Energy storage battery multi-fault cloud collaborative diagnosis method and system

Abstract

The invention relates to an energy storage battery multi-fault cloud collaborative diagnosis method and system, and the method comprises the steps: collecting the real-time voltage data of an energy storage battery, and carrying out the enhancement of the real-time voltage data, and obtaining an enhanced voltage signal; extracting an abnormal voltage segment in the enhanced voltage signal; using the abnormal voltage fragment to train an MT-AGAN generator to obtain a trained MT-AGAN model; using the trained MT-AGAN model to generate a virtual full charge curve; training a long short-term memory network model by using a virtual full charge curve to obtain a fault diagnosis model; and using the energy storage battery fault diagnosis model to complete identification of the fault type of the target energy storage battery. According to the invention, by using the virtual data to train the long short-term memory network (LSTM) model, the generalization ability of the fault diagnosis model can be enhanced, so that the fault diagnosis model still has relatively high diagnosis accuracy and stability in the face of unknown fault types and in a new environment.

Description

Technical Field

[0001] This invention relates to the field of battery diagnostic technology, and in particular to a cloud-based collaborative diagnostic method and system for multiple faults in energy storage batteries. Background Technology

[0002] Driven by energy transition, the energy storage industry has developed rapidly, with installed capacity continuously expanding. The safety and operational reliability of battery systems have become core challenges for the healthy development of the industry. The high-density characteristics of energy storage systems mean that a single battery failure can easily trigger a chain reaction, leading to safety accidents or operational interruptions, causing significant losses and impacting power plant efficiency. Therefore, building a more intelligent and adaptable fault prevention and diagnosis system is a key support for ensuring the safe and reliable application of energy storage on a large scale.

[0003] The core challenge facing current energy storage battery fault diagnosis technology lies in its inability to effectively address critical issues in complex real-world operating environments. First, early fault signals are extremely weak and easily masked by noise, making it difficult for traditional methods to identify faults at their critical initiation stages. Second, energy storage batteries are typically not fully charged and discharged, which significantly differs from diagnostic models built on standard full-charge data in laboratories, causing these models to generally fail under fragmented and non-standard operating conditions. Furthermore, existing diagnostic algorithms lack robustness to factors such as battery aging, environmental temperature variations, and operating condition fluctuations, relying on preset thresholds or rules under fixed operating conditions, which easily leads to high false alarm rates. Finally, existing technologies themselves have significant limitations: some voltage evolution analysis methods are ill-suited to random and irregular charging and discharging scenarios; some fault location schemes based on theories such as Shannon entropy require complete discharge curve data and cannot handle the prevalent data fragmentation problem; while other schemes rely on high-precision sensors, resulting in high costs and a lack of effective data transmission compression mechanisms, making it difficult to support the economical and efficient deployment of large-scale energy storage systems. Summary of the Invention

[0004] To address the aforementioned problems, the purpose of this invention is to provide a cloud-based collaborative diagnosis method and system for multiple faults in energy storage batteries.

[0005] A cloud-based collaborative diagnosis method for multiple faults in energy storage batteries includes:

[0006] Step 1: Collect real-time voltage data of the energy storage battery, and enhance the real-time voltage data to obtain an enhanced voltage signal;

[0007] Step 2: Extract the abnormal voltage segments from the enhanced voltage signal;

[0008] Step 3: Use the abnormal voltage fragment to train the MT-AGAN generator to obtain the trained MT-AGAN model;

[0009] Step 4: Generate a virtual full-fill curve using the trained MT-AGAN model;

[0010] Step 5: Train the Long Short-Term Memory network model using a virtual full-fill curve to obtain the fault diagnosis model;

[0011] Step 6: Use the energy storage battery fault diagnosis model to identify the fault type of the target energy storage battery.

[0012] Preferably, step 1: acquiring real-time voltage data of the energy storage battery and enhancing the real-time voltage data to obtain an enhanced voltage signal includes:

[0013] Step 1.1: Use the edge device to collect real-time voltage data of the energy storage battery, and normalize the real-time voltage data to obtain a normalized voltage signal;

[0014] Step 1.2: The normalized voltage signal is enhanced using the Caputo fractional derivative method to obtain the enhanced voltage signal; the Caputo fractional derivative processing formula is as follows:

[0015]

[0016] Among them, V norm (k) is the normalized voltage value of the kth sampling point; α is the fractional order; Δt is the sampling interval; L is the memory length; Γ(·) is the gamma function; j is the time offset index, j=0 represents the current sampling point, j=L-1 represents the earliest historical point.

[0017] Preferably, step 2: extracting abnormal voltage segments from the enhanced voltage signal includes:

[0018] Step 2.1: Calculate the voltage entropy change index and expansion entropy trend index of the enhanced voltage signal within the set monitoring window;

[0019] In step 2.1, at each scale factor τ, the fractional probability distribution is used. Define multiscale fractional Boltzmann-Shannon entropy:

[0020]

[0021] Calculate the Boltzmann-Shannon entropy under a selected scale factor as an indicator of voltage entropy change:

[0022] ΔMFBS=|MFBS_entropy(t)-MFBS_entropy(t-ΔT)|

[0023] Where ΔT is the monitoring window, V diffThe signal is the enhanced voltage signal; N is the number of data points at the current scale τ. j is the data segment index; α is the fractional order; t is the current time.

[0024] Step 2.2: Extract abnormal voltage segments based on voltage entropy change index and expansion entropy trend index.

[0025] Preferably, in step 2.1, the enhanced voltage signal is processed using a hierarchical decomposition method to obtain intrinsic mode function components, and the spread entropy of each intrinsic mode function component is calculated based on symbolic dynamics:

[0026]

[0027] Among them, HDE entropy Let π be the spread entropy, Π be the symbol sequence space, p(π) be the probability of pattern π occurring, and IMF be the probability of occurrence. K It is the k-th eigenmode function component generated by variational mode decomposition;

[0028] The spread entropy is used to calculate the expansion entropy trend index:

[0029]

[0030] Among them, HDE slope It is an indicator of the expansion entropy trend.

[0031] Preferably, step 3: training the MT-AGAN generator using abnormal voltage segments to obtain the trained MT-AGAN model includes:

[0032] Using an abnormal voltage segment as input and the reconstructed full-fill curve as output, the MT-AGAN generator is trained to obtain the trained MT-AGAN model; wherein, during the training of the MT-AGAN model, the loss function is represented by Wasserstein distance:

[0033]

[0034] x and y represent the reconstructed full-fill curve and the actual full-fill curve, respectively, L Wass Let E be the representation of the Wasserstein distance. (x,y) For feature extraction of the distribution mean, Γ is the set of all possible joint distributions, and its marginal distributions are the distributions of the generated curve and the true curve, respectively; γ belongs to one of the joint distributions in the set of joint distributions Γ.

[0035] Preferably, in step 3, a maximum mean difference loss is introduced to reduce the distribution difference between the source domain and the target domain:

[0036]

[0037] Where Φ(·) represents the kernel function mapping feature; X s For source domain data; X t For target domain data; N s N represents the number of samples in the source domain. t The number of samples in the target domain.

[0038] Preferably, after step 5, the method further includes:

[0039] Global aging data is collected in the cloud, and the threshold at the edge is dynamically adjusted and sent to the edge to replace the local threshold. The edge uses the new threshold to recalculate the voltage entropy change index and expansion entropy trend index of the current monitoring window, and extracts the corresponding abnormal voltage segments. Steps 3-5 are repeated to retrain the long short-term memory network model to obtain the updated fault diagnosis model. The edge threshold update formula is as follows:

[0040]

[0041] In the formula, N cycle ΔMFBS represents the number of cell cycles. th This is the new threshold.

[0042] This invention also provides a cloud-based collaborative diagnostic system for multiple faults in energy storage batteries, comprising:

[0043] The voltage data acquisition module is used to acquire real-time voltage data of the energy storage battery and enhance the real-time voltage data to obtain an enhanced voltage signal.

[0044] An abnormal voltage extraction module is used to extract abnormal voltage segments from the enhanced voltage signal.

[0045] The generator training module is used to train the MT-AGAN generator using abnormal voltage fragments to obtain the trained MT-AGAN model.

[0046] The full-fill curve generation module is used to generate virtual full-fill curves using the trained MT-AGAN model.

[0047] The Long Short-Term Memory Network Training Module is used to train a Long Short-Term Memory Network model using a virtual full-fill curve to obtain a fault diagnosis model.

[0048] The fault diagnosis module is used to identify the fault type of the target energy storage battery using the energy storage battery fault diagnosis model.

[0049] According to specific embodiments provided by the present invention, the present invention discloses the following technical effects:

[0050] This invention relates to a cloud-based collaborative diagnosis method for multiple faults in energy storage batteries. Compared with existing technologies, this invention enhances the generalization ability of the fault diagnosis model by using virtual data to train a Long Short-Term Memory (LSTM) network model, enabling it to maintain high diagnostic accuracy and stability even when facing unknown fault types and new environments.

[0051] 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

[0052] 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.

[0053] Figure 1 This invention provides a flowchart of a cloud-based collaborative diagnosis method for multiple faults in energy storage batteries.

[0054] Figure 2 A schematic diagram of the architecture of the fragmented data reconstruction technology (MT-AGAN) provided by this invention;

[0055] Figure 3 A collaborative workflow diagram of the federated learning diagnostic architecture provided by this invention;

[0056] Figure 4 A schematic diagram illustrating the generator training principle provided by this invention. Detailed Implementation

[0057] 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.

[0058] 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.

[0059] 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.

[0060] Please see Figure 1-4 A cloud-based collaborative diagnosis method for multiple faults in energy storage batteries, comprising:

[0061] Step 1: Collect real-time voltage data of the energy storage battery, and enhance the real-time voltage data to obtain an enhanced voltage signal;

[0062] Further step 1 includes:

[0063] Step 1.1: Acquire voltage signals at the edge and normalize the voltage data to eliminate the influence of dimensions and adapt to subsequent feature amplification. Min-Max normalization is typically used.

[0064]

[0065] In the formula, V represents the real-time voltage data of the energy storage battery collected by the edge device; V min and V max These are the lowest and highest voltages within the evaluation period, respectively.

[0066] Step 1.2: Based on the Caputo fractional derivative method, the weak voltage drop characteristics at the initial stage of internal short circuit are amplified to improve fault sensitivity and facilitate subsequent abnormal state identification.

[0067] The normalized signal V is processed using Caputo fractional derivative (order α = 0.65, memory length L = 20). norm The discrete approximation formula is:

[0068]

[0069] In the formula, V norm(k) is the normalized voltage value of the kth sampling point; α is the fractional order (usually taken as 0.3 to 0.7); Δt is the sampling interval; L is the memory length (usually taken as 15-30, adjusted according to the sampling rate); Γ(·) is the gamma function: j is the time offset index (j=0 indicates the current sampling point, j=L-1 indicates the earliest historical point), technical effect: amplify the voltage drop of 0.5mV to 3.2mV (signal-to-noise ratio improved by 6.4 times).

[0070] Step 2: Extract the abnormal voltage segments from the enhanced voltage signal;

[0071] In step 2, the multi-scale fractional Boltzmann-Shannon entropy (MFBS_entropy) of the voltage signal and the hierarchical spread entropy (HDE_entropy) and trend (HDE_slope) of the expansion signal are calculated. When the safety threshold is exceeded, a battery fault warning command is triggered.

[0072] First, for the voltage signal after normalization and fractional derivative enhancement, calculate its multi-scale fractional Boltzmann-Shannon entropy:

[0073] -Use the scale factor set τ={τ1,τ2,…,τ M}(Usually, M = 10, τ min =1,τ max =20)

[0074] - At each scale factor τ, through the fractional probability distribution Define entropy:

[0075]

[0076] Where V diff The signal is a Caputo fractional derivative enhancement signal; N is the number of data points at the current scale τ. j is the index of the data segment (1≤j≤N).

[0077] - Calculate the voltage entropy change index at a selected characteristic scale (e.g., τ = 5):

[0078] ΔMFBS=|MFBS_entropy(t)-MFBS_entropy(t-ΔT)|

[0079] ΔT is the monitoring window (usually 5 minutes).

[0080] Then, for the real-time acquired expansion signal (such as battery deformation displacement), its hierarchical scattering entropy (HDE_entropy) is calculated:

[0081] - The intrinsic mode function components are obtained by hierarchical decomposition (EMD / VMD).

[0082] IMF1,…,IMF K

[0083] - Calculate the spread entropy of each IMF based on symbolic dynamics:

[0084]

[0085] In the formula, Π represents the symbol sequence space, p(π) represents the probability of pattern π appearing, and k is the total number of intrinsic mode function (IMF) components generated by variational mode decomposition (VMD).

[0086] Each IMF corresponds to physical characteristics at different frequency scales;

[0087] Low-frequency IMFs (such as IMF1) reflect battery aging trends;

[0088] High-frequency IMF (such as IMF_K) captures abrupt changes in mechanical expansion caused by internal short circuits;

[0089] Technical verification: Confirmed by Hilbert transform, 95% of the effective features can be separated when K=4 (K<4 loses details, K>5 introduces noise).

[0090] - Extract the expansion entropy trend indicator (HDE) slope ):

[0091]

[0092] The determination of whether a fault warning mechanism is triggered is based on voltage entropy change indicators and expansion entropy trend indicators. The typical judgment criteria are as follows:

[0093]

[0094] When the cloud issues a new threshold ΔMFBS th Then, the edge end immediately recalculates the ΔMFBS of the current window and determines whether the new threshold is met.

[0095] Step 3: Use the abnormal voltage fragment to train the MT-AGAN generator to obtain the trained MT-AGAN model;

[0096] Step 3 includes:

[0097] Step 3.1: Generator Training and Loss Function Design

[0098] Input: Tagged abnormal voltage segment

[0099] Output: Reconstructed full-fill curve (SOC 0%~100%)

[0100] The loss function uses the Wasserstein distance (Earth-Mover distance) to ensure that the generated curves are smooth and continuously distributed, thus avoiding mode collapse.

[0101]

[0102] In the formula, x and y represent the generated curve and the actual full-fill curve, respectively. The gradient penalty term is used... Constraining the discriminator D forces the generated curve gradient to be continuous (smoothness improved by 82%), L Wass Let E be the representation of the Wasserstein distance. (x,y) This is for extracting the distribution mean feature, corresponding to the expectation calculation in the MMD loss.

[0103]

[0104] Technical Summary: L Wass It's like a ruler for stabilizing vibrations -> flattening out sudden voltage changes; gradient penalty is like a bulldozer -> turning steep slopes into gentle slopes; E (x,y) The quality control director ensures the smoothness of every curve.

[0105] In one embodiment of the present invention: Input: a local voltage segment (duration ≤ 5 min) within the SOC range of 10%-70%.

[0106] 1. Causal Convolution

[0107]

[0108] Purpose: To ensure that the output at time t depends only on historical data (without future data leakage).

[0109] 2. Dilated Convolution

[0110] Expansion rate formula:

[0111] d=2 i (i is the number of layers)

[0112] Receptive field calculation:

[0113] R n =(k-1)*d n +R n-1

[0114] Example: When k=3, the receptive field of a 3-layer TCN reaches 15 steps (covering long-term dependencies).

[0115] 3. Residual Block

[0116] Formula: Output = ReLU(F(x) + x)

[0117] Function: To solve the vanishing gradient problem and accelerate the convergence of deep networks.

[0118] Output: High-dimensional temporal features

[0119] Z∈R T*D (T is the time step, D is the feature dimension)

[0120] Adversarial generation module

[0121] Input: TCN encoded feature Z

[0122] The adversarial mechanism is divided into: a generator (G) that maps Z to a complete curve. Discriminator (D): Joint input (fragments, Calculate the Wasserstein distance between (fragment, true curve) and (fragment).

[0123] Loss function:

[0124]

[0125] (Gradient penalty term enhances curve smoothness)

[0126] Step 3.2: Discriminator Domain Adaptation Mechanism

[0127] Introducing maximum mean difference (MMD) loss reduces the distribution difference between the source domain (historical data from laboratory batteries) and the target domain (generated curves), addressing the data distribution offset problem in energy storage scenarios and improving reconstruction generalization:

[0128]

[0129] Where Φ(·) represents the kernel function mapping feature; X s For source domain data; X t For target domain data; N s N represents the number of source domain samples (the total amount of historical full-data from the laboratory). s ≥500); N t The number of samples in the target domain (the number of generated curves, dynamically adjusted according to the power plant).

[0130] Step 4: Generate a virtual full-fill curve using the trained MT-AGAN model;

[0131] Step 5: Train the Long Short-Term Memory network model using a virtual full-fill curve to obtain the fault diagnosis model;

[0132] Step 5 includes:

[0133] Step 5.1: The edge device performs initial entropy screening based on the real-time acquired battery voltage / expansion signal and completes entropy value compression and uploading.

[0134] Step 5.2: For the marked abnormal segments, generate a "virtual full-fill curve" based on MT-AGAN from Step 4, and complete the abnormal classification.

[0135] Input the reconstruction curve into the pre-trained long short-term memory network model;

[0136] Model processing

[0137] An LSTM network (3 layers) extracts temporal features, and the output layer uses Softmax to calculate the probability vector: [P 内短路 ,P 不一致性 ,P 传感器故障 ].

[0138] Decision rule: If max(P) > 0.8, then determine the corresponding fault type (e.g., P). 内短路 =0.92>0.8).

[0139] P 内短路 =0.92>0.8).

[0140] Output the probability of the fault type.

[0141] To identify fault types: internal short circuit (voltage drop), battery pack inconsistency (entropy drift), sensor failure (noise spike).

[0142] Step 5.3: Collect global aging data (e.g., N) in the cloud. cycle >500 times), dynamically adjust the edge threshold, and send it to the edge to replace the local threshold (and the formula ΔMFBS) th Used to update the threshold of the original step 1), the entropy index (such as ΔMFBS and HDE) of the current monitoring window is recalculated at the edge using the new threshold. slope This triggers a secondary diagnosis, thereby adapting to the age-related increase in baseline entropy and maintaining diagnostic specificity.

[0143]

[0144] In the formula, N cycle This represents the number of battery cell cycles.

[0145] This solution achieves low-latency, high-precision diagnosis of multiple fault types in energy storage batteries through three technological innovations: edge computing feature compression, cloud-based generative reconstruction, and federated dynamic optimization. Its core value lies in:

[0146] ①Breaking the dependency on fragmented data and adapting to random charging and discharging scenarios;

[0147] ② Entropy-driven feature compression transmission breaks through transmission bottlenecks;

[0148] ③ An aging adaptive threshold optimization mechanism ensures reliability throughout the entire life cycle;

[0149] ④ The weak fault characteristics based on fractional derivative (α=0.65) are significantly enhanced, improving the early fault detection rate;

[0150] ⑤ The edge-cloud federated collaborative architecture, combined with encrypted communication, enables efficient and secure distributed diagnostics and optimization.

[0151] This invention also provides a cloud-based collaborative diagnostic system for multiple faults in energy storage batteries, comprising:

[0152] The voltage data acquisition module is used to acquire real-time voltage data of the energy storage battery and enhance the real-time voltage data to obtain an enhanced voltage signal.

[0153] An abnormal voltage extraction module is used to extract abnormal voltage segments from the enhanced voltage signal.

[0154] The generator training module is used to train the MT-AGAN generator using abnormal voltage fragments to obtain the trained MT-AGAN model.

[0155] The full-fill curve generation module is used to generate virtual full-fill curves using the trained MT-AGAN model.

[0156] The Long Short-Term Memory Network Training Module is used to train a Long Short-Term Memory Network model using a virtual full-fill curve to obtain a fault diagnosis model.

[0157] The fault diagnosis module is used to identify the fault type of the target energy storage battery using the energy storage battery fault diagnosis model.

[0158] Compared with the prior art, the beneficial effects of the cloud-based collaborative diagnosis system for multiple faults of energy storage batteries provided by the present invention are the same as those of the cloud-based collaborative diagnosis method for multiple faults of energy storage batteries described in the above technical solutions, and will not be repeated here.

[0159] 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 cloud-based collaborative diagnosis method for multiple faults in energy storage batteries. 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 cloud-based collaborative diagnosis method for multiple faults in energy storage batteries described in the above-described technical solution, and will not be repeated here.

[0160] 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 cloud-based collaborative diagnosis method for multiple faults in energy storage batteries, characterized in that, include: Step 1: Collect real-time voltage data of the energy storage battery, and enhance the real-time voltage data to obtain an enhanced voltage signal; Step 2: Extract the abnormal voltage segments from the enhanced voltage signal; Step 3: Use the abnormal voltage fragment to train the MT-AGAN generator to obtain the trained MT-AGAN model; Step 4: Generate a virtual full-fill curve using the trained MT-AGAN model; Step 5: Train the Long Short-Term Memory network model using a virtual full-fill curve to obtain the fault diagnosis model; Step 6: Use the energy storage battery fault diagnosis model to identify the fault type of the target energy storage battery.

2. The cloud-based collaborative diagnosis method for multiple faults in energy storage batteries according to claim 1, characterized in that, Step 1: Acquire real-time voltage data of the energy storage battery, and enhance the real-time voltage data to obtain an enhanced voltage signal, including: Step 1.1: Use the edge device to collect real-time voltage data of the energy storage battery, and normalize the real-time voltage data to obtain a normalized voltage signal; Step 1.2: The normalized voltage signal is enhanced using the Caputo fractional derivative method to obtain the enhanced voltage signal; the Caputo fractional derivative processing formula is as follows: Among them, V norm (k) is the normalized voltage value of the kth sampling point; α is the fractional order; Δt is the sampling interval; L is the memory length; Γ(·) is the gamma function; j is the time offset index, j=0 represents the current sampling point, j=L-1 represents the earliest historical point.

3. The cloud-based collaborative diagnosis method for multiple faults in energy storage batteries according to claim 2, characterized in that, Step 2: Extracting abnormal voltage segments from the enhanced voltage signal, including: Step 2.1: Calculate the voltage entropy change index and expansion entropy trend index of the enhanced voltage signal within the set monitoring window; In step 2.1, at each scale factor τ, the fractional probability distribution is used. Define multiscale fractional Boltzmann-Shannon entropy: Calculate the Boltzmann-Shannon entropy under a selected scale factor as an indicator of voltage entropy change: ΔMFBS=|MFBS_entropy(t)-MFBS_entropy(t-ΔT)| Where ΔT is the monitoring window, V diff The signal is the enhanced voltage signal; N is the number of data points at the current scale τ. j is the data segment index; α is the fractional order; t is the current time. Step 2.2: Extract abnormal voltage segments based on voltage entropy change index and expansion entropy trend index.

4. The cloud-based collaborative diagnosis method for multiple faults in energy storage batteries according to claim 3, characterized in that, In step 2.1, the enhanced voltage signal is processed using a hierarchical decomposition method to obtain the intrinsic mode function components, and the spread entropy of each intrinsic mode function component is calculated based on symbolic dynamics: Among them, HDE entropy Let π be the spread entropy, Π be the symbol sequence space, p(π) be the probability of pattern π occurring, and IMF be the probability of occurrence. K It is the k-th eigenmode function component generated by variational mode decomposition; The spread entropy is used to calculate the expansion entropy trend index: Among them, HDE slope It is an indicator of the expansion entropy trend.

5. The cloud-based collaborative diagnosis method for multiple faults in energy storage batteries according to claim 4, characterized in that, Step 3: Using abnormal voltage segments to train the MT-AGAN generator to obtain the trained MT-AGAN model, including: Using an abnormal voltage segment as input and the reconstructed full-fill curve as output, the MT-AGAN generator is trained to obtain the trained MT-AGAN model; wherein, during the training of the MT-AGAN model, the loss function is represented by Wasserstein distance: x and y represent the reconstructed full-fill curve and the actual full-fill curve, respectively, L Wass Let E be the representation of the Wasserstein distance. (x,y) For feature extraction of the distribution mean, Γ is the set of all possible joint distributions, and its marginal distributions are the distributions of the reconstructed curve and the true curve, respectively. γ belongs to a joint distribution in the set of joint distributions Γ.

6. The cloud-based collaborative diagnosis method for multiple faults in energy storage batteries according to claim 5, characterized in that, In step 3, the maximum mean difference loss is introduced to reduce the distribution difference between the source and target domains: Where Φ(·) represents the kernel function mapping feature; X s For source domain data; X t For target domain data; N s N represents the number of samples in the source domain. t The number of samples in the target domain.

7. The cloud-based collaborative diagnosis method for multiple faults in energy storage batteries according to claim 6, characterized in that, Following step 5, the following is also included: Global aging data is collected in the cloud, and the threshold at the edge is dynamically adjusted and sent to the edge to replace the local threshold. The edge uses the new threshold to recalculate the voltage entropy change index and expansion entropy trend index of the current monitoring window, and extracts the corresponding abnormal voltage segments. Steps 3-5 are repeated to retrain the long short-term memory network model to obtain the updated fault diagnosis model. The edge threshold update formula is as follows: In the formula, N cycle ΔMFBS represents the number of cell cycles. th This is the new threshold.

8. A cloud-based collaborative diagnostic system for multiple faults in energy storage batteries, characterized in that, include: The voltage data acquisition module is used to acquire real-time voltage data of the energy storage battery and enhance the real-time voltage data to obtain an enhanced voltage signal. An abnormal voltage extraction module is used to extract abnormal voltage segments from the enhanced voltage signal. The generator training module is used to train the MT-AGAN generator using abnormal voltage fragments to obtain the trained MT-AGAN model. The full-fill curve generation module is used to generate virtual full-fill curves using the trained MT-AGAN model. The Long Short-Term Memory Network Training Module is used to train a Long Short-Term Memory Network model using a virtual full-fill curve to obtain a fault diagnosis model. The fault diagnosis module is used to identify the fault type of the target energy storage battery using the energy storage battery fault diagnosis model.

9. 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 cloud-based collaborative diagnosis method for multiple faults in an energy storage battery as described in any one of claims 1-7.

10. A 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 cloud-based collaborative diagnosis method for multiple faults in energy storage batteries as described in any one of claims 1-7.

Images