Method for estimating peak value of lithium battery

Abstract

The invention discloses a peak value estimation method of a lithium battery, which comprises the following steps of: 1, analyzing a gradient value and a peak value, taking a gradient value mean value of points contained in each type of curve on an SOC (State of Charge) curve as a limit initial value, carrying out limit value replacement iteration by taking 1 / 20 of addition and subtraction of the mean value as a step length, and calculating to obtain a high-precision limit value; step 2, establishing an SOC peak value identification model and performing training by using a Mask R-CNN method; step 3, establishing a gradient map and calculating a limit value, performing primary identification according to the SOC peak value identification model, extracting an SOC curve and applying the SOC curve to the gradient map to obtain an SOC gradient map, and then identifying the SOC gradient map by using the model to obtain an estimated peak value; according to the method, the peak value model and the SOC curve characteristics are fully considered, the peak value SOC is accurately estimated, and the practicability of the battery can be protected to the maximum extent, so that the energy efficiency of the lithium battery is improved, and the service life of the battery is prolonged.

Description

Technical Field

[0001] The present invention relates to the technical field of lithium batteries, and in particular to a peak value estimation method for lithium batteries. Background Art

[0002] PSOC (Peak SOC) is the peak SOC of a lithium battery. This value represents the maximum value of the battery capacity. Accurately obtaining this value is beneficial to battery charging, discharging, and battery management.

[0003] In recent years, the demand for lithium iron phosphate batteries in electronic products has greatly increased. Lithium batteries can meet the needs of electric vehicles. Such batteries play an important role in the fields of power and energy storage. The peak value of lithium batteries is dynamically adjusted and is less than the rated capacity of the battery. If the battery is charged and discharged at the rated capacity for a long time without real-time adjustment according to the actual working conditions, the efficiency of the battery cannot be effectively utilized, resulting in adverse effects such as false labeling, which ultimately leads to market deterioration, impact on the lithium battery industry and collapse of new energy. At this stage, domestic and foreign scholars have focused their attention on the prediction of SOC and the research of battery management systems, without analyzing and studying the maximum value of battery charge and discharge. Therefore, the present invention proposes a peak estimation method for lithium batteries to solve the problems existing in the prior art. Summary of the Invention

[0004] In response to the above problems, the purpose of the present invention is to propose a peak estimation method for lithium batteries. This peak estimation method for lithium batteries proposes the concept of peak SOC for the first time. After accurately estimating the PSOC threshold, it can maximize the protection of battery use, improve the efficiency of lithium batteries, and thus extend the battery life.

[0005] To achieve the purpose of the present invention, the present invention is implemented through the following technical solutions: A peak estimation method for a lithium battery comprises the following steps:

[0006] Step 1: Analyze the gradient value and peak value, use the mean gradient value of each type of curve on the SOC curve as the initial limit value, and iterate the limit value with a step size of 1 / 20 of the mean value to obtain a high-precision limit value;

[0007] Step 2: Establish a model to identify SOC peaks and train it using the Mask R-CNN method;

[0008] Step 3: Establish a gradient map and limit calculation, perform initial identification based on the SOC peak model, extract the SOC curve and apply it to the gradient map to obtain the SOC gradient map, and then use the model to identify the SOC gradient map to obtain the estimated peak value.

[0009] A further improvement is that in step 1, the peak value is calculated using a gradient formula, which is expressed as follows:

[0010]

[0011] Where S(j+1) and S(j-1) are the SOC values ​​at time j+1 and j-1 respectively, and ΔT is the time unit on the SOC curve.

[0012] A further improvement is that: when the limit value is replaced and iterated in step one, the gradient value average of all peak points contained in each type of peak is taken in turn as the peak threshold calculation ratio, and the condition for terminating the iteration is that when a certain average value is used as the threshold standard, the iteration is terminated when more than three peaks overlap.

[0013] A further improvement is that the SOC peak model identified in step 2 is expressed by the following formula:

[0014]

[0015] S is the SOC value of the battery, Indicates the differential value of the polarization voltage, V s is the polarization voltage of the battery in the equivalent circuit, R s is the polarization internal resistance of the battery, C s is the polarization capacitance of the battery, i is the charge and discharge current of the battery, V is the voltage at the battery terminal, V oc (S) is the open circuit voltage of the battery in the equivalent circuit, R i is the ohmic internal resistance of the battery; the SOC value of the battery is expressed by the following formula

[0016]

[0017] Among them, S(0) is the SOC value of the battery at the initial moment, η is the battery charge and discharge efficiency, C t is the rated capacity of the battery, C t It gradually becomes smaller as the battery ages, i is the battery charge and discharge current, d is the differential symbol, and τ represents the object of calculus time.

[0018] Further improvements are as follows: the samples for model training in step 2 are divided into three categories: high, medium, and low SOC, as well as high, low temperature, and strong impact SOC, totaling 9 combined samples. During training, ablation experiments are performed on epoch, learning rate, and learning impulse.

[0019] A further improvement is that in step 2, when using the Mask R-CNN method for training, the four corner positions relative to the SOC curve in each RoI block are selected, the precise value of each position is calculated using the double line difference, and then the results are summarized to complete the model training.

[0020] The further improvement lies in: the specific process of step three is to first read the gradient map, then perform deep learning, traverse the deep learning recognition result map, mark the corresponding points in the original gradient map as peak points according to the position of the front edge points in the deep learning recognition result map, traverse the gradient values ​​of all peak points of the same type of peak in the deep learning recognition result map, obtain the threshold of this type of peak according to the threshold calculation method, traverse the peak points in the gradient map, and finally output the recognition result.

[0021] A further improvement is that in step three, when identifying the model, the strategy of expanding the curve range of SOC and expanding the sample is adopted to perform identification adjustment.

[0022] The beneficial effects of the present invention are as follows: by fully considering the peak model and SOC curve characteristics and accurately estimating the peak SOC, the present invention can maximize the protection of the battery's practicality, thereby improving the energy efficiency of the lithium battery and extending the battery's service life. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 Flow chart of the method of the present invention.

[0024] Figure 2 This is a flow chart for the refined identification of SOC limits of the present invention.

[0025] Figure 3 This is the flowchart of the Mask R-CNN of the present invention.

[0026] Figure 4 This is the equivalent circuit diagram of the present invention. DETAILED DESCRIPTION

[0027] In order to deepen the understanding of the present invention, the present invention will be further described in detail below with reference to the examples. The examples are only used to explain the present invention and do not constitute a limitation on the scope of protection of the present invention.

[0028] according to Figures 1-4 As shown, this embodiment provides a peak estimation method for a lithium battery, comprising the following steps:

[0029] Step 1: Analyze the gradient value and peak value, use the mean gradient value of each type of curve on the SOC curve as the initial limit value, and iterate the limit value with a step size of 1 / 20 of the mean value to obtain a high-precision limit value;

[0030] The peak value is calculated using the gradient formula, which is expressed as follows:

[0031]

[0032] Where S(j+1) and S(j-1) are the SOC values ​​at time j+1 and j-1 respectively, and ΔT is the time unit on the SOC curve;

[0033] During the limit replacement iteration, the gradient value mean of all peak points contained in each type of peak is taken in turn as the peak threshold calculation ratio. The condition for the iteration to terminate is that when a certain mean value is used as the threshold standard, the iteration is terminated when more than three peaks overlap.

[0034] Step 2: Establish a model to identify SOC peaks and train it using the Mask R-CNN method;

[0035] The SOC peak identification model is expressed by the following formula

[0036]

[0037] S is the SOC value of the battery, Indicates the differential value of the polarization voltage, V s is the polarization voltage of the battery in the equivalent circuit, R s is the polarization internal resistance of the battery, C s is the polarization capacitance of the battery, i is the charge and discharge current of the battery, V is the voltage at the battery terminal, V oc (S) is the open circuit voltage of the battery in the equivalent circuit, R i Is the ohmic internal resistance of the battery. The equivalent circuit diagram is as attached in the manual. Figure 4 As shown; the SOC value of the battery is expressed by the following formula

[0038]

[0039] Among them, S(0) is the SOC value of the battery at the initial moment, η is the battery charge and discharge efficiency, C t is the rated capacity of the battery, C t It gradually becomes smaller as the battery ages, i is the battery charge and discharge current, d is the differential symbol, and τ represents the object of calculus time action;

[0040] The training samples are divided into three categories: high, medium, and low SOC, as well as high, low, and strong impact SOC, totaling nine combined samples. During training, ablation experiments were conducted on epochs, learning rates, and learning impulses.

[0041] When using the Mask R-CNN method for training, the four corner positions relative to the SOC curve in each RoI block are selected, the precise value of each position is calculated using the double-line difference, and then the results are summarized to complete the model training.

[0042] Mask R-CNN uses three outputs: the first is the class label, the second is the bounding-box offset, and the third is the target mask. Mask R-CNN improves RoIPool in Faster R-CNN to RoIAlign to eliminate the coarse quantization of the former and accurately align the extracted features.

[0043] The Mask R-CNN architecture enhances the Faster R-CNN framework by introducing an additional Mask branch that can perform pixel-level segmentation on the image, thereby combining object detection and instance segmentation into one model. Figure 3 As shown in the figure, it is the flowchart of Mask R-CNN. First, the input image is processed by the feature extraction network or the backbone network to obtain a series of significant features. Then, these features are passed to the RPN (region proposal network) for generating and preliminarily screening candidate regions. Next, the ROIAlign (region of interest matching) layer standardizes these candidate regions into regions of interest of fixed size to ensure spatial consistency. Finally, these processed regions of interest enter the Head layer of the network, where the fully connected layer is used for classification and bounding box regression, and the Mask branch is equipped with an FCN (convolutional neural network) to achieve pixel-level instance segmentation.

[0044] Step 3: Establish a gradient map and limit calculation, perform initial identification based on the SOC peak recognition model, extract the SOC curve and apply it to the gradient map to obtain the SOC gradient map, and then use the model to identify the SOC gradient map to obtain the estimated peak value;

[0045] The specific process is to first read the gradient map, then perform deep learning, traverse the deep learning recognition result map, mark the corresponding points in the original gradient map as peak points according to the position of the front edge points in the deep learning recognition result map, traverse the gradient values ​​of all peak points of the same type of peak in the deep learning recognition result map, obtain the threshold of this type of peak according to the threshold calculation method, traverse the peak points in the gradient map, and finally output the recognition result. The process of SOC limit refinement recognition is shown in the attached manual. Figure 2 As shown;

[0046] Since the limit values ​​identified by deep learning have problems such as missed detection, false detection, and unclear edges, the model recognition is adjusted by adopting a strategy of expanding the SOC curve range and expanding the sample.

[0047] Regarding the limit calculation in step three, refer to the DZSOC algorithm for lithium iron phosphate batteries, which has been authorized with the patent announcement number CN113049961B.

[0048] The basic principles, main features, and advantages of the present invention are shown and described above. Those skilled in the art should understand that the present invention is not limited to the foregoing embodiments. The foregoing embodiments and descriptions are merely illustrative of the principles of the present invention. Various changes and modifications may be made to the present invention without departing from the spirit and scope of the present invention. Such changes and modifications are intended to fall within the scope of the present invention. The scope of protection claimed in the present invention is defined by the appended claims and their equivalents.

Claims

1. A peak estimation method for a lithium battery, characterized in that: The following steps are involved: Step 1: Analyze the gradient value and peak value, use the mean gradient value of each type of curve on the SOC curve as the initial limit value, and iterate the limit value with a step size of 1 / 20 of the mean value to calculate the high-precision limit value; Step 2: Establish a model to identify SOC peaks and train it using the Mask R-CNN method; Step 3: Establish a gradient map and limit calculation, perform initial identification based on the SOC peak model, extract the SOC curve and apply it to the gradient map to obtain the SOC gradient map, and then use the model to identify the SOC gradient map to obtain the estimated peak value.

2. The method for estimating the peak value of a lithium battery according to claim 1, wherein: In step 1, the high-precision limit is calculated using a gradient formula, which is expressed as follows: Where S(j+1) and S(j-1) are the SOC values ​​at time j+1 and j-1 respectively, and ΔT is the time unit on the SOC curve.

3. The method for estimating the peak value of a lithium battery according to claim 1, wherein: During the iteration of the limit value replacement in step 1, the gradient value average of all peak points contained in each type of peak is taken in turn as the peak threshold calculation ratio. The condition for terminating the iteration is that when a certain average value is used as the threshold standard, the iteration is terminated when more than three peaks overlap.

4. The method for estimating the peak value of a lithium battery according to claim 1, wherein: The SOC peak model identified in step 2 is expressed by the following formula S is the SOC value of the battery, Indicates the differential value of the polarization voltage, V s is the polarization voltage of the battery in the equivalent circuit, R s is the polarization internal resistance of the battery, C s is the polarization capacitance of the battery, i is the charge and discharge current of the battery, V is the voltage at the battery terminal, V oc (S) is the open circuit voltage of the battery in the equivalent circuit, R i is the ohmic internal resistance of the battery; the SOC value of the battery is expressed by the following formula Among them, S(0) is the SOC value of the battery at the initial moment, η is the battery charge and discharge efficiency, C t is the rated capacity of the battery, C t It gradually becomes smaller as the battery ages, i is the battery charge and discharge current, d is the differential symbol, and τ represents the object of calculus time.

5. The method for estimating the peak value of a lithium battery according to claim 1, wherein: The samples for model training in step 2 are divided into three categories: high, medium, and low SOC, as well as high, low temperature, and strong impact SOC, totaling 9 combined samples. During training, ablation experiments are performed on epoch, learning rate, and learning impulse.

6. The method for estimating the peak value of a lithium battery according to claim 1, wherein: In the second step, when using the Mask R-CNN method for training, the four corner positions relative to the SOC curve in each RoI block are selected, the precise value of each position is calculated using the double line difference, and then the results are summarized to complete the model training.

7. The method for estimating the peak value of a lithium battery according to claim 1, wherein: The specific process of step three is to first read the gradient map, then perform deep learning, traverse the deep learning recognition result map, mark the corresponding points in the original gradient map as peak points according to the positions of the front edge points in the deep learning recognition result map, traverse the gradient values ​​of all peak points of the same type of peak in the deep learning recognition result map, obtain the threshold of the peak according to the threshold calculation method, traverse the peak points in the gradient map, and finally output the recognition result.

8. The method for estimating the peak value of a lithium battery according to claim 1, wherein: In the step 3, the model is identified and adjusted by adopting a strategy of expanding the SOC curve range and the sample.

Images