Energy storage lithium battery degradation curve generation and prediction method based on conditional diffusion model

Abstract

The present application relates to a method for generating and predicting degradation curves of energy storage lithium batteries based on a conditional diffusion model, comprising: uniformly and randomly sampling a real degradation curve and working condition from a set time step to a battery charge and discharge cycle sequence; adding Gaussian noise to the real degradation curve to generate a noisy curve, generating a condition mask with Bernoulli distribution, and constructing a masked working condition using the condition mask and the working condition; inputting the noisy curve and the masked working condition into a conditional 1D U-Net network to train a trained EMA network; using the trained EMA network to output the complete SOH degradation curve of the target battery, and calculating the remaining service life of the target battery based on the complete SOH degradation curve. The present application takes the battery SOH degradation curve as the generation target of the diffusion model, takes the working condition matrix as the condition, trains the noise prediction network through forward diffusion, generates the complete SOH degradation trajectory through backward step-by-step denoising, and greatly improves the stability of the life prediction of the energy storage lithium battery.

Description

Technical Field

[0001] This invention relates to the field of energy storage lithium battery health management technology, specifically to a method for generating and predicting energy storage lithium battery degradation curves based on a conditional diffusion model. Background Technology

[0002] Lithium-ion batteries are core components of new energy storage systems and power battery systems. Their State of Health (SOH) and Remaining Useful Life (RUL) directly determine the safety, stability, and service life of the energy storage system. During long-term charge-discharge cycles, batteries exhibit continuous and non-linear capacity decay characteristics. A complete SOH degradation trajectory curve is the core basis for accurately assessing battery aging, predicting failure risks, and formulating maintenance and replacement strategies. Therefore, accurately, completely, and reliably predicting lithium-ion battery SOH degradation trajectories and lifespan projections is a core research direction in the field of battery health management.

[0003] Current battery SOH prediction technologies are mainly divided into two categories: traditional machine learning deterministic prediction and novel generative model probabilistic prediction. Publicly available patents and academic literature have disclosed a variety of battery degradation prediction schemes, which can basically achieve basic assessment and short-term prediction of battery health status. However, there are still significant technical shortcomings in terms of degradation trajectory integrity, prediction uncertainty characterization, operating condition adaptability, and inference accuracy stability, making it difficult to meet the long-term, high-precision, and high-reliability health monitoring requirements of large-scale energy storage batteries.

[0004] Among existing mainstream technical solutions, deterministic deep learning prediction methods are typically represented by Chinese patent CN122345805A. This solution constructs a dedicated prediction network that includes a basic feature extraction module, a time-parameter interaction module, a local and global feature extraction module, a local attention fusion module, and a regressor. Relying on multi-dimensional operating parameters such as battery voltage, current, and temperature, it extracts spatiotemporal features and fuses local and global information, outputting the battery SOH prediction result through regression. It can be applied to online monitoring scenarios of lithium-ion battery health status. However, this type of deterministic regression prediction architecture has inherent defects. It can only output a single SOH value or a finite-step prediction sequence, and cannot generate a complete degradation trajectory curve for the entire battery life cycle. At the same time, the basic architecture of feedforward feature extraction and regression does not have the ability to model the randomness and multimodal uncertainty distribution of the battery degradation process, and cannot provide confidence intervals for degradation trajectories and multi-dimensional degradation paths. The prediction results are dimensional and have poor fault tolerance, making it difficult to adapt to the dynamic changes in battery degradation under complex operating conditions. Summary of the Invention

[0005] To address the aforementioned problems, the purpose of this invention is to provide a method for generating and predicting degradation curves of energy storage lithium batteries based on a conditional diffusion model.

[0006] A method for generating and predicting degradation curves of energy storage lithium batteries based on a conditional diffusion model includes:

[0007] Step 1: Obtain the complete charge-discharge cycle sequence for each battery cell; the charge-discharge cycle sequence includes the SOH measurement value and corresponding operating parameters for each cycle;

[0008] Step 2: Construct a working condition matrix based on the working condition parameters, and normalize the SOH sequence to obtain the SOH degradation curve to form a sample set;

[0009] Step 3: Uniformly and randomly sample the sample set at the set time step to obtain the real degradation curve and the corresponding operating conditions;

[0010] Step 4: Add Gaussian noise to the real degradation curve to generate a noisy curve, generate a condition mask using Bernoulli distribution, and use the condition mask and operating conditions to construct the masked operating conditions.

[0011] Step 5: Input the noisy curve and the masked operating conditions into the Condition 1D U-Net network to output the noise prediction value, continuously train the Condition 1D U-Net network, and update the weights of the shadow Condition 1D U-Net network in an exponential moving average manner to obtain the trained EMA network.

[0012] Step 6: Extract the operating condition matrix from the target battery's operating data and input it into the trained EMA network to output the complete SOH degradation curve of the target battery, and calculate the remaining service life of the target battery based on the complete SOH degradation curve.

[0013] Preferably, in step 4, based on diffusion theory, Gaussian noise is added to the true degradation curve using a cumulative coefficient to generate a noisy curve; wherein, the noisy curve is:

[0014]

[0015] in, Represents the cumulative coefficient. This represents the true degradation curve. Indicates a noisy curve. This represents Gaussian noise.

[0016] Preferably, the encoder residual block group of the conditional 1D U-Net network consists of multiple downsampled residual blocks, and the encoder residual blocks are injected only with time embeddings, not with operating condition embeddings.

[0017] Preferably, the decoder residual block group of the Conditional 1D U-Net network consists of multiple upsampled residual blocks, and the decoder residual blocks fuse operating conditions and time information in a gated manner.

[0018] Preferably, step 6: extracting the operating condition matrix from the target battery's operating data and inputting it into the trained EMA network to output the complete SOH degradation curve of the target battery, and calculating the remaining service life of the target battery based on the complete SOH degradation curve, includes:

[0019] Step 6.1: Extract the operating condition matrix from the target battery's operating data and convert it into a condition vector using a Transformer encoder;

[0020] Step 6.2: Sample the initial pure noise from the standard Gaussian distribution;

[0021] Step 6.3: Using the conditional vector and the initial pure noise as input, call the EMA network with an all-1 mask and an all-0 mask respectively to output the conditional noise prediction and the unconditional noise prediction;

[0022] Step 6.3: Fuse conditional noise prediction and unconditional noise prediction, and generate multiple candidate degradation curves by changing different initial pure noise;

[0023] Step 6.4: For each candidate curve, calculate the root mean square error between it and the actual measured cycle SOH value of the target battery, and select the candidate curve with the smallest root mean square error as the complete SOH degradation curve of the target battery.

[0024] Preferably, in step 6.3, the two predictions are fused using a classifier-free guided formula, and the candidate curve is calculated using the posterior distribution sampling formula; wherein, the classifier-free guided formula is:

[0025]

[0026] in, To guide the strength parameters, For conditional noise prediction, This is for unconditional noise prediction.

[0027] The present invention also provides an electronic device, including a bus, a transceiver, a memory, a processor, and a computer program stored in the memory and executable on the processor. The transceiver, the memory, and the processor are connected via the bus. The computer program, when executed by the processor, implements the steps in the above-described method for generating and predicting degradation curves of energy storage lithium batteries based on a conditional diffusion model.

[0028] The present invention also provides a computer-readable storage medium having a computer program stored thereon, characterized in that, when the computer program is executed by a processor, it implements the steps in the above-described method for generating and predicting degradation curves of energy storage lithium batteries based on a conditional diffusion model.

[0029] According to specific embodiments provided by the present invention, the present invention discloses the following technical effects:

[0030] This invention relates to a method for generating and predicting degradation curves of energy storage lithium batteries based on a conditional diffusion model. Compared with the prior art, this invention uses the battery SOH degradation curve as the generation target of the diffusion model, with the operating condition matrix as the condition, and generates a complete SOH degradation trajectory by forward diffusion training of a noise prediction network and reverse stepwise denoising. This significantly improves the stability of energy storage lithium battery lifetime prediction.

[0031] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0032] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0033] Figure 1 Flowchart of the method for generating and predicting degradation curves of energy storage lithium batteries based on the conditional diffusion model provided by the present invention;

[0034] Figure 2 This invention provides a framework diagram of an end-to-end conditional diffusion model. Detailed Implementation

[0035] In the description of this invention, it should be understood that the terms "center," "longitudinal," "lateral," "length," "width," "thickness," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," "outer," "clockwise," and "counterclockwise," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0036] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0037] In this invention, unless otherwise explicitly specified and limited, the terms "installation," "connection," "linking," and "fixing," etc., should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0038] Please see Figure 1 A method for generating and predicting degradation curves of energy storage lithium batteries based on a conditional diffusion model, including:

[0039] Step 1: Obtain the complete charge-discharge cycle sequence for each battery cell; the charge-discharge cycle sequence includes the SOH measurement value and corresponding operating parameters for each cycle;

[0040] Step 2: Construct a working condition matrix based on the working condition parameters, and normalize the SOH sequence to obtain the SOH degradation curve to form a sample set;

[0041] Step 3: Uniformly and randomly sample the sample set at the set time step to obtain the real degradation curve and the corresponding operating conditions;

[0042] Step 4: Add Gaussian noise to the real degradation curve to generate a noisy curve, generate a condition mask using Bernoulli distribution, and use the condition mask and operating conditions to construct the masked operating conditions.

[0043] In step 4, based on diffusion theory, Gaussian noise is added to the true degradation curve using a cumulative coefficient to generate a noisy curve; wherein, the noisy curve is:

[0044]

[0045] in, Represents the cumulative coefficient. This represents the true degradation curve. Indicates a noisy curve. This represents Gaussian noise.

[0046] Step 5: Input the noisy curve and the masked operating conditions into the Condition 1D U-Net network to output the noise prediction value, continuously train the Condition 1D U-Net network, and update the weights of the shadow Condition 1D U-Net network in an exponential moving average manner to obtain the trained EMA network.

[0047] The encoder residual block group of the Conditional 1D U-Net network consists of multiple downsampled residual blocks, and the encoder residual blocks are injected only with time embeddings, not with operating condition embeddings. The decoder residual block group of the Conditional 1D U-Net network consists of multiple upsampled residual blocks, and the decoder residual blocks fuse operating condition and time information in a gated manner.

[0048] Step 6: Extract the operating condition matrix from the target battery's operating data and input it into the trained EMA network to output the complete SOH degradation curve of the target battery, and calculate the remaining service life of the target battery based on the complete SOH degradation curve.

[0049] Furthermore, step 6 includes:

[0050] Step 6.1: Extract the operating condition matrix from the target battery's operating data and convert it into a condition vector using a Transformer encoder;

[0051] Step 6.2: Sample the initial pure noise from the standard Gaussian distribution;

[0052] Step 6.3: Using the conditional vector and the initial pure noise as input, call the EMA network with an all-1 mask and an all-0 mask respectively to output the conditional noise prediction and the unconditional noise prediction;

[0053] Step 6.3: Fuse conditional noise prediction and unconditional noise prediction, and generate multiple candidate degradation curves by changing different initial pure noise;

[0054] In step 6.3, the two predictions are fused using a classifier-free guided formula, and the candidate curve is calculated using the posterior distribution sampling formula; wherein, the classifier-free guided formula is:

[0055]

[0056] in, To guide the strength parameters, For conditional noise prediction, This is for unconditional noise prediction.

[0057] Step 6.4: For each candidate curve, calculate the root mean square error between it and the actual measured cycle SOH value of the target battery, and select the candidate curve with the smallest root mean square error as the complete SOH degradation curve of the target battery.

[0058] like Figure 2 As shown, it should be noted that this invention also proposes an end-to-end conditional diffusion model framework: the battery degradation curve is used as the generation target of the diffusion model, and the battery operating condition matrix (capacity matrix, which includes a multi-dimensional description of operating condition parameters such as charge / discharge rate, cut-off voltage, and temperature) is used as the conditional input; during the training phase, Gaussian noise is gradually added to the real degradation curve through a forward diffusion process, and a one-dimensional U-Net network with asymmetric encoding and decoding is trained to predict the added noise from the noisy curve; during the inference phase, starting from pure Gaussian noise and guided by operating conditions, a complete SOH degradation curve is generated by progressively denoising in reverse; furthermore, through multi-sample generation and early cyclic observation optimization, high-precision and high-stability degradation curve prediction and RUL estimation are output.

[0059] (1) Data preprocessing layer

[0060] Training data loading: Load the complete charge-discharge cycle sequence of each battery cell from the battery degradation experiment dataset, including the SOH measurement value and corresponding operating parameters for each cycle.

[0061] Operating condition matrix extraction: Extract multi-dimensional operating condition parameters (charge / discharge rate, cut-off voltage, temperature, SOC range, etc.) from the original operating condition data, and construct a fixed-dimensional operating condition matrix as the condition input.

[0062] Sequence normalization: Normalize the SOH sequence to The interval is used to adapt to the Gaussian noise assumption of the diffusion model.

[0063] (2) Conditional diffusion model layer

[0064] Forward diffusion module: scheduled according to preset linear variance ( ), towards the true degradation curve Gaussian noise is gradually added to generate noisy curves with different noise levels. This is used to train the network's noise prediction capabilities.

[0065] Condition 1 DU-Net network: Core noise prediction network, receiver noise degradation curve diffusion time step The operating condition matrix and condition mask are used to output an estimate of the added noise. Its internal structure employs an asymmetric encoding and decoding architecture.

[0066] Inverse denoising module: During the inference phase, it denoises pure Gaussian noise. Starting with the noise predicted by the trained network, we gradually denoise it through backsampling, ultimately generating a clean degradation curve. .

[0067] (3) Condition 1 DU-Net internal structure

[0068] Position encoder: Employing one-dimensional sinusoidal position encoding, a unique encoding vector is assigned to each cyclic position in the sequence, enhancing the network's ability to perceive cyclic position information in the degradation curve. Position encoding can be enabled or disabled via a configuration switch, supporting ablation experiments to verify its contribution.

[0069] Time embedding module: Employs sine-cosine time embedding to diffuse time steps The vector is mapped to a fixed-dimensional temporal embedding vector and injected into each layer of the network to identify the current diffusion stage.

[0070] Conditional Encoder (Transformer): The Transformer encoder compresses the multi-dimensional load case matrix into a uniform-dimensional condition vector, which serves as the load case constraint for generating the degradation curve. This condition vector may be randomly zeroed during the training phase (Bernoulli mask) to support classifier-free training.

[0071] Encoder residual block group: Composed of multiple downsampled residual blocks, it extracts degradation features at different resolutions step by step. Key design: The encoder residual blocks are injected only with temporal embedding (TEMB), not with condition embedding—this allows the encoder to focus on extracting multi-scale degradation morphological features related to the diffusion stage from signals with different noise levels, unaffected by specific operating conditions.

[0072] Decoder residual block group: Composed of multiple upsampled residual blocks, progressively restoring curve resolution. Key design: The decoder residual block adopts a gated conditional fusion strategy, injecting features jointly with conditional embedding (cemb) and temporal embedding (temb) in the form of cemb*x+temb. Conditional embedding uses multiplicative gating to control feature activation intensity, and temporal embedding uses additive bias to calibrate the diffusion stage.

[0073] Multi-resolution self-attention: Introducing a multi-head self-attention module at a specific resolution level (high-resolution layer) to capture long-range dependencies between distant cycles in the degradation curve.

[0074] (4) Reasoning and Optimal Selection Layer

[0075] Multi-sample generator: For the same operating conditions, starting from different initial states of random noise, it generates multiple candidate degradation curves (reps) to cover the possible multimodal distribution of the degradation process.

[0076] Early Cyclic Optimization Module: This module calculates the root mean square error (RMSE) between each candidate curve and the actual observation within an early cyclic window (e.g., the first 100 cycles), and selects the candidate curve with the smallest error as the final prediction. This strategy utilizes the limited amount of early charge / discharge data available in real-world scenarios.

[0077] Evaluation output module: Based on the selected optimal degradation curve, calculate and output the complete SOH prediction curve, RUL estimate and related error index.

[0078] Training phase:

[0079] 1. Data loading: Read batch SOH degradation curves and their corresponding operating condition matrices from the preprocessed training set.

[0080] 2. Time-step sampling: For each sample, from... Medium uniform random sampling diffusion time step .

[0081] 3. Conditional mask generation: using a Bernoulli distribution (probability) (Typical value is 0.2) Generate condition mask When the mask is 0, the operating conditions are completely discarded, and the model relies only on the noisy curve and time step to predict noise—the unconditional generation capability of the training model; when the mask is 1, the operating conditions are injected normally—the conditional generation capability of the training model.

[0082] 4. Forward noise addition: Based on diffusion theory, using the accumulation coefficient Towards the true degradation curve Add Gaussian noise Generate noisy curves Noise level varies It increases and grows.

[0083] The process of gradually adding noise to the real curve is as follows:

[0084]

[0085] in , Follow Increase monotonically decreasing (the proportion of noise gradually increases).

[0086] 5. Noise Prediction: Condition 1 DU-Net Receiver Noise-Containing Curve Time step and the operating conditions after masking Output noise prediction value .

[0087] 6. Loss Calculation and Parameter Update: Calculate the mean square error (MSE) between the predicted noise and the actual noise, and update the network weights through backpropagation; at the same time, update the shadow network weights using the exponential moving average (EMA) method for stable sampling during the inference phase.

[0088] 7. Iterative training: Repeat steps 2–6 until the preset number of training rounds is reached, and save the optimal model checkpoint.

[0089] Reasoning process description:

[0090] 8. Model Loading: Load the EMA network weights saved during the training phase. The EMA network provides a more stable denoising effect during the inference phase.

[0091] 9. Condition Extraction: Extract the operating condition matrix from the target battery's operating data. The vector is then converted into a conditional vector by the Transformer encoder.

[0092] 10. Noise Initialization: Sample initial pure noise from a standard Gaussian distribution. .

[0093] 11. Stepwise reverse noise reduction: From Decrease to 0, execute each step as follows:

[0094] 12. Call the EMA network with an all-1 mask (conditional mode) and an all-0 mask (unconditional mode) respectively to obtain conditional noise prediction. And unconditional noise prediction ;

[0095] 13. Fusing two predictions using a classifier-free guided (CFG) formula: ,in For guiding strength parameters;

[0096] 14. Using the posterior distribution sampling formula from calculate The process is as follows:

[0097]

[0098] in, For the posterior variance (when ), For random noise (when hour (It degenerates into a deterministic mean).

[0099] 15. Multiple Sample Generation: For the same operating conditions, starting from different initial values ​​of random noise, repeat steps 3-4 a total of reps to generate multiple candidate degradation curves.

[0100] 16. Early Cycle Selection: For each candidate curve, calculate its relationship with the actual measured early cycle SOH value of the target battery (as before). The strategy calculates the RMSE (Recovery Rate of Error) between each cycle and selects the candidate curve with the smallest RMSE as the final prediction result. This strategy simulates a real-world application scenario—after the battery has run for a certain number of cycles, the best-matching degradation trajectory can be selected from the generated distribution using existing early observation data. The process of selecting the optimal candidate curve is as follows:

[0101]

[0102] in, For the first The actual SOH observation values ​​for each cycle For the first The candidate curve in the th The predicted values ​​for each cycle, and the final predicted curve is... .

[0103] 17. Output: Output the selected complete SOH degradation curve and calculate RUL (the number of cycles before SOH first drops below the failure threshold) based on the curve.

[0104] Table 1. Explanation of symbols used in diffusion processes

[0105]

[0106] Table 2 Network and Condition Symbols

[0107]

[0108] Table 3 Evaluation Symbols

[0109]

[0110] Table 4 Key Technical Parameters

[0111]

[0112] Example:

[0113] Application Scenario: A group of lithium iron phosphate battery modules (referred to as "dataset A") operating in an energy storage power station has experienced a degradation of the State of Harm (SOH) of some modules to below 80% of their initial capacity after approximately 2500 charge-discharge cycles. The task is to predict the future degradation trajectory and estimate the RUL based on the historical operating data (operating parameters and measured SOH sequences) of each module.

[0114] Data Scale: Dataset A contains complete degradation curves of several individual battery cells. Each curve records the SOH measurements and corresponding operating parameters for each cycle from the initial state to the end of the life (SOH < 80%). The dataset is divided into a training set and a test set. The training set is used for model training, and the test set is used to evaluate prediction performance.

[0115] Implementation steps:

[0116] (1) Data preprocessing: Extract the SOH sequence and operating condition matrix (charge-discharge rate, cutoff voltage range, temperature range, etc.) of each degradation curve, normalize the SOH to the [-1,1] interval, and truncate or fill to a fixed sequence length. .

[0117] (2) Model training: Follow the training process described in 3.4, using parameters , , , Learning rate Batch size 45, train conditional diffusion model until convergence.

[0118] (3) Inference and prediction: For each target battery cell in the test set, extract its operating condition matrix and set... , Ten candidate degradation curves are generated according to the reasoning process in 3.4. The best curve is selected by the measured SOH of the previous 100 cycles, and the final predicted curve is output.

[0119] (4) Evaluation: The prediction curve was upsampled to 2560 cycles, and the SOH predicted RMSE and RUL estimated RMSE were calculated.

[0120] According to specific embodiments of the present invention, the following technical effects are disclosed:

[0121] 1. An asymmetric conditional fusion architecture for encoding and decoding improves the accuracy of conditional control. The encoder residual block injects only the diffusion time embedding (TEMB), allowing it to focus on extracting multi-scale morphological features related to the degradation trend from signals with different noise levels, unaffected by specific operating conditions. The decoder residual block fuses operating conditions and time information in a gated manner using cemb*x+temb, enabling the operating conditions to finely control the feature activation intensity through multiplicative gating, and the time embedding to calibrate the diffusion stage with additive bias. Compared to the symmetric conditional injection design of schemes such as DiffBatt, this invention better meets the differentiated information type requirements of encoding (feature extraction) and decoding (conditional control generation) in degradation curve generation tasks.

[0122] 2. The Bernoulli mask CFG training strategy enables flexible condition control. During the training phase, the working conditions are randomly discarded using a Bernoulli mask, allowing a single model to learn both conditional and unconditional distributions simultaneously. During the inference phase, the strength of conditional compliance is continuously adjusted using a single parameter w via the CFG formula, eliminating the need to train multiple models or switch network branches. Compared to CDUA's external feature selection + general conditional injection approach, this scheme offers finer-grained conditional control and is simpler to implement.

[0123] 3. Early Cyclic Observation Optimization Improves Prediction Stability and Accuracy. Utilizing a limited number of readily available early cyclic SOH measurements, the optimal prediction is selected from multiple candidate generation curves, calibrating the generated distribution with actual observations. Compared to single-sampling or simple averaging strategies like DiffBatt and CDUA, this approach translates the potential of the diffusion model's multi-sample generation into a practical improvement in prediction accuracy. Furthermore, the early cyclic data upon which the optimization relies is naturally available in real-world scenarios, demonstrating practical feasibility.

[0124] 4. The end-to-end unified conditional diffusion framework reduces system complexity. Compared to the VAE+DDPM two-stage hybrid architecture of VDGM, this invention adopts a pure conditional DDPM framework, which unifies the training and inference processes, eliminates the need for additional training of VAEs or other prior models, and features a simple architecture, stable training, and convenient deployment.

[0125] The present invention also provides an electronic device, including a bus, a transceiver, a memory, a processor, and a computer program stored in the memory and executable on the processor. The transceiver, the memory, and the processor are connected via the bus. The computer program, when executed by the processor, implements the steps in the above-described method for generating and predicting degradation curves of energy storage lithium batteries based on a conditional diffusion model. Compared with the prior art, the beneficial effects of the electronic device provided by the present invention are the same as those of the above-described method for generating and predicting degradation curves of energy storage lithium batteries based on a conditional diffusion model, and will not be elaborated upon here.

[0126] The present invention also provides a computer-readable storage medium having a computer program stored thereon, characterized in that, when the computer program is executed by a processor, it implements the steps in the above-described method for generating and predicting degradation curves of energy storage lithium batteries based on a conditional diffusion model. Compared with the prior art, the beneficial effects of the computer-readable storage medium provided by the present invention are the same as the beneficial effects of the above-described method for generating and predicting degradation curves of energy storage lithium batteries based on a conditional diffusion model, and will not be elaborated here.

[0127] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for generating and predicting degradation curves of energy storage lithium batteries based on a conditional diffusion model, characterized in that, include: Step 1: Obtain the complete charge-discharge cycle sequence for each battery cell; the charge-discharge cycle sequence includes the SOH measurement value and corresponding operating parameters for each cycle; Step 2: Construct a working condition matrix based on the working condition parameters, and normalize the SOH sequence to obtain the SOH degradation curve to form a sample set; Step 3: Uniformly and randomly sample the sample set at the set time step to obtain the real degradation curve and the corresponding operating conditions; Step 4: Add Gaussian noise to the real degradation curve to generate a noisy curve, generate a condition mask using Bernoulli distribution, and use the condition mask and operating conditions to construct the masked operating conditions. Step 5: Input the noisy curve and the masked operating conditions into the Condition 1D U-Net network to output the noise prediction value, continuously train the Condition 1D U-Net network, and update the weights of the shadow Condition 1D U-Net network in an exponential moving average manner to obtain the trained EMA network. Step 6: Extract the operating condition matrix from the target battery's operating data and input it into the trained EMA network to output the complete SOH degradation curve of the target battery, and calculate the remaining service life of the target battery based on the complete SOH degradation curve.

2. The method for generating and predicting degradation curves of energy storage lithium batteries based on the conditional diffusion model according to claim 1, characterized in that, In step 4, based on diffusion theory, Gaussian noise is added to the true degradation curve using a cumulative coefficient to generate a noisy curve; wherein, the noisy curve is: in, Represents the cumulative coefficient. This represents the true degradation curve. Indicates a noisy curve. This represents Gaussian noise.

3. The method for generating and predicting degradation curves of energy storage lithium batteries based on a conditional diffusion model according to claim 1, characterized in that, The encoder residual block group of the conditional 1D U-Net network consists of multiple downsampled residual blocks, and the encoder residual blocks are injected only with time embeddings, not with operating condition embeddings.

4. The method for generating and predicting degradation curves of energy storage lithium batteries based on the conditional diffusion model according to claim 1, characterized in that, The decoder residual block group of the conditional 1D U-Net network consists of multiple upsampled residual blocks, and the decoder residual blocks fuse operating conditions and time information in a gated manner.

5. The method for generating and predicting degradation curves of energy storage lithium batteries based on a conditional diffusion model according to claim 1, characterized in that, Step 6: Extract the operating condition matrix from the target battery's operating data and input it into the trained EMA network to output the complete SOH degradation curve of the target battery. Then, calculate the remaining service life of the target battery based on the complete SOH degradation curve, including: Step 6.1: Extract the operating condition matrix from the target battery's operating data and convert it into a condition vector using a Transformer encoder; Step 6.2: Sample the initial pure noise from the standard Gaussian distribution; Step 6.3: Using the conditional vector and the initial pure noise as input, call the EMA network with an all-1 mask and an all-0 mask respectively to output the conditional noise prediction and the unconditional noise prediction; Step 6.3: Fuse conditional noise prediction and unconditional noise prediction, and generate multiple candidate degradation curves by changing different initial pure noise; Step 6.4: For each candidate curve, calculate the root mean square error between it and the actual measured cycle SOH value of the target battery, and select the candidate curve with the smallest root mean square error as the complete SOH degradation curve of the target battery.

6. The method for generating and predicting degradation curves of energy storage lithium batteries based on the conditional diffusion model according to claim 5, characterized in that, In step 6.3, the two predictions are fused using a classifier-free guided formula, and the candidate curve is calculated using the posterior distribution sampling formula; wherein, the classifier-free guided formula is: in, To guide the strength parameters, For conditional noise prediction, This is for unconditional noise prediction.

7. An electronic device comprising a bus, a transceiver, a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the transceiver, the memory, and the processor are connected via the bus, characterized in that, When the computer program is executed by the processor, it implements the steps in the method for generating and predicting degradation curves of energy storage lithium batteries based on the conditional diffusion model as described in any one of claims 1-6.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps in the method for generating and predicting degradation curves of energy storage lithium batteries based on the conditional diffusion model as described in any one of claims 1-6.

Images