete cycle, respectively. Furthermore, it determines the front-end compensation amount and the rear-end compensation amount based on the position of the start time and end time of the capacity statistics window within the corresponding incomplete cycle. In this way, the present invention utilizes the technical characteristics that the output ripple is periodic and the boundary section error is mainly affected by the phase interception position, so that the capacity measurement processing is no longer evenly distributed throughout the entire measurement range, but is corrected for the first and last boundary sections that are more prone to deviation, thereby improving the stability and consistency of the capacity measurement results.

[0050] This invention further introduces a stable segment identification mechanism. Before constructing the front-end and back-end local ripple templates, it simultaneously acquires the output voltage sequence and step instruction sequence corresponding to the target channel, and uses the stable segment identification model to identify unstable intervals, removing complete ripple cycles located within unstable intervals from the candidate cycles for template construction. This design leverages the characteristic that step switching, setpoint changes, and control recovery during the capacity formation process can introduce overshoot, swayback, and waveform distortion in a short period of time. This ensures that the local ripple template is built on a complete ripple cycle with consistent steps and a stable output state, avoiding transient switching entering the template and thus preventing boundary compensation from being built on distorted waveforms. Therefore, this invention further solves the problem that although boundary sections require compensation, the template itself may be contaminated by unstable waveforms.

[0051] This invention further introduces a common ripple auxiliary template mechanism. When the number of complete ripple cycles meeting the conditions near the target channel is insufficient, the output current sequence of a reference channel set belonging to the same power supply unit channel group and with the same process step identifier as the target channel is obtained. Synchronous ripple segments are extracted from these sequences, and a common ripple auxiliary template is obtained through time alignment, average current normalization, and aggregation by channel weight. This template is then combined with the mapping record corresponding to the target channel to obtain the target auxiliary template. This design utilizes the structural characteristics of the device—multiple channels under the same power supply unit are affected by a common ripple source, and different channels share a common ripple pattern—to ensure that even when the target channel lacks sufficient complete ripple cycles near the boundary, a template basis for boundary compensation can still be obtained. This solves the problem of difficulty in continuing compensation under short measurement intervals, insufficient samples in the boundary neighborhood, or local abnormal interference.

[0052] Therefore, this invention divides the capacity statistical interval into a central stable integration section and a beginning and end boundary compensation section, and combines two technical means, stable section identification and common ripple auxiliary template, to make the capacity measurement processing more closely match the actual error distribution characteristics during the capacity division process. In particular, it addresses the beginning and end boundary sections, which are the main source of error, thereby improving the repeatability, channel consistency, and comparability between steps of the capacity measurement results under the influence of output ripple. Attached Figure Description

[0053] Figure 1 This is a schematic diagram of the overall capacity compensation method of the present invention;

[0054] Figure 2 This is a schematic diagram of the waveform division of the ripple period boundary and statistical window in this invention;

[0055] Figure 3 This is a flowchart of the reference channel auxiliary template construction and determination process of the present invention;

[0056] Figure 4 This is a physical system topology diagram of the present invention, which includes a power supply unit and a channel;

[0057] Figure 5 This is a schematic diagram illustrating the calculation of the final compensation capacity value of this invention. Detailed Implementation

[0058] The specific embodiments of the present invention will now be described with reference to the accompanying drawings.

[0059] The method of this invention is mainly applied to lithium-ion cell formation and capacity testing equipment, and can also be applied to other battery testing systems that need to perform capacity integration during charge and discharge testing.

[0060] For ease of understanding, the capacity statistics window mentioned in this invention refers to the start and end time interval of the integration corresponding to a single capacity measurement, which is defined by the start and end times of the capacity statistics window. The front-end incomplete period refers to the segment of the ripple period that falls within the capacity statistics window at the start time but does not form a complete period. The rear-end incomplete period refers to the segment of the ripple period that falls within the capacity statistics window at the end time but does not form a complete period. The middle complete period segment refers to the continuous segment located between the front-end and rear-end incomplete periods and composed of several complete ripple periods. The purpose of this definition is to narrow the location where capacity measurement errors are most likely to occur from the entire charging and discharging process to the vicinity of the window boundary, because in actual processes, the capacity statistics window often does not coincide with the boundary of the ripple period, and the area deviation due to truncation is most likely to occur near the boundary. Furthermore, when the start time of the capacity statistics window is exactly on the boundary of the ripple cycle, the front-end incomplete cycle is an empty segment; when the end time of the capacity statistics window is exactly on the boundary of the ripple cycle, the rear-end incomplete cycle is an empty segment; when there is no complete ripple cycle within the capacity statistics window, the middle complete cycle segment is an empty segment, and the main capacity value is recorded as 0, with subsequent capacity values ​​given by the boundary compensation part.

[0061] In one embodiment, such as Figure 4 As shown, the system includes a power supply unit, multiple test channels, a sampling circuit, a controller, and a memory. The power supply unit outputs charging or discharging current to each test channel, and each test channel is connected to a corresponding battery cell. The sampling circuit includes a current sampling module and a voltage sampling module. The current sampling module can use a shunt resistor with a differential amplifier or a Hall current sensor; the voltage sampling module can use an isolated sampling module or an analog-to-digital converter. The controller is used to acquire the output current sequence, output voltage sequence, and step instruction sequence of each test channel, and calls the program stored in the memory to execute the capacity measurement compensation method. The power supply unit can drive multiple channels simultaneously, therefore multiple channels may be affected by ripple from the same power supply unit at the same time. The common ripple auxiliary template described later in this invention is designed using this system structural feature.

[0062] In one embodiment, such as Figure 1As shown, the method first obtains the output current sequence of the target channel within the capacity statistics window. The output current sequence consists of current sample values ​​collected according to a uniform sampling period and sampling timestamps corresponding to each current sample value. The uniform sampling period can be selected according to the device switching frequency, ripple frequency, and control board processing capability. In the capacity assessment equipment, the uniform sampling period can be selected from 0.5ms to 10ms, and in most cases, it can be selected from 1ms to 2ms. If the sampling period is too large, ripple details will be lost; if the sampling period is too small, the processing burden will increase. The capacity statistics window usually corresponds to the effective capacity measurement stage within a certain process step, such as a certain measurement interval in the constant current discharge process step or a certain interval in the constant current charging process step. Its start and end times can be recorded by the process control system or issued by the host computer according to the process script. In order to ensure the traceability of the integration results, this embodiment requires that each current sample value be bound to a sampling timestamp, so that subsequent integration can be performed directly according to the time interval, rather than assuming that all intervals are completely equal.

[0063] This invention does not directly accumulate all current sample values ​​within the entire capacity statistics window in a single step, but instead first identifies the boundaries of the ripple period. The reason for this design is that the ripples themselves may not completely cancel each other out during long-term integration. When the start or end time of the capacity statistics window falls in the middle phase of a certain ripple period, the areas of the first and last incomplete periods are no longer simply divided into equal parts of the area of ​​the complete period. Without processing, the truncation error will be directly introduced into the capacity value.

[0064] In one embodiment, the ripple cycle boundary is obtained through a ripple cycle identification model. The controller inputs a segment of the output current sequence of the target channel near the capacity statistics window into the ripple cycle identification model. The model outputs a boundary label for each sampling timestamp, indicating whether the sampling point corresponds to a ripple cycle boundary. To avoid the model outputting multiple boundary labels that are too close to each other, a label merging process can be performed after obtaining the original labels, that is, multiple labels with a time interval less than the number of boundary tolerance samples are merged into one boundary point. The number of boundary tolerance samples can be selected from 2 to 10 sampling points, which can be determined according to the uniform sampling period and the device ripple frequency. After the boundary points output by the model are arranged in chronological order, a boundary sequence of continuous ripple cycles can be obtained. Subsequently, the controller determines the front-end incomplete cycle, the middle complete cycle segment, and the rear-end incomplete cycle based on the position of the start and end times of the capacity statistics window in the boundary sequence. Figure 2As shown, if the starting time of the capacity statistics window is within a certain ripple cycle, the segment between the starting time and the boundary of the next ripple cycle is defined as the front-end incomplete cycle; if the ending time of the capacity statistics window is within a certain ripple cycle, the segment between the boundary of the previous ripple cycle and the ending time is defined as the back-end incomplete cycle; the continuous ripple cycles completely surrounded by all the boundaries between the two constitute the middle complete cycle segment.

[0065] In one embodiment, the ripple period recognition model employs a temporal segmentation neural network. This temporal segmentation neural network includes an input layer, a temporal convolutional layer, a bidirectional recurrent layer, and a sequence labeling layer. The input layer receives output current segments of length L, where L can be selected as 256, 512, 1024, or 2048 sampling points. The temporal convolutional layer consists of 2 to 5 layers of one-dimensional convolutional units, with each convolutional kernel length selectable as 3, 5, or 7, and the number of channels selectable as 32, 64, or 128. Its function is to extract local waveform morphology features from the original current sequence, enabling the model to identify the rising edge, falling edge, peaks and valleys, and repetition of adjacent periods of the ripple. The bidirectional recurrent layer can employ bidirectional gated recurrent units or a bidirectional long short-term memory network, with the number of hidden units selectable as 32 to 128. It is used to understand the contextual information of multiple consecutive ripple periods from a temporal perspective, avoiding misjudging a single local spike as a boundary. The sequence labeling layer outputs a boundary probability value for each sampling point and generates boundary labels using Softmax or Sigmoid.

[0066] The training process of this model can be implemented as follows: First, output current sequences under different operating steps, different current levels, and different power supply unit load states are extracted from the historical operating data of the capacity-measuring equipment to form a training sample set. Each sequence in the training sample set needs to be pre-labeled with ripple cycle boundaries. The labeling can be done manually with the help of auxiliary software, or the initial boundaries can be obtained through rule-based methods and then corrected by technicians. To ensure consistent boundary definitions, this embodiment preferably uses the first stable sampling point after the ripple cycle rises from the trough as the boundary point. The label can be a binary label, where boundary points are marked as 1 and non-boundary points are marked as 0. The cross-entropy loss function is used during training, the optimizer can be Adam, and the learning rate can be... to The number of training rounds can be selected from 20 to 200. After training, the boundary recognition accuracy, boundary offset error, and false detection rate on the validation set are used as evaluation metrics, and the optimal parameter combination is selected as the final ripple period recognition model. The purpose of this model setting is to make boundary recognition no longer dependent on a single threshold or fixed frequency assumption, but to adapt to the situation of slight ripple period drift under different power supply unit operating conditions.

[0067] Once the complete periodic segment in the middle is determined, the controller integrates the output current sequence of that segment to obtain the main capacity value. To ensure integration accuracy, this embodiment uses an integration method based on sampling timestamps. Let the complete periodic segment in the middle contain current sample values ​​from the m-th sampling point to the n-th sampling point, and let the k-th current sample value be denoted as... The corresponding sampling timestamp is Then the main body capacity value It can be calculated using the trapezoidal integral method as follows:

[0068] ;

[0069] in, The unit is Ah. The unit is A. The unit is seconds (s). The advantage of this approach is that even if there are slight fluctuations in the actual sampling timestamps, the consistency of the capacity integral can be maintained without introducing additional errors due to the default fixed sampling interval.

[0070] The key to this invention lies not only in calculating the main capacity value, but also in determining the compensation amount for the front-end and back-end residual cycles separately. To ensure that the compensation is based on a ripple pattern consistent with the current operating condition, this invention introduces a front-end local ripple template and a back-end local ripple template. A local ripple template refers to a representative periodic waveform formed by uniformly processing a set of complete ripple cycles near the target channel that are consistent with the current step and have stable waveforms. "Local" here refers to two aspects: firstly, it means being close to the boundary of the capacity statistics window in time; secondly, it means that the step identifier is consistent with the step identifier corresponding to the capacity statistics window. The purpose of this design is to avoid directly using ripple cycles from other steps or far from the current time for compensation, thereby reducing template distortion caused by load changes and control mode changes.

[0071] To ensure the availability of the cycle for template construction, this invention acquires the output voltage sequence and step instruction sequence corresponding to the target channel before template construction, and uses a stable segment identification model to output the unstable interval. The step instruction sequence is the control command sequence recorded by the controller on the time axis, and may include at least the step identifier, set current value, set voltage value, and step switching time. The output voltage sequence and output current sequence are acquired synchronously to reflect the sway, overshoot, or short-term recovery process during step switching. The unstable interval refers to the time interval unsuitable for constructing a local ripple template, typically including a transition interval after step switching, an interval where a sudden load change causes a significant change in the current envelope, and an interval where voltage and current oscillate significantly simultaneously. This determination is introduced because although the target channel's ripple cycles after step switching are formally complete cycles, their waveforms are often affected by the control loop recovery process. If directly included in the template, temporary overshoots will be treated as normal ripples, thus amplifying the compensation error.

[0072] In one embodiment, the stable segment identification model employs a temporal classification neural network. This network includes a feature encoding layer, a temporal fusion layer, and an interval output layer. The feature encoding layer performs one-dimensional convolutional encoding on the output current sequence and the output voltage sequence, respectively, and vectorized encoding on the step instruction sequence. The step instruction sequence can be expanded according to the sampling time to form a step feature vector corresponding to each sampling point, which includes at least the step identifier, the set current value, and the set voltage value. The temporal fusion layer can employ a bidirectional gated recurrent unit or a lightweight Transformer structure to fuse current, voltage, and step instruction features on a unified time axis. The interval output layer outputs a probability value for each sampling point indicating whether it belongs to a stable or unstable interval, and obtains the start and end times of the unstable interval by merging consecutive labels.

[0073] The training process for this model can be performed as follows: First, a large number of output current sequences, output voltage sequences, and step command sequences containing step switching processes, steady-state step processes, and abnormal fluctuation processes are collected to establish a training sample set. Then, technicians label the stable and unstable intervals in each sample. The labeling criteria can be set according to the actual characteristics of the equipment. For example, an interval lasting 1 to 5 ripple cycles after the step switching command is issued is usually labeled as an unstable interval; if the change in the current envelope within a continuous interval exceeds 5% to 20% of the average current in that interval, and the duration reaches 2 to 4 ripple cycles, it can also be labeled as an unstable interval. The purpose of providing the range here is to facilitate different equipment to select an appropriate training caliber based on the ripple amplitude and control loop speed, rather than limiting the unstable interval to a single threshold. During training, the cross-entropy loss function or Focal Loss can be used, and the learning rate can be selected as [missing information]. to Once trained, the model can output the non-stationary interval for real-time data.

[0074] After obtaining the unstable interval, the complete ripple cycle within the unstable interval is removed from the front-end template candidate set and the back-end template candidate set. In a further embodiment, the template candidate set also requires that the step identifier be consistent with the step identifier corresponding to the capacity statistics window. This is because the control mode of the same power supply unit may be different under different steps, and directly constructing a template across steps will destroy the consistency of the cycle shape.

[0075] In one embodiment, the construction process of the front-end local ripple template and the back-end local ripple template is as follows. First, a template configuration table is established, which includes at least the template quantity configuration value, the template resampling point number, and the template smoothing method. The template quantity configuration value refers to the number of complete ripple cycles used to generate a local template, which is typically selected from 3 to 8, and more preferably from 4 to 6. If the quantity is too small, the template is easily affected by single-cycle anomalies; if the quantity is too large, it may introduce old cycles that are too far from the current boundary. The template resampling point number is used to unify waveforms with different actual cycle lengths to the same phase coordinate, and is typically selected from 64, 128, or 256 points. The template smoothing method can use moving average or Savitzky-Golay smoothing to reduce single-point sampling noise.

[0076] For the front-end local ripple template, within complete ripple cycles where the step identifier matches the step identifier corresponding to the capacity statistics window and is outside the unstable range, the number of complete ripple cycles indicated by the template quantity configuration value is selected in ascending order of time distance from the start time of the capacity statistics window, forming the front-end template cycle set. For the back-end local ripple template, within complete ripple cycles that also satisfy the condition of the step identifier matching the step identifier corresponding to the capacity statistics window and being outside the unstable range, the same number of complete ripple cycles are selected in ascending order of time distance from the end time of the capacity statistics window, forming the back-end template cycle set. Here, the time distance can be the absolute value of the difference between the center time of the complete ripple cycle and the start or end time of the capacity statistics window. The nearest neighbor principle is used because the ripple pattern within the same step may still drift slowly with temperature rise, channel load, and control state; complete cycles closer to the boundary better reflect the true ripple near the boundary.

[0077] After each complete ripple cycle is selected, phase resampling is performed according to the number of template resampling points in the template configuration table, mapping each complete cycle to a unified phase coordinate. Let the resampled _th_ cycle be... Each phase point is denoted as , The value of is between 0 and 1, representing the normalized phase position within a complete ripple cycle. Subsequently, the current values ​​at the same phase point for all selected complete cycles are averaged to form a local ripple template. In a further embodiment, weights can also be set according to the distance between the cycle and the boundary, with closer distances resulting in larger weights, thus obtaining a weighted average template. The front-end local ripple template is denoted as . The back-end local ripple template is denoted as .

[0078] like Figure 5 As shown, after obtaining the local ripple template, it is necessary to determine the front-end compensation amount and the back-end compensation amount, which are obtained by integrating the template in the corresponding front-end and back-end residual periods, respectively. Specifically, first, the position of the capacity statistics window's starting time within the front-end residual period is determined, denoted as the phase position. Next, determine the position of the capacity statistics window termination time within the back-end residual period, and record it as the phase position. The phase position can be determined by linear normalization of the boundary times. If the initial boundary time of the incomplete period at the front end is... The termination boundary time is The capacity statistics window starts at time . Then we have:

[0079] ;

[0080] If the starting boundary time of the incomplete backend cycle is The termination boundary time is The capacity statistics window ends at [time]. Then we have:

[0081] ;

[0082] in, and The values ​​range from 0 to 1. The front-end compensation amount represents the current area within the capacity statistics window during the front-end residual period, while the back-end compensation amount represents the current area within the capacity statistics window during the back-end residual period. If a discrete integration method is used, the front-end compensation amount can be obtained from the local ripple template at the front end. The integral is obtained by performing an integral over the template interval up to 1. The back-end compensation amount can be obtained by adjusting the back-end local ripple template from 0 to... The template interval is used to perform integration. This design means that the middle complete period segment only undertakes the true integration of the complete period, while the missing parts at the beginning and end are recovered by the template, thus making up for the truncated boundary area. The final compensation capacity value... It can be represented as the main capacity value Compensation amount at the front end Backend compensation amount The sum, that is:

[0083] ;

[0084] Under certain operating conditions, the number of complete ripple cycles meeting the requirements near the target channel may be insufficient. For example, the capacity statistics window may coincide with a process switch, the target channel may experience a short-term anomaly, or the available cycles near the front and rear boundaries may be largely occupied by unstable intervals. To address this, this invention designs a common ripple auxiliary template mechanism to ensure that compensation can still be completed when templates are insufficient.

[0085] In one embodiment, such as Figure 3 As shown, when the number of complete ripple cycles that are outside the unstable range and whose step identifiers match the step identifiers corresponding to the capacity statistics window is less than the template quantity configuration value in the template configuration table, the controller triggers the auxiliary template construction process. At this time, the output current sequence of the reference channel set, which belongs to the same power supply unit channel group as the target channel and whose step identifiers match the step identifiers corresponding to the capacity statistics window, is obtained. The so-called same power supply unit channel group refers to the channel set consisting of the target channel and the reference channel set powered by the same power supply unit. Because the ripple output by the same power supply unit has the same origin on multiple channels, the ripple on the reference channel can be used as an external reference for the ripple shape of the target channel. The number of channels in the reference channel set can be selected from 2 to 16, preferably selecting reference channels that currently have no fault identifiers, have complete sampling, and whose step identifiers match the step identifiers corresponding to the capacity statistics window.

[0086] When extracting synchronous ripple segments from the output current sequence of the reference channel set, it is preferable to select complete ripple cycles that overlap in time with the front or rear boundary of the target channel. Here, a synchronous ripple segment refers to a complete ripple cycle segment within the same time window, corresponding to the same working step state, and with a clearly defined boundary. Since the actual cycle length and phase start point may differ slightly between different reference channels, it is necessary to first perform time alignment on each synchronous ripple segment according to the sampling timestamp. Time alignment can be performed according to the starting boundary time of each complete ripple cycle, and then uniform resampling is performed to map them all to the same number of phase points.

[0087] To reduce amplitude differences caused by inconsistencies in the average current levels of different reference channels, this invention normalizes the ripple segments based on the average current of the complete ripple cycle in which each synchronous ripple segment exists. Let the... After resampling of each synchronous ripple segment, at the phase point The current value on The average current of this segment is Then the normalized fragment can be represented as:

[0088] ;

[0089] When the absolute value of the average current is less than the preset lower current limit, normalization can be achieved by subtracting the average value and then dividing by the rated test current to avoid numerical amplification. The preset lower current limit can be set according to the minimum effective test current of the equipment, typically ranging from 1% to 5% of the rated test current. The purpose of normalization is to extract the common ripple pattern of different reference channels and minimize the influence of the differences in the set current of each channel.

[0090] In a further embodiment, the controller performs aggregation processing on each normalized synchronization ripple segment according to the channel weights in the auxiliary template configuration table to obtain a common ripple auxiliary template. The auxiliary template configuration table can record the correlation weight of each reference channel to the target channel. The weights can be calculated offline from historical data or updated periodically during operation. The channel weights range from 0 to 1, and the sum of the weights of all reference channels is 1. Preferably, reference channels with higher correlation to the historical ripple of the target channel can be assigned larger weights. The aggregation method can be a weighted average to obtain the common ripple auxiliary template. The purpose of this is to utilize the characteristic of multiple channels sharing the power supply unit ripple, so that when the target channel sample is insufficient, other channels can provide a stable common configuration.

[0091] Since the target channel and the reference channel may differ in sampling link gain, wiring impedance, and single-channel load conditions, directly using a common ripple auxiliary template for target channel compensation will result in amplitude deviation. Therefore, this invention further introduces a channel mapping template library. The channel mapping template library stores mapping records corresponding to the target channel, indexed by the channel identifier and process step identifier. Mapping records may include at least a scaling factor. and bias coefficient It can also include segmented scaling factors or nonlinear mapping parameters. When used online, the controller queries the channel mapping template library to obtain the mapping record corresponding to the channel identifier and the step identifier corresponding to the capacity statistics window of the target channel. It then performs amplitude mapping on the common ripple auxiliary template according to this mapping record to obtain the target auxiliary template. When using a linear mapping, it can be performed as follows:

[0092] ;

[0093] proportionality coefficient The bias coefficient is typically between 0.7 and 1.3. Generally, the value can be taken within the range of -5% to 5% of the rated test current of the target channel. The mapping coefficient can be obtained by fitting historical complete cycle data. The method is to collect the correspondence between the local template and the common template of the target channel during a period with sufficient complete ripple cycle of the target channel, and then fit it using the least squares method. and And write it into the template library.

[0094] After the target auxiliary template is obtained, when the number of complete ripple cycles used to construct the front-end local ripple template is less than the template quantity configuration value, the controller determines the front-end compensation amount based on the position of the target auxiliary template and the capacity statistics window at the start time within the front-end incomplete cycle; when the number of complete ripple cycles used to construct the back-end local ripple template is less than the template quantity configuration value, the controller determines the back-end compensation amount based on the position of the target auxiliary template and the capacity statistics window at the end time within the back-end incomplete cycle. In a further embodiment, if the front-end or back-end does not reach the template quantity configuration value, but some complete ripple cycles of the target channel are still available, a small number of local templates can be weighted and fused with the target auxiliary template, with the fusion weight proportional to the number of available complete ripple cycles. This approach can take into account both the target channel's own characteristics and common ripple characteristics without deviating from the main method. To maintain simplicity, the corresponding compensation amount can also be directly calculated using the target auxiliary template in the main implementation.

[0095] To illustrate the execution process of this method more intuitively, a specific flow is given below:

[0096] After the start of the process step, the controller continuously acquires the output current and voltage sequences of the target channel and records the process step command sequence. Once the start and end times of the capacity statistics window are determined, the controller calls the ripple cycle identification model to identify the boundaries of the current sequence near the window, determining the front-end incomplete cycle, the middle complete cycle segment, and the rear-end incomplete cycle. Subsequently, the controller calls the stable segment identification model to output the unstable interval, and within the complete ripple cycle where the process step identifier matches the corresponding process step identifier of the capacity statistics window and is not in the unstable interval, it selects the nearest template cycle set for the front-end and rear-end respectively, constructing the front-end local ripple template and the rear-end local ripple template. Next, the controller performs trapezoidal integration on the middle complete cycle segment based on a unified timestamp to obtain the main capacity value, and then calculates the front-end compensation amount and the rear-end compensation amount based on the template and boundary phase position. The three are added together to obtain the final compensated capacity value. Figure 1 and Figure 5 As shown, the entire process retains the measured integral of the complete cycle in the middle and performs template compensation on the first and last boundaries, which are most prone to distortion. Therefore, it is suitable for online use in the process of breaking down and composting.

[0097] In another embodiment, the ripple period identification model and the stable segment identification model can be deployed in the device's local controller, or in a host computer or server that communicates with the device. Local deployment is suitable for scenarios with high real-time requirements; host computer deployment is suitable for scenarios that require centralized training and unified distribution of model parameters. When updating model parameters, offline training and online replacement can be used to avoid affecting the ongoing fractionation and capacity-building task.

[0098] In another embodiment, the template configuration table, auxiliary template configuration table, and channel mapping template library can be stored in the device's local database. The template configuration table can be categorized and stored according to step identifier, rated current level, and power supply unit type, thereby enabling different steps to use different template quantity configuration values ​​and resampling point numbers. The auxiliary template configuration table can be automatically updated based on the historical correlation between the target channel and the reference channel. The channel mapping template library can be generated during the device's factory calibration phase or automatically refreshed using historical complete cycle data after the device has operated for a certain period. The purpose of this design is to ensure that the present invention not only works during the initial installation phase but also maintains its compensation effect as the device ages and channel characteristics change.

[0099] In another embodiment, if the device sampling frequency is high and the power supply unit switching frequency is relatively stable, the ripple period boundary, in addition to model identification, can also be corrected using rule-based assistance. Rule-based correction can manifest as constraining the time interval between the model output boundary and adjacent boundaries within a set frequency band. The set frequency band range can be calculated based on the device switching frequency. For example, when the period corresponding to the device's main ripple frequency is 2ms to 10ms, the effective range of the adjacent boundary interval can be set to the engineering allowable range near this interval, thereby eliminating obviously unreasonable isolated boundaries. Using rule-based correction does not change the core idea of ​​this invention that relies on model-based boundary identification, but rather enhances engineering robustness.

[0100] In another embodiment, the identification results of unstable intervals can be used not only for template elimination but also to control the template update strategy. When the proportion of unstable intervals is too high within a certain time period, the controller can pause the writing of new channel mapping records and auxiliary template weights to avoid abnormal operating conditions contaminating the template library. Although this implementation is an enhancement at the system operation and maintenance level, it is still consistent with the template reliability control logic of the present invention.

[0101] This invention also allows for the selection of different parameter ranges in different types of formation and capacity-deployment devices. For devices with high output ripple frequency and a large number of channels, the uniform sampling period can be appropriately reduced and the number of template resampling points increased to obtain finer phase resolution. For devices with milder output ripple and limited processor computing power, the uniform sampling period can be appropriately increased and the number of template resampling points reduced to reduce computational load. It is generally not recommended that the number of templates be less than 3, as templates are easily affected by accidental disturbances when there are fewer than 3 complete periods; it is also not recommended to have a value greater than 8 for a long period, as too many periods will dilute locality. It is generally not recommended that the number of reference channels be less than 2, as a single reference channel cannot represent the common ripple; it should also not be too many, to avoid introducing channels with weak correlation to the target channel.

[0102] In summary, this invention fully realizes a capacity measurement compensation method under the influence of output ripple during capacity balancing by performing ripple period boundary identification, segmentation, integration of the middle complete period, local ripple template compensation, unstable interval elimination, and common ripple auxiliary template mapping on the output current sequence within the capacity statistics window. The above embodiments can be combined with each other, and all equivalent substitutions and conventional technical transformations made based on the concept of this invention can be performed within the scope of this invention.

Claims

1. A method for compensating for capacity measurement under the influence of output ripple during the formation and capacity setting process, characterized in that, include: Obtain the output current sequence of the target channel within the capacity statistics window; Identify the ripple period boundary corresponding to the output current sequence, and divide the capacity statistics window into a front-end incomplete period, a middle complete period segment, and a rear-end incomplete period based on the ripple period boundary; Integrate the output current sequence within the complete central period segment to obtain the main capacity value; A front-end local ripple template and a back-end local ripple template are constructed based on the set of complete ripple cycles adjacent to the front-end incomplete cycle and the set of complete ripple cycles adjacent to the back-end incomplete cycle, respectively. The front-end local ripple template is integrated with respect to the front-end incomplete cycle to obtain the front-end compensation amount, and the back-end local ripple template is integrated with respect to the back-end incomplete cycle to obtain the back-end compensation amount. The final compensation capacity value is obtained by adding the main capacity value, the front-end compensation amount, and the back-end compensation amount.

2. The capacity measurement compensation method according to claim 1, characterized in that, The output current sequence includes current sample values ​​collected according to a uniform sampling period and sampling timestamps corresponding to each current sample value; the capacity statistics window is defined by the start time and end time of the capacity statistics window; integrating the output current sequence within the middle complete period segment includes performing integration calculations on each current sample value within the middle complete period segment according to the sampling timestamps.

3. The capacity measurement compensation method according to claim 1 or 2, characterized in that, The step of identifying the ripple period boundary corresponding to the output current sequence includes: The output current sequence is input into the ripple period identification model; The ripple period recognition model outputs boundary labels corresponding to each sampling timestamp; The boundary labels determine the ripple period boundary, the front-end incomplete period, the middle complete period segment, and the rear-end incomplete period.

4. The capacity measurement compensation method according to claim 3, characterized in that, The ripple period recognition model is a temporal segmentation neural network, which includes an input layer, a temporal convolutional layer, a bidirectional recurrent layer, and a sequence labeling layer. The training process of the ripple period recognition model includes: acquiring an output current sequence with ripple period boundary labels as a training sample set; inputting the output current sequence in the training sample set into the time-series segmentation neural network; using the corresponding ripple period boundary labels as supervision labels; and performing parameter training on the time-series segmentation neural network based on the supervision labels to obtain the ripple period recognition model.

5. The capacity measurement compensation method according to claim 1, characterized in that, Before constructing the front-end local ripple template and the back-end local ripple template, the following is also included: Obtain the output voltage sequence and process instruction sequence corresponding to the target channel; The output current sequence, the output voltage sequence, and the step instruction sequence are input into the stable segment identification model; The unstable interval is output by the stable segment identification model; Complete ripple cycles located within the unstable interval are removed from the set of complete ripple cycles adjacent to the front-end incomplete cycle and the set of complete ripple cycles adjacent to the rear-end incomplete cycle.

6. The capacity measurement compensation method according to claim 5, characterized in that, The stable segment identification model is a temporal classification neural network, which includes a feature encoding layer, a temporal fusion layer, and an interval output layer. The training process of the stable segment identification model includes: acquiring the output current sequence, output voltage sequence, and step instruction sequence labeled with stable and unstable intervals as the training sample set; The output current sequence, output voltage sequence, and step instruction sequence from the training sample set are input into the time-series classification neural network; the corresponding stable interval labels and unstable interval labels are used as supervision labels; the parameters of the time-series classification neural network are trained based on the supervision labels to obtain the stable segment recognition model.

7. The capacity measurement compensation method according to claim 1, 5, or 6, characterized in that, The steps for constructing the front-end local ripple template and the back-end local ripple template include: According to the order of time distance from the start time of the capacity statistics window from small to large, select the complete ripple cycles indicated by the template quantity configuration value in the template configuration table to form a front-end template cycle set. Phase alignment is performed on each complete ripple cycle in the front-end template cycle set, and the current values ​​at the same phase positions are averaged to obtain the front-end local ripple template. According to the order of time distance from the end time of the capacity statistics window from small to large, select the complete ripple cycles indicated by the template quantity configuration value in the template configuration table to form a backend template cycle set. Phase alignment is performed on each complete ripple cycle in the back-end template period set, and the current values ​​at the same phase position are averaged to obtain the back-end local ripple template.

8. The capacity measurement compensation method according to claim 7, characterized in that, When the number of complete ripple cycles is less than the template quantity configuration value in the template configuration table, the following is also included: Obtain the output current sequence of the reference channel set that belongs to the same power supply unit channel group as the target channel and whose step identifier is consistent with the step identifier corresponding to the capacity statistics window; Extract the synchronization ripple segment from the output current sequence of the reference channel set; Construct a common ripple auxiliary template based on the synchronous ripple segment; The common ripple auxiliary template is used to determine the front-end compensation amount and the back-end compensation amount.

9. The capacity measurement compensation method according to claim 8, characterized in that, The same power supply unit channel group is a channel set consisting of the target channel and the reference channel set powered by the same power supply unit; The step of constructing a common ripple auxiliary template based on the synchronous ripple segment includes: Perform time alignment on each of the aforementioned synchronous ripple segments according to the sampling timestamp; Normalization is performed on each synchronous ripple segment according to the average current of the complete ripple cycle in which each synchronous ripple segment is located. Based on the channel weights in the auxiliary template configuration table, the normalized synchronization ripple segments are aggregated to obtain the common ripple auxiliary template.

10. The capacity measurement compensation method according to claim 8 or 9, characterized in that, The step of using the common ripple auxiliary template to participate in determining the front-end compensation amount and the back-end compensation amount includes: Query the channel mapping template library to obtain the mapping record corresponding to the channel identifier of the target channel and the step identifier corresponding to the capacity statistics window; Amplitude mapping is performed on the common ripple auxiliary template according to the mapping record to obtain the target auxiliary template; When the number of complete ripple cycles used to construct the front-end local ripple template is less than the template quantity configuration value in the template configuration table, the target auxiliary template is integrated corresponding to the front-end incomplete cycle to obtain the front-end compensation amount. When the number of complete ripple cycles used to construct the back-end local ripple template is less than the template quantity configuration value in the template configuration table, the target auxiliary template is integrated corresponding to the back-end incomplete cycle to obtain the back-end compensation amount.

Images