Energy storage power station battery health status diagnosis method and system based on big data

Abstract

The present invention provides a method and system for diagnosing the health status of batteries in energy storage power stations based on big data, which relates to the technical field of energy storage batteries. The method and system include collecting multimodal operation data of batteries in energy storage power stations, homomorphically encrypting to obtain ciphertext data, performing analysis and calculation, extracting explicit and implicit correlation features, and obtaining fused time series data; based on the fused time series data, extracting quantum entanglement features of multiple paths, combining with a self-evolving quantum attention mechanism, dynamically adjusting the phase and amplitude of quantum bits, generating enhanced representations of diagnostic features, realizing entanglement storage through quantum circuits, and predicting health status sequences in combination with gated units; determining battery health prediction results, searching the diagnostic strategy combination space online by combining quantum state superposition and quantum parallelism, embedding a multi-order causal relationship tensor model, modeling the causal relationship strength between battery performance indicators, determining battery health degradation indicators, generating a causal diagnosis map, and obtaining battery health status diagnosis results.

Description

Technical Field

[0001] The present invention relates to the technical field of energy storage batteries, and in particular to a method and system for diagnosing the health status of batteries in energy storage power stations based on big data. Background Art

[0002] With the rapid development of new energy power generation technology, the proportion of renewable energy power generation in the power system continues to increase. However, new energy power generation is intermittent and volatile, which brings challenges to the stable operation of the power grid. As a key technology to smooth the fluctuations of new energy power generation and improve the flexibility of the power grid, energy storage power stations play an increasingly important role in the power system. Energy storage power stations usually use large-scale battery arrays to store and release electrical energy. The health status of batteries directly affects the reliability, safety and economy of energy storage power stations. During long-term use, batteries will experience performance degradation such as capacity decay and increased internal resistance, resulting in a decrease in the available capacity and efficiency of energy storage power stations, and even causing safety hazards.

[0003] Traditional battery health status diagnosis methods mainly include model-based methods and data-driven methods. Model-based methods estimate the battery state variables by establishing an equivalent circuit model or electrochemical model of the battery. They usually rely on accurate physical models and parameter identification, and are difficult to adapt to the complex nonlinear characteristics and long-term degradation behavior of the battery. Data-driven methods use machine learning algorithms to extract the characteristics of battery health degradation from a large amount of historical operating data and establish a mapping relationship between health status and characteristics. However, existing data-driven methods mainly focus on the health diagnosis of a single battery, lack consideration of the overall health status of the battery array, and are difficult to fully explore the correlation information and degradation patterns between different batteries.

[0004] In summary, there is an urgent need for a battery health status diagnosis method for energy storage power stations based on big data, which can make full use of the massive multi-source heterogeneous data generated during the operation of energy storage power stations, improve the accuracy, robustness and efficiency of battery health diagnosis, and provide strong support for the safe and stable operation of energy storage power stations. The present invention can solve the problems in the prior art. Summary of the invention

[0005] The embodiments of the present invention provide a method and system for diagnosing the health status of batteries in energy storage power stations based on big data, which can solve the problems in the prior art.

[0006] According to a first aspect of the embodiments of the present invention,

[0007] A method for diagnosing the health status of batteries in energy storage power stations based on big data is provided, including:

[0008] Collect multimodal operation data at the PACK and cell levels of the energy storage power station battery, homomorphically encrypt the multimodal operation data to obtain ciphertext data, analyze and calculate the ciphertext data based on the Transformer self-attention data fusion mechanism, adaptively extract explicit and implicit correlation features, and obtain fused time series data;

[0009] Based on the fused time series data, a battery health diagnosis model is constructed. Through the convolution operation of the quantum walk convolution kernel, the quantum entanglement features of multiple paths are extracted. Combined with the self-evolving quantum attention mechanism, the phase and amplitude of the quantum bit are dynamically adjusted, and the combination weights of different feature subspaces are explored to generate diagnostic feature enhanced representations. Based on the diagnostic feature enhanced representations, the entangled storage of long-term and short-term memory is realized through quantum circuits, and combined with gated units, the health state sequence is predicted.

[0010] Based on the predicted health status sequence, the battery health prediction result is determined, and a quantum reinforcement learning algorithm is constructed. By combining quantum state superposition and quantum parallelism, the diagnostic strategy combination space is searched online. In the loop of the quantum reinforcement learning algorithm, a multi-order causal relationship tensor model is embedded to model the causal relationship strength between battery performance indicators, determine the battery health degradation indicator, generate a causal diagnostic map, and obtain the battery health status diagnosis result.

[0011] In an optional embodiment,

[0012] Collect multimodal operation data of the battery pack and cell level of the energy storage power station, perform homomorphic encryption on the multimodal operation data to obtain ciphertext data, analyze and calculate the ciphertext data based on the self-attention data fusion mechanism of Transformer, adaptively extract explicit and implicit correlation features, and obtain fused time series data including:

[0013] Collect multimodal operation data at the PACK and cell levels of the energy storage power station battery, including voltage, current, temperature and state of charge. During the collection process, introduce edge computing devices based on the MQTT protocol to collect and preprocess the multimodal operation data in real time, and perform hash operations on the multimodal operation data through a pre-built blockchain data authentication mechanism, and store the hash value in the blockchain;

[0014] Perform fully homomorphic encryption based on the lattice cryptographic system on multimodal operation data, convert the original data into encrypted ciphertext data, introduce a noise mechanism in the encryption process, encrypt the original data one by one based on the preset public key, and divide the preset private key into multiple shares through the key management protocol based on secure multi-party computing, and distribute them to multiple energy storage power stations, where the public key and secret key are distributed and updated through secure multi-party computing;

[0015] The ciphertext data is input into the Transformer-based self-attention data fusion model, and the explicit and implicit correlation features between different modal data are calculated through the self-attention mechanism, wherein the self-attention data fusion model adopts an encoder-decoder architecture, the encoder introduces a position encoding mechanism to capture the position information of the ciphertext data, the encoder and the decoder adopt a multi-head attention mechanism to capture the correlation features between the ciphertext data from different subspaces, and the corresponding features are extracted from the ciphertext data and fused through a forward propagation algorithm based on fully homomorphic encryption to obtain fused time series ciphertext data;

[0016] Based on the homomorphic decryption algorithm corresponding to the fully homomorphic encryption algorithm, the fused time series ciphertext data is homomorphically decrypted, and the ciphertext data is decrypted one by one using the private key. Combined with the GPU-based parallel decryption framework, the ciphertext data is transferred to the GPU in batches, and the decryption algorithm is executed in parallel on the GPU to obtain the fused time series decrypted data.

[0017] In an optional embodiment,

[0018] Based on the fused time series data, a battery health diagnosis model is constructed. Through the convolution operation of the quantum walk convolution kernel, the quantum entanglement features of multiple paths are extracted. Combined with the self-evolving quantum attention mechanism, the phase and amplitude of the quantum bit are dynamically adjusted, and the combination weights of different feature subspaces are explored to generate diagnostic feature enhanced representation. Based on the diagnostic feature enhanced representation, the entangled storage of long-term and short-term memory is realized through quantum circuits. Combined with the gate control unit, the health state sequence is predicted, including:

[0019] Construct a quantum walk convolution kernel, map the fused time series data to the quantum bit space, each time step corresponds to a quantum bit, create a translation operator and a coin operator, generate a quantum walk convolution kernel, and perform a convolution operation in the quantum bit space through the quantum walk convolution kernel. The translation operator controls the movement of the quantum bit in space, and the coin operator controls the superposition state of the quantum bit. By adjusting the movement parameters corresponding to the translation operator and the superposition parameters corresponding to the coin operator, the step size and direction of the quantum walk are controlled, and the quantum entanglement characteristics are extracted;

[0020] The quantum entangled features are input into a self-evolving quantum attention module based on the attention mechanism. Through quantum gate operations, the phase and amplitude of the quantum bit are dynamically adjusted to explore the combined weights of different feature subspaces. For each quantum entangled feature, quantum gate operations are performed according to a set of preset parameters to obtain a primary feature representation. The parameters corresponding to the quantum gate operations are adjusted in combination with the self-evolution algorithm, and the diagnostic feature enhanced representation is generated according to the self-attention allocation.

[0021] Input the diagnostic feature enhancement representation into a pre-built quantum long short-term memory circuit, and perform entanglement storage and update on the timing information corresponding to the diagnostic feature enhancement representation. The quantum long short-term memory circuit is composed of an input gate, a forget gate, an output gate, and a state update gate. The input gate controls the new diagnostic feature enhancement representation to enter the memory unit, the forget gate controls the forgetting of historical information, the output gate controls the influence of the memory unit on the current state, and the state update gate updates the state of the memory unit;

[0022] The output of the quantum long short-term memory circuit is entangled with the diagnostic feature enhancement representation through a parameterized quantum circuit to obtain a predicted state. The predicted state is continuously generated by recursively calling the quantum long short-term memory circuit. Each predicted state is measured in combination with quantum measurement operations, and the predicted state quantum state is converted into a classical state to generate a health state prediction sequence.

[0023] In an optional embodiment,

[0024] The method further comprises:

[0025] The quantum walk convolution kernel extracts quantum entanglement features through multi-step quantum walks. The formula is as follows:

[0026] ;

[0027] Among them, U walk represents the quantum walk convolution kernel, l represents the quantum walk step number, L represents the total number of quantum walk steps, represents the translation operator, represents the movement parameter, represents the hidden operator, represents the superposition parameter, I represents the identity matrix, Represents a tensor product operation.

[0028] In an optional embodiment,

[0029] The output of the quantum long short-term memory circuit is entangled with the diagnostic feature enhancement representation through a parameterized quantum circuit to obtain the predicted states including:

[0030] The predicted state is formulated as follows:

[0031] ;

[0032] in, represents the predicted quantum state of the health state at the tth time step, p represents the quantum gate operation sequence, N p represents the total number of quantum gate operations, γ p,t represents the weight of the p-th quantum gate operation at the t-th time step, Vp represents the pth parameterized quantum gate operation at the tth time step, λ p,t represents the parameters controlling the p-th quantum gate operation at the t-th time step, represents the hidden state quantum state at the t-th time step, Represents the adjusted quantum entanglement feature representation at the t-th time step.

[0033] In an optional embodiment,

[0034] Based on the predicted health status sequence, the battery health prediction results are determined, and a quantum reinforcement learning algorithm is constructed. By combining quantum state superposition and quantum parallelism, the online search diagnostic strategy combination space includes:

[0035] Map the performance indicators of the battery under different health states to the probability amplitude of the quantum state, construct a quantum representation of the battery health state, add the quantum phase as a hidden variable to the quantum state, and capture the nonlinear dynamic characteristics of the battery health degradation process;

[0036] Based on quantum circuits, value iteration and strategy iteration algorithms are constructed. Strategy search is accelerated through quantum parallel computing. The optimal diagnostic strategy combination is located using quantum amplitude amplification. The diagnostic strategy combination space is sampled through quantum gate sequences.

[0037] Based on the diagnostic strategy combination space, a search tree of diagnostic strategy combinations is dynamically generated, branches of the search tree are pruned based on heuristic knowledge, and adaptive step length exploration and utilization of the diagnostic strategy combination space is performed based on a quantum walk algorithm.

[0038] In an optional embodiment,

[0039] In the loop of the quantum reinforcement learning algorithm, a multi-order causal relationship tensor model is embedded to model the causal relationship strength between battery performance indicators, determine the battery health degradation indicators, generate a causal diagnosis map, and obtain the battery health status diagnosis results including:

[0040] Perform high-order singular value decomposition on the battery performance indicators to generate multi-order tensor causal relationships between battery performance indicators. Based on the causal connection strength between the performance indicators, determine the causal strength tensor. Based on the causal discovery algorithm, set the loss function and combine the causal intervention regularization term to construct the causal graph structure between the performance indicators, generate a multi-order causal relationship tensor model, and embed the multi-order causal relationship tensor model in the loop of the quantum reinforcement learning algorithm.

[0041] A quantum-classical hybrid optimization algorithm is used to search for battery health degradation indicators online, and a data-driven quantum causal enhancement method is used to extract high-order causal statistics, dynamically update the weights corresponding to battery health degradation indicators, and generate a combination of performance degradation indicators.

[0042] Based on the combination of multi-order tensor causal relationships and performance degradation indicators, a causal diagnosis graph is constructed. The quantum representation of the causal diagnosis graph is learned using a pre-trained quantum graph neural network, and the semantic consistency of the graph is maintained through graph embedding regularization constraints.

[0043] Based on the causal diagnostic map, the contribution of different indicators to the diagnostic results is quantified, and the probability distribution of the diagnostic results is determined based on the quantum density matrix. The current health diagnostic result is determined by selecting the diagnostic result corresponding to the maximum probability.

[0044] According to a second aspect of the embodiments of the present invention,

[0045] Provided is a big data-based energy storage power station battery health status diagnosis system, including:

[0046] The first unit is used to collect multimodal operation data at the PACK level and the cell level of the energy storage power station battery, homomorphically encrypt the multimodal operation data to obtain ciphertext data, analyze and calculate the ciphertext data based on the self-attention data fusion mechanism of Transformer, adaptively extract explicit and implicit correlation features, and obtain fused time series data;

[0047] The second unit is used to build a battery health diagnosis model based on the fused time series data, extract the quantum entanglement features of multiple paths through the convolution operation of the quantum walk convolution kernel, dynamically adjust the phase and amplitude of the quantum bit in combination with the self-evolving quantum attention mechanism, explore the combination weights of different feature subspaces, generate diagnostic feature enhanced representation, based on the diagnostic feature enhanced representation, realize the entanglement storage of long and short-term memory through quantum circuits, and combine with the gate control unit to predict the health state sequence;

[0048] The third unit is used to determine the battery health prediction result based on the predicted health status sequence, construct a quantum reinforcement learning algorithm, and search the diagnostic strategy combination space online by combining quantum state superposition and quantum parallelism. In the loop of the quantum reinforcement learning algorithm, a multi-order causal relationship tensor model is embedded to model the causal relationship strength between battery performance indicators, determine the battery health degradation indicator, generate a causal diagnostic map, and obtain the battery health status diagnosis result.

[0049] According to a third aspect of the embodiments of the present invention,

[0050] An electronic device is provided, comprising:

[0051] processor;

[0052] a memory for storing processor-executable instructions;

[0053] The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0054] A fourth aspect of the embodiments of the present invention is:

[0055] A computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the aforementioned method is implemented.

[0056] In the embodiment of the present invention, multimodal operation data is processed by homomorphic encryption, which ensures the privacy and security of data during analysis and calculation, and reduces the risk of data leakage; it can adaptively extract explicit and implicit correlation features, automatically discover important patterns and relationships in the data, and improve the accuracy and efficiency of data analysis; through the convolution operation of the quantum walk convolution kernel, the quantum entanglement features of multiple paths are extracted, the complex correlations between different battery states are captured, and the richness and diversity of feature extraction are improved; the phase and amplitude of quantum bits are dynamically adjusted, the combined weights of different feature subspaces are explored, and diagnostic feature enhanced representations are generated, which improves the accuracy and effectiveness of feature representation; the entangled storage of long and short-term memory is realized by quantum circuits, and combined with gated units, it can effectively process time series data and capture the health status of batteries. Long-term and short-term dependencies enhance predictive capabilities; the enhanced representation of diagnostic features generated based on quantum entanglement features and self-evolving quantum attention mechanisms provides higher quality input data for health status sequence prediction, improving the performance and accuracy of the prediction model; through quantum state superposition and quantum parallelism, the quantum reinforcement learning algorithm can simultaneously explore and evaluate multiple strategy combinations, significantly improving the efficiency of strategy search and optimization, enabling the algorithm to find the best diagnostic strategy more quickly; by embedding multi-order causal tensor models, the algorithm can model and quantify the complex causal relationships between battery performance indicators, improving the accurate judgment of battery health degradation indicators, thereby generating a more accurate causal diagnostic map; the causal diagnostic map provides a detailed causal analysis of the battery health status, revealing the causal relationship and impact intensity between different performance indicators. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] Figure 1 A schematic diagram of a process for diagnosing the health status of batteries in an energy storage power station based on big data according to an embodiment of the present invention;

[0058] Figure 2 It is a structural schematic diagram of a battery health status diagnosis system for an energy storage power station based on big data according to an embodiment of the present invention. DETAILED DESCRIPTION

[0059] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0060] The technical solution of the present invention is described in detail with specific embodiments below. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described in detail in some embodiments.

[0061] Figure 1 FIG. 1 is a flow chart of a method for diagnosing the health status of batteries in an energy storage power station based on big data according to an embodiment of the present invention. Figure 1 As shown, the method includes:

[0062] S101. Collect multimodal operation data of the battery pack level and the cell level of the energy storage power station, homomorphically encrypt the multimodal operation data to obtain ciphertext data, analyze and calculate the ciphertext data based on the self-attention data fusion mechanism of Transformer, adaptively extract explicit and implicit correlation features, and obtain fused time series data;

[0063] The PACK level specifically refers to the overall unit of the battery module. A battery PACK is usually composed of multiple cells and includes a management and monitoring system to ensure the safety, efficiency and performance of the battery. The cell level specifically refers to the most basic unit that makes up the battery PACK, that is, a single battery cell. Each cell is an independent electrochemical unit with independent voltage and capacity characteristics.

[0064] In this embodiment, multimodal operating data is processed through homomorphic encryption to ensure the privacy and security of the data during the analysis and calculation process. Even in the encrypted state, the data can still be processed, reducing the risk of data leakage; the Transformer-based self-attention data fusion mechanism can adaptively extract explicit and implicit correlation features, and the system can automatically discover important patterns and relationships in the data, thereby improving the accuracy and efficiency of data analysis; by extracting explicit and implicit features, fused time series data is obtained, which can more comprehensively reflect the battery operating status, and contribute to more accurate status monitoring, fault prediction and life estimation.

[0065] In an optional embodiment, multimodal operation data of the battery PACK level and the battery cell level of the energy storage power station are collected, the multimodal operation data are homomorphically encrypted to obtain ciphertext data, the ciphertext data are analyzed and calculated based on the self-attention data fusion mechanism of Transformer, explicit and implicit correlation features are adaptively extracted, and the fused time series data obtained includes:

[0066] Collect multimodal operation data at the PACK and cell levels of the energy storage power station battery, including voltage, current, temperature and state of charge. During the collection process, introduce edge computing devices based on the MQTT protocol to collect and preprocess the multimodal operation data in real time, and perform hash operations on the multimodal operation data through a pre-built blockchain data authentication mechanism, and store the hash value in the blockchain;

[0067] Perform fully homomorphic encryption based on the lattice cryptographic system on multimodal operation data, convert the original data into encrypted ciphertext data, introduce a noise mechanism in the encryption process, encrypt the original data one by one based on the preset public key, and divide the preset private key into multiple shares through the key management protocol based on secure multi-party computing, and distribute them to multiple energy storage power stations, where the public key and secret key are distributed and updated through secure multi-party computing;

[0068] The ciphertext data is input into the Transformer-based self-attention data fusion model, and the explicit and implicit correlation features between different modal data are calculated through the self-attention mechanism, wherein the self-attention data fusion model adopts an encoder-decoder architecture, the encoder introduces a position encoding mechanism to capture the position information of the ciphertext data, the encoder and the decoder adopt a multi-head attention mechanism to capture the correlation features between the ciphertext data from different subspaces, and the corresponding features are extracted from the ciphertext data and fused through a forward propagation algorithm based on fully homomorphic encryption to obtain fused time series ciphertext data;

[0069] Based on the homomorphic decryption algorithm corresponding to the fully homomorphic encryption algorithm, the fused time series ciphertext data is homomorphically decrypted, and the ciphertext data is decrypted one by one using the private key. Combined with the GPU-based parallel decryption framework, the ciphertext data is transferred to the GPU in batches, and the decryption algorithm is executed in parallel on the GPU to obtain the fused time series decrypted data.

[0070] In the energy storage power station, sensors and data acquisition equipment are installed to collect the operation data of the battery PACK level and the battery cell level. The collected data includes multimodal data such as voltage, current, temperature and state of charge. An edge computing device based on the MQTT (Message Queue Telemetry Transmission) protocol is introduced, such as Raspberry Pi or industrial gateway. The edge computing device is connected to the sensor and data acquisition device, and receives the collected multimodal operation data in real time through the MQTT protocol. The data preprocessing program is run on the edge computing device to perform real-time preprocessing on the received multimodal operation data; the blockchain data authentication program is deployed on the edge computing device to perform hash operations on the preprocessed multimodal operation data to generate a hash value of the data. The hash algorithm can select a more secure algorithm, preferably SHA-256; the generated hash value is uploaded to the blockchain network and stored in the blockchain transaction to form a digital fingerprint of the data. Through the decentralized characteristics and consensus mechanism of the blockchain, the immutability and traceability of the hash value are ensured, and the trusted authentication of the data is realized.

[0071] Deploy a fully homomorphic encryption program on the edge computing device to encrypt the preprocessed multimodal operation data. Select a fully homomorphic encryption algorithm based on the lattice cryptographic system, exemplarily the BGV or CKKS scheme; the fully homomorphic encryption algorithm requires the generation of public and private keys. In the initialization phase, multiple energy storage power stations jointly participate in the generation of public and private keys through a secure multi-party computing protocol; each participant generates its own private key share and keeps it confidential; the participants jointly calculate the public key to obtain a complete public key; the public key is disclosed to all energy storage power stations for data encryption. In the data encryption process, the preprocessed multimodal operation data is encrypted one by one using the public key. For each data point, perform the following steps: encode the data point into a polynomial form and convert it into an input format supported by the fully homomorphic encryption algorithm; use the public key to encrypt the encoded data point to generate the corresponding ciphertext data; to enhance security, introduce a noise mechanism in the encryption process. Generate random noise, add it to the encoded data point, and then encrypt it to obtain the noisy ciphertext data. The encrypted ciphertext data will be used for subsequent data fusion and calculation, and privacy protection and secure computing can be achieved without decryption. To ensure the security and recoverability of the private key, a key management protocol of secure multi-party computing is adopted. The private key is divided into multiple shares and distributed to multiple energy storage power stations. Each power station holds a private key share, and the complete private key cannot be recovered alone. Through the secure multi-party computing protocol, the private key shares are regularly updated and rotated to ensure the long-term security of the key. At the same time, through the key recovery protocol involving multiple parties, when the data needs to be decrypted, the complete private key can be reconstructed to ensure the recoverability of the data.

[0072] The encrypted ciphertext data is sent to the central server for data fusion and analysis. The central server has high-performance computing resources and storage capacity for processing large-scale encrypted data. The Transformer-based self-attention data fusion model is deployed on the central server. Transformer is a deep learning model based on the self-attention mechanism, which is suitable for processing time series data. The self-attention data fusion model adopts an encoder-decoder architecture. The encoder is used to convert the input ciphertext data into a hidden representation, and the decoder is used to generate the fused output based on the hidden representation. In the encoder, a position encoding mechanism is introduced to embed the position information of the ciphertext data into the input representation. Position encoding can use sine and cosine functions or learnable position embedding vectors to capture the temporal dependency of the data. Both the encoder and the decoder use a multi-head attention mechanism. Multi-head attention learns the correlation features between data from different subspaces through multiple parallel attention heads. Each attention head independently calculates the attention weight to capture different correlation patterns. In the self-attention mechanism, the attention weight matrix is ​​obtained through pairwise calculations between ciphertext data. The attention weight represents the correlation and importance between data points and is used to aggregate information at different time steps. In order to realize the self-attention calculation on the ciphertext data, a forward propagation algorithm based on fully homomorphic encryption is adopted. Utilizing the homomorphic property of fully homomorphic encryption, matrix multiplication, addition and other operations are performed on the encrypted data to obtain the encrypted attention weights and hidden representations. Through multiple rounds of self-attention calculation and forward propagation, the final fused time series ciphertext data is obtained. The fused ciphertext data contains the explicit and implicit correlation features between different modal data, realizing the secure fusion of multimodal data.

[0073] The fused time series ciphertext data is sent to the decryption server for data decryption. The decryption server has GPU acceleration capability for efficient execution of the decryption algorithm. A GPU-based parallel decryption framework, such as CUDA or OpenCL, is deployed on the decryption server. The parallel decryption framework allows multiple decryption algorithms to be executed simultaneously on the GPU, making full use of the parallel computing capabilities of the GPU. The fused time series ciphertext data is transferred to the GPU in batches. Each batch contains a certain amount of ciphertext data, which is divided according to the GPU memory and computing power. A thread is allocated to each ciphertext data on the GPU to execute the decryption algorithm in parallel. The decryption algorithm corresponds to the fully homomorphic encryption algorithm, and the ciphertext data is decrypted one by one using the private key. During the decryption process, private key shares are first collected from multiple energy storage power stations, and the complete private key is reconstructed through a secure multi-party computing protocol. Each ciphertext data is decrypted using the reconstructed private key. The decryption algorithm restores the ciphertext data to plaintext form according to the specific steps of the fully homomorphic encryption scheme, namely, lattice-based decryption. The decrypted plaintext data is the fused time series data, which contains the feature representation after the fusion of multimodal data. The decrypted data is transmitted back to the central server for subsequent battery health diagnosis.

[0074] In this embodiment, edge computing devices are used for real-time data preprocessing, which reduces data transmission delays and improves system response speed. Communication based on the MQTT protocol ensures the reliability and real-time nature of data transmission. The preprocessed data is hashed through the blockchain to generate a tamper-proof digital fingerprint to ensure the credibility and traceability of the data. Multimodal operating data is fully homomorphically encrypted to ensure that the data can be processed and analyzed even in an encrypted state, thereby protecting data privacy. A noise mechanism is introduced during the encryption process to further enhance data security. Public and private keys are generated using a secure multi-party computing protocol, and private key shares are held through multiple energy storage power stations to ensure the security and recoverability of the keys. Private key shares are regularly updated and rotated to ensure long-term security of the keys and avoid single point failures. A Transformer-based self-attention data fusion model is deployed on the central server to achieve efficient fusion of multimodal data. Position encoding mechanisms and multi-head attention mechanisms are introduced to capture the temporal dependencies of data and improve the accuracy of data analysis.

[0075] S102. Based on the fused time series data, a battery health diagnosis model is constructed. Through the convolution operation of the quantum walk convolution kernel, the quantum entanglement features of the multi-path are extracted. Combined with the self-evolving quantum attention mechanism, the phase and amplitude of the quantum bit are dynamically adjusted, and the combination weights of different feature subspaces are explored to generate diagnostic feature enhanced representations. Based on the diagnostic feature enhanced representations, the entangled storage of long-term and short-term memory is realized through quantum circuits, and combined with the gate control unit, the health state sequence is predicted;

[0076] The quantum entanglement feature specifically refers to a feature extracted based on the principle of quantum entanglement, which uses the non-locality and multipath interference of quantum mechanics to reveal the underlying complex relationships in the data. Quantum entanglement is a special correlation between quantum states. Even at a long distance, the change of one quantum state will immediately affect another quantum state. It can be used to capture the deep correlation between different battery states and provide richer feature representation than classical methods. The self-evolving quantum attention mechanism specifically refers to a quantum computing method that dynamically adjusts the phase and amplitude of quantum bits to explore the combined weights of feature subspaces to improve the effect of feature representation. Self-evolution means that the system can autonomously adjust the quantum state, optimize the state of the quantum bit according to the input data, dynamically adjust the phase and amplitude of the quantum bit, optimize the feature extraction process, and improve the model's prediction accuracy for the battery health status.

[0077] In this embodiment, through the convolution operation of the quantum walk convolution kernel, the quantum entanglement features of multiple paths are extracted, the complex correlations between different battery states are captured, and the richness and diversity of feature extraction are improved; the phase and amplitude of the quantum bit are dynamically adjusted, the combination weights of different feature subspaces are explored, and the diagnostic feature enhanced representation is generated, which improves the accuracy and effectiveness of the feature representation; the entangled storage of long-term and short-term memory is realized through quantum circuits, and combined with the gated unit, it can effectively process time series data, capture the long-term and short-term dependencies of the battery health state, and enhance the prediction ability; the diagnostic feature enhanced representation generated based on the quantum entanglement features and the self-evolving quantum attention mechanism provides higher quality input data for the health state sequence prediction, and improves the performance and accuracy of the prediction model.

[0078] In an optional embodiment, based on the fused time series data, a battery health diagnosis model is constructed, and the quantum entanglement features of multiple paths are extracted through the convolution operation of the quantum walk convolution kernel. The phase and amplitude of the quantum bit are dynamically adjusted in combination with the self-evolving quantum attention mechanism, and the combination weights of different feature subspaces are explored to generate diagnostic feature enhanced representations. Based on the diagnostic feature enhanced representations, the entangled storage of long and short-term memory is realized through quantum circuits, and combined with the gate control unit, the health state sequence is predicted, including:

[0079] Construct a quantum walk convolution kernel, map the fused time series data to the quantum bit space, each time step corresponds to a quantum bit, create a translation operator and a coin operator, generate a quantum walk convolution kernel, and perform a convolution operation in the quantum bit space through the quantum walk convolution kernel. The translation operator controls the movement of the quantum bit in space, and the coin operator controls the superposition state of the quantum bit. By adjusting the movement parameters corresponding to the translation operator and the superposition parameters corresponding to the coin operator, the step size and direction of the quantum walk are controlled, and the quantum entanglement characteristics are extracted;

[0080] The quantum entangled features are input into a self-evolving quantum attention module based on the attention mechanism. Through quantum gate operations, the phase and amplitude of the quantum bit are dynamically adjusted to explore the combined weights of different feature subspaces. For each quantum entangled feature, quantum gate operations are performed according to a set of preset parameters to obtain a primary feature representation. The parameters corresponding to the quantum gate operations are adjusted in combination with the self-evolution algorithm, and the diagnostic feature enhanced representation is generated according to the self-attention allocation.

[0081] Input the diagnostic feature enhancement representation into a pre-built quantum long short-term memory circuit, and perform entanglement storage and update on the timing information corresponding to the diagnostic feature enhancement representation. The quantum long short-term memory circuit is composed of an input gate, a forget gate, an output gate, and a state update gate. The input gate controls the new diagnostic feature enhancement representation to enter the memory unit, the forget gate controls the forgetting of historical information, the output gate controls the influence of the memory unit on the current state, and the state update gate updates the state of the memory unit;

[0082] The output of the quantum long short-term memory circuit is entangled with the diagnostic feature enhancement representation through a parameterized quantum circuit to obtain a predicted state. The predicted state is continuously generated by recursively calling the quantum long short-term memory circuit. Each predicted state is measured in combination with quantum measurement operations, and the predicted state quantum state is converted into a classical state to generate a health state prediction sequence.

[0083] Map the fused time series data to the quantum bit space. Each time step corresponds to a quantum bit, forming a quantum bit sequence. The quantum bit is the basic unit in quantum computing and can represent the superposition of quantum states. Create a translation operator and a coin operator to generate a quantum walk convolution kernel. The translation operator controls the movement of the quantum bit in space, and the coin operator controls the superposition state of the quantum bit. Define the movement parameter corresponding to the translation operator and the superposition parameter corresponding to the coin operator. The movement parameter determines the step size and direction of the movement of the quantum bit in space, and the superposition parameter determines the weight of the superposition state of the quantum bit. By adjusting the movement parameter and the superposition parameter, different quantum walk convolution kernels are generated. The quantum walk convolution kernel will perform convolution operations in the quantum bit space to extract quantum entanglement features. Apply the quantum walk convolution kernel to the quantum bit sequence, and establish entanglement relationships between different quantum bits through the process of quantum walks. Quantum entanglement is a non-classical correlation between quantum states, which can capture long-range dependencies in time series data. Through multiple iterations of quantum walk convolution operations, quantum entanglement features are extracted at different scales and directions. The step size and direction of the quantum walk determine the granularity and scope of feature extraction.

[0084] The extracted quantum entanglement features are input into the self-evolving quantum attention module. Based on the attention mechanism, the self-evolving quantum attention module dynamically adjusts the phase and amplitude of quantum bits through quantum gate operations to explore the combination weights of different feature subspaces. For each quantum entanglement feature, quantum gate operations are performed according to a preset set of parameters. Quantum gates are the basic operation units in quantum computing, which can transform quantum states by rotating and shifting phases. By applying different quantum gate combinations, such as Hadamard gates, CNOT gates, rotation gates, etc., quantum entanglement features are transformed and combined to obtain primary feature representations. The primary feature representation contains feature combination information of different subspaces. The self-evolution algorithm is introduced to dynamically adjust the parameters corresponding to the quantum gate operation. The self-evolution algorithm learns the optimal quantum gate parameter combination by optimizing the objective function, such as minimizing the reconstruction error or maximizing the mutual information. According to the parameters obtained by the self-evolution algorithm, the quantum gate operation is updated to obtain the optimized feature representation. Through the iterative optimization process, the quality and expression ability of the feature representation are continuously improved. Combined with the self-attention mechanism, the optimized feature representation is weighted combined. The self-attention mechanism generates attention weights by calculating the correlation between different features, and dynamically weights the features. The weighted feature representation is used as the diagnostic feature enhancement representation and passed to the subsequent quantum long short-term memory circuit.

[0085] The diagnostic feature enhancement representation is input into the pre-built quantum long short-term memory circuit. The quantum long short-term memory circuit is a recurrent neural network based on quantum computing, which is used to process time series data. The quantum long short-term memory circuit consists of an input gate, a forget gate, an output gate, and a state update gate. These gates control the flow and update of information in the memory unit. The input gate controls the entry of new diagnostic feature enhancement representation into the memory unit. Through quantum gate operation, the new feature is entangled with the state of the memory unit, which determines the degree of influence of the new information on the memory. The forget gate controls the forgetting of historical information. Through quantum gate operation, the state of the memory unit is partially reset, and no longer relevant historical information is forgotten. The output gate controls the influence of the memory unit on the current state. Through quantum gate operation, the state of the memory unit is entangled with the current input to generate an output signal. The state update gate updates the state of the memory unit. Through quantum gate operation, the state of the memory unit is updated by combining the signals of the input gate, forget gate, and output gate to maintain the continuity and temporal dependency of the memory. By recursively calling the quantum long short-term memory circuit, the temporal information of the diagnostic feature enhancement representation is entangled, stored, and updated. The output of each time step depends on the previous memory state and the current input.

[0086] The output of the quantum long short-term memory circuit is processed by a parameterized quantum circuit. The parameterized quantum circuit is a trainable quantum circuit that transforms and maps the input through parameterized quantum gate operations. Through the parameterized quantum circuit, the output of the quantum long short-term memory circuit and the diagnostic feature enhancement representation are entangled. The entanglement operation links two quantum states and makes them correlated. By optimizing the parameters of the parameterized quantum circuit, the optimal mapping relationship between the output and the feature is learned. The optimization process can use a gradient descent algorithm to update the parameters by minimizing the prediction error. After the parameterized quantum circuit transformation, the predicted state is obtained. The predicted state is a quantum state that contains the predicted information of the future health state. The predicted state is continuously generated by recursively calling the quantum long short-term memory circuit and the parameterized quantum circuit. Each predicted state is based on the previous prediction and the current feature to form a prediction sequence. A quantum measurement operation is performed on each predicted state. The quantum measurement converts the quantum state into a classical state to obtain a certain prediction result. The measurement operation can be based on different measurement bases, such as the computational basis, the Hadamard basis, etc. Through multiple measurements and statistics, the classical probability distribution corresponding to each predicted state is obtained. The probability distribution represents the possibility of different health states. Based on the probability distribution of the predicted state, a health state prediction sequence is generated. The prediction sequence contains the battery health state prediction for a period of time in the future.

[0087] In this embodiment, the fused time series data is mapped to the quantum bit space so that each time step corresponds to a quantum bit. The superposition and entanglement characteristics of quantum states are used to improve the efficiency and accuracy of time series data processing. By creating a translation operator and a coin operator, a quantum walk convolution kernel is generated, and a convolution operation is performed on the quantum bit sequence to extract multi-path quantum entanglement features and capture long-range dependencies in the time series data. The self-evolving quantum attention module dynamically adjusts the phase and amplitude of the quantum bit through quantum gate operations to optimize the feature representation and ensure the quality and expressiveness of the feature. Through an iterative optimization process combined with a self-attention mechanism, the features are dynamically weighted and combined to improve the accuracy and reliability of the enhanced representation of diagnostic features. A quantum long-short-term memory circuit is constructed to process time series data. Through the input gate, forget gate, output gate and state update gate, the flow and update of information in the memory unit are controlled to ensure the continuity and timing dependence of memory; the quantum long short-term memory circuit realizes the entangled storage of timing information represented by diagnostic feature enhancement, captures the deep correlation between historical information and current input, and provides efficient state prediction; the output of the quantum long short-term memory circuit is processed by parameterized quantum circuit, the quantum gate parameters are optimized, the optimal mapping relationship between output and features is learned, and an accurate prediction state is generated; quantum measurement operations are performed on each prediction state to convert the quantum state into a classical state, and a definite prediction result and probability distribution are obtained to ensure the accuracy and reliability of the prediction; the introduction of the self-evolving quantum attention mechanism enables the model to adaptively optimize the feature extraction and representation process according to the input data, thereby improving the generalization ability and stability of the prediction model.

[0088] In an optional embodiment, the method further includes:

[0089] The quantum walk convolution kernel extracts quantum entanglement features through multi-step quantum walks. The formula is as follows:

[0090] ;

[0091] Among them, U walk represents the quantum walk convolution kernel, l represents the quantum walk step number, L represents the total number of quantum walk steps, represents the translation operator, represents the movement parameter, represents the hidden operator, represents the superposition parameter, I represents the identity matrix, Represents a tensor product operation.

[0092] Through the translation and superposition operations in multiple quantum walk steps, the complex correlation between quantum states in different positions and states is gradually extracted. Each step of operation will affect the current state of the quantum state. Through multiple steps of iteration, the quantum walk convolution kernel can capture the long-range dependencies and quantum entanglement characteristics in the data. Represents the tensor product operation, which is used to combine two matrices into a larger matrix. The product symbol in the formula indicates the continuous application of these operations, ensuring that each step is performed on the basis of the previous steps to form the final convolution kernel.

[0093] According to the formula, through the multi-step quantum walk process, the formula can extract the complex entanglement features between quantum states. The features can reveal the deep-level correlations and patterns hidden in the data, and can capture long-range dependencies better than traditional methods; the combination of translation operators and hidden operators realizes efficient information conversion of quantum states. The translation operator moves the quantum state in space, and the hidden operator superimposes and adjusts the weight of the quantum state. This information conversion capability enables the convolution kernel to effectively transmit information and extract features between multiple quantum bits; through the continuous application of multiple quantum walk steps, the formula realizes multi-level and multi-scale feature extraction of data. This multi-step iterative process enhances the data representation capability, making the extracted features richer and more comprehensive; using the superposition and entanglement characteristics of quantum bits, the formula can operate in high-dimensional quantum state space and capture high-dimensional features that are difficult to discover with traditional computing methods. It helps to improve the accuracy and effect of feature extraction.

[0094] In an optional embodiment, the output of the quantum long short-term memory circuit is entangled with the diagnostic feature enhancement representation through a parameterized quantum circuit to obtain a predicted state including:

[0095] The predicted state is formulated as follows:

[0096] ;

[0097] in, represents the predicted quantum state of the health state at the tth time step, p represents the quantum gate operation sequence, N p represents the total number of quantum gate operations, γ p,t represents the weight of the p-th quantum gate operation at the t-th time step, V p represents the pth parameterized quantum gate operation at the tth time step, λ p,t represents the parameters controlling the p-th quantum gate operation at the t-th time step, represents the hidden state quantum state at the t-th time step, Represents the adjusted quantum entanglement feature representation at the t-th time step.

[0098] By performing quantum gate operations and combining the output state of the quantum long short-term memory circuit and the adjusted quantum entanglement feature representation, a predicted quantum state of the battery health state is generated.

[0099] First, the output state of the quantum long short-term memory circuit and the adjusted quantum entanglement feature representation are expressed by a tensor product operation ( ) to form a composite state of two quantum states, forming a joint quantum state. A series of parameterized quantum gate operations are applied to the joint quantum state, each of which is controlled by specific parameters. These operations can transform the quantum state by rotating it, shifting its phase, etc. The results of all quantum gate operations are weighted and summed according to their respective weight factors, and the effects of different operations are combined to generate the final predicted quantum state. Finally, the predicted state quantum state at the tttth time step is generated through the above steps. This quantum state contains the predicted information about the battery health status, and is converted into a specific prediction result through subsequent quantum measurement operations.

[0100] According to the formula, by performing a tensor product operation on the output state of the quantum long-short term memory circuit and the adjusted quantum entanglement feature representation, the formula effectively combines historical memory and real-time features, enhancing the comprehensive representation capability of the battery health state; the parameterized quantum gate operation and weighted summation mechanism enable the model to flexibly adjust the feature processing process, thereby improving the accuracy of health state prediction; by transforming the combined quantum state through a series of quantum gate operations, the formula can explore the combined weights of different feature subspaces. This flexible mapping capability enables the model to adapt to different feature distributions, improving the adaptability and accuracy of the prediction model; weighted summation of the results of quantum gate operations allows the model to integrate the results of multiple operations. This multi-level feature integration method helps to capture more detailed and complex data patterns, thereby improving the comprehensive prediction performance of the model; using the characteristics of quantum computing, such as quantum superposition and entanglement, the formula can process higher-dimensional data features and perform complex operations in quantum state space.

[0101] S103. Based on the predicted health state sequence, the battery health prediction result is determined, and a quantum reinforcement learning algorithm is constructed. By combining quantum state superposition and quantum parallelism, the diagnosis strategy combination space is searched online. In the loop of the quantum reinforcement learning algorithm, a multi-order causal relationship tensor model is embedded to model the causal relationship strength between battery performance indicators, determine the battery health degradation indicator, generate a causal diagnosis map, and obtain the battery health state diagnosis result;

[0102] The quantum state superposition specifically refers to a basic phenomenon in quantum computing, in which a quantum bit can be in a superposition of multiple states at the same time, rather than being in only one certain state like a classical bit. Quantum superposition allows quantum systems to process multiple information at the same time, thereby achieving more efficient data processing and computing; the quantum parallelism specifically refers to the ability of quantum computing to use quantum superposition and quantum entanglement to execute multiple computing paths or operations at the same time. Through parallel processing, quantum computing can complete multiple tasks that traditional computing cannot perform simultaneously in a single computing step. The causal diagnostic map specifically refers to a graphical model used to represent the causal relationship between variables. Through causal relationship modeling, the intensity and direction of the influence of each factor on the target variable can be determined, thereby revealing the causal structure in complex systems.

[0103] In this embodiment, through quantum state superposition and quantum parallelism, the quantum reinforcement learning algorithm can simultaneously explore and evaluate multiple strategy combinations, significantly improving the efficiency of strategy search and optimization, so that the algorithm can find the best diagnostic strategy more quickly; by embedding a multi-order causal relationship tensor model, the algorithm can model and quantify the complex causal relationship between battery performance indicators, improve the accurate judgment of battery health degradation indicators, and thus generate a more accurate causal diagnostic map; the causal diagnostic map provides a detailed causal analysis of the battery health status, reveals the causal relationship and influence intensity between different performance indicators, makes the diagnostic process more transparent, and helps to identify potential problems and optimize maintenance strategies; the online search capability of the quantum reinforcement learning algorithm makes it possible to dynamically adjust the diagnostic strategy in the real-time data stream, ensuring the real-time monitoring and accurate prediction of the battery health status.

[0104] In an optional embodiment, based on the predicted health state sequence, the battery health prediction result is determined, a quantum reinforcement learning algorithm is constructed, and by combining quantum state superposition and quantum parallelism, the online search of the diagnostic strategy combination space includes:

[0105] Map the performance indicators of the battery under different health states to the probability amplitude of the quantum state, construct a quantum representation of the battery health state, add the quantum phase as a hidden variable to the quantum state, and capture the nonlinear dynamic characteristics of the battery health degradation process;

[0106] Based on quantum circuits, value iteration and strategy iteration algorithms are constructed. Strategy search is accelerated through quantum parallel computing. The optimal diagnostic strategy combination is located using quantum amplitude amplification. The diagnostic strategy combination space is sampled through quantum gate sequences.

[0107] Based on the diagnostic strategy combination space, a search tree of diagnostic strategy combinations is dynamically generated, branches of the search tree are pruned based on heuristic knowledge, and adaptive step length exploration and utilization of the diagnostic strategy combination space is performed based on a quantum walk algorithm.

[0108] Collect performance indicator data of the battery under different health states, such as voltage, current, temperature, internal resistance, etc. These indicators reflect the health status and degradation degree of the battery. Preprocess and normalize the performance indicator data to eliminate the scale difference and noise interference of the data. Preprocessing can include operations such as outlier removal, smoothing filtering, and standardization. Select an appropriate quantum state dimension and determine the number of quantum bits required for the quantum state according to the number and range of performance indicators. The quantum state is the state of a quantum system, which consists of complex amplitude and phase. Map the preprocessed performance indicators to the probability amplitude of the quantum state. The probability amplitude represents the projection coefficient of the quantum state on the corresponding basis vector, reflecting the probability distribution of different health states. Introduce the quantum phase as a hidden variable and add it to the quantum state. The quantum phase is the argument of the complex number, which can capture the nonlinear dynamic characteristics of the battery health degradation process, such as the change in degradation rate, mutation point, etc. By adjusting the quantum phase, optimize the quantum state's ability to represent the battery health state. The optimization goal can be to minimize the difference between the quantum state and the classical probability distribution, such as KL divergence, Wasserstein distance, etc. Iterate and optimize the quantum phase until the quantum state can accurately represent the battery health state distribution. The optimization process can use the gradient descent algorithm to update the phase by calculating the parameter gradient of the quantum circuit.

[0109] Based on quantum circuits, value iteration and policy iteration algorithms are constructed. The value iteration algorithm estimates the long-term benefits under different diagnostic strategies, and the policy iteration algorithm searches for the optimal diagnostic strategy combination. The diagnostic strategy combination is mapped to the input state of the quantum circuit. Each diagnostic strategy corresponds to a quantum bit, and the strategy combination is represented by the tensor product of multiple quantum bits. The structure of the quantum circuit and the sequence of quantum gates are designed to realize the calculation process of value iteration and policy iteration. Through quantum parallel computing, the benefit values ​​of multiple diagnostic strategy combinations are evaluated simultaneously. Quantum parallelism allows an exponential number of strategy combinations to be calculated in a single quantum circuit operation, which greatly accelerates the search process. The quantum amplitude amplification technology, preferably the Grover algorithm, is used to amplify the quantum state amplitude corresponding to the optimal diagnostic strategy combination. Amplitude amplification can quickly locate the optimal solution and reduce the number of searches. The result of the optimal diagnostic strategy combination is obtained by measuring the output state of the quantum circuit. The measurement operation collapses the quantum state into a classical state and obtains a determined strategy combination. The quantum circuit is iteratively executed, and the value function and strategy are continuously updated until convergence to the optimal diagnostic strategy combination. The convergence condition can be that the change in the strategy combination is less than the preset threshold or the maximum number of iterations is reached.

[0110] Based on the diagnostic strategy combination space, a search tree is dynamically generated. The nodes of the search tree represent different diagnostic strategy combinations, and the edges represent the transfer relationship between strategies. The root node represents the initial diagnostic strategy combination, and the child node represents the new strategy combination obtained by adding or deleting a diagnostic operation based on the current strategy. The search tree is pruned using heuristic knowledge to remove infeasible or low-quality strategy combination branches. Heuristic knowledge can come from expert experience, historical data analysis, etc., and is used to evaluate the pros and cons of strategy combinations. A quantum walk algorithm is designed to explore and utilize the search tree. Quantum walk is the evolution process of quantum systems on graph structures, which can achieve efficient state transfer and information propagation. The nodes of the search tree are mapped to quantum states, and the edges are mapped to the transfer operators of quantum walks. The transfer operator determines the transfer probability and phase accumulation of quantum states between nodes. By adjusting the parameters of the transfer operator, such as step size and direction, the exploration and utilization strategy of quantum walks on the search tree is controlled. A larger step size is conducive to global exploration, and a smaller step size is conducive to local utilization. An adaptive step size mechanism is introduced to dynamically adjust the parameters of quantum walks according to the feedback of the search process. The adaptive mechanism can adjust the step size to balance exploration and utilization based on the node evaluation value, access frequency and other information. The quantum walk algorithm is iteratively executed until the preset termination condition is reached, such as finding a satisfactory diagnosis strategy combination, reaching the maximum number of iterations, etc. By measuring the quantum state, the node information at the end of the quantum walk is obtained as the output of the optimized diagnosis strategy combination.

[0111] In this embodiment, the quantum phase is introduced as a hidden variable, which can capture the nonlinear characteristics of the battery health degradation process; by optimizing the quantum phase, the quantum state can more accurately represent the battery health status distribution, thereby improving the accuracy of health status diagnosis; based on the value iteration and strategy iteration algorithm of the quantum circuit, the benefit values ​​of multiple diagnostic strategy combinations are simultaneously evaluated through quantum parallel computing, thereby accelerating the strategy search process; the quantum amplitude amplification technology is used to quickly locate the optimal diagnostic strategy combination and reduce the number of searches; the search tree is dynamically generated and pruned, and the low-quality strategy combinations are removed using heuristic knowledge to improve the search efficiency; a quantum walk algorithm is designed for efficient exploration and utilization, balancing global exploration and local utilization to ensure that the optimal strategy combination is found; an adaptive step size mechanism is introduced to dynamically adjust the quantum walk parameters according to the feedback of the search process, optimize the search path, and improve the search effect; through multiple iterations and measurements of the quantum state, it is ensured that the diagnostic strategy combination finally output is fully optimized and verified, and has high robustness and reliability.

[0112] In an optional embodiment, a multi-order causal relationship tensor model is embedded in the loop of the quantum reinforcement learning algorithm to model the causal relationship strength between battery performance indicators, determine the battery health degradation indicator, generate a causal diagnosis map, and obtain the battery health status diagnosis result including:

[0113] Perform high-order singular value decomposition on the battery performance indicators to generate multi-order tensor causal relationships between battery performance indicators. Based on the causal connection strength between the performance indicators, determine the causal strength tensor. Based on the causal discovery algorithm, set the loss function and combine the causal intervention regularization term to construct the causal graph structure between the performance indicators, generate a multi-order causal relationship tensor model, and embed the multi-order causal relationship tensor model in the loop of the quantum reinforcement learning algorithm.

[0114] A quantum-classical hybrid optimization algorithm is used to search for battery health degradation indicators online, and a data-driven quantum causal enhancement method is used to extract high-order causal statistics, dynamically update the weights corresponding to battery health degradation indicators, and generate a combination of performance degradation indicators.

[0115] Based on the combination of multi-order tensor causal relationships and performance degradation indicators, a causal diagnosis graph is constructed. The quantum representation of the causal diagnosis graph is learned using a pre-trained quantum graph neural network, and the semantic consistency of the graph is maintained through graph embedding regularization constraints.

[0116] Based on the causal diagnostic map, the contribution of different indicators to the diagnostic results is quantified, and the probability distribution of the diagnostic results is determined based on the quantum density matrix. The current health diagnostic result is determined by selecting the diagnostic result corresponding to the maximum probability.

[0117] Collect battery performance indicator data, such as voltage, current, temperature, internal resistance, etc.; preprocess the performance indicator data, including outlier removal, data normalization, missing value filling, etc. The purpose of preprocessing is to improve data quality and consistency; perform high-order singular value decomposition (HOSVD) on the preprocessed performance indicator data. HOSVD is a multilinear algebraic technique that can decompose high-dimensional tensors into the product of multiple low-rank matrices to reveal the potential structure and relationship in the data; through HOSVD, generate multi-order tensor causal relationships between battery performance indicators. Multi-order tensors represent high-order interactions and dependencies between indicators; based on the strength of causal connections between performance indicators, determine the causal strength tensor. The causal strength tensor quantifies the degree of causal influence between each indicator, and the larger the value, the stronger the causal relationship; use causal discovery algorithms, such as PC algorithm, GES algorithm, etc., to construct a causal graph structure between performance indicators. The causal graph is a directed acyclic graph that represents the causal dependency between indicators; in the causal discovery process, set a suitable loss function, preferably the minimum description length criterion, Bayesian information criterion, etc., to measure the degree of fit between the causal graph and the data. At the same time, a causal intervention regularization term is introduced to encourage the learning of sparse and interpretable causal structures. By optimizing the loss function and considering the regularization term, a multi-order causal relationship tensor model is generated. The multi-order causal relationship tensor model combines high-order tensor decomposition and causal discovery algorithms, and can capture the complex causal relationship between battery performance indicators.

[0118] A quantum reinforcement learning algorithm is designed to search and optimize battery health degradation indicators. Quantum reinforcement learning implements policy iteration and value function estimation through quantum circuits, and uses quantum parallelism to accelerate the learning process. A multi-order causal tensor model is embedded in the loop of the quantum reinforcement learning algorithm. The parameters of the tensor model are mapped to the gate parameters of the quantum circuit, so that the quantum state can encode causal information. A quantum-classical hybrid optimization algorithm is used to search for battery health degradation indicators online. The quantum part is responsible for policy evaluation and generation, and the classical part is responsible for policy improvement and parameter update. A data-driven quantum causal enhancement method is introduced to extract high-order causal statistics. By designing specific quantum circuits, statistics such as conditional probability distribution and mutual information between performance indicators are calculated to reveal potential causal relationships. Based on high-order causal statistics, the weights corresponding to battery health degradation indicators are dynamically updated. The weight reflects the degree of influence of each indicator on health degradation, and the larger the weight, the more important the indicator. Through the iterative optimization process, the optimal combination of performance degradation indicators is searched and generated. This combination can best characterize the health degradation state of the battery.

[0119] Based on the combination of multi-order tensor causal relationships and performance degradation indicators, a causal diagnostic graph is constructed. The nodes of the graph represent performance indicators, the edges represent the causal relationships between indicators, and the weights represent the causal strength. The quantum representation of the causal diagnostic graph is learned using a pre-trained quantum graph neural network (QGNN). QGNN maps the graph to the quantum state space and encodes and learns the graph structure through quantum circuits. During the training process of QGNN, graph embedding regularization constraints are introduced. The regularization constraints encourage the learned quantum representation to maintain the semantic consistency of the graph, that is, similar nodes should have similar quantum states. By optimizing the objective function of QGNN, such as minimizing the reconstruction error and maximizing the mutual information, the optimal quantum representation of the graph is learned. The quantum representation can capture the structural characteristics and semantic information of the graph.

[0120] The contribution of different indicators to the diagnosis results is quantified by using the learned quantum representation of the causal diagnosis graph. The contribution can be measured by calculating the quantum state amplitude of the node, node centrality and other indicators. Based on the quantum density matrix, the probability distribution of the diagnosis results is determined. The quantum density matrix represents the state distribution of the quantum system. By measuring the density matrix, the probabilities of different diagnosis results can be obtained. A quantum measurement circuit is designed to measure the quantum state of the causal diagnosis graph. The measurement results correspond to different diagnosis results, such as "normal", "mild attenuation", "severe attenuation", etc. The probability distribution of the diagnosis results is obtained by multiple measurements and statistical probabilities. The probability distribution reflects the possibility of different health states. The diagnosis result corresponding to the maximum probability is selected as the current health diagnosis result. The result with the highest probability represents the most likely battery health state under given performance indicators and causal relationships. The health diagnosis results are output, and a diagnosis report and maintenance recommendations are generated as needed. The diagnosis report includes battery health status assessment, key performance indicator analysis, degradation cause inference, etc., to provide support for subsequent maintenance decisions.

[0121] In this embodiment, a multi-order causal relationship tensor model is generated through high-order singular value decomposition (HOSVD) and causal discovery algorithm to capture the complex causal relationship between battery performance indicators and provide a deeper health status analysis; the quantum reinforcement learning algorithm is used to accelerate strategy search and optimization, and the learning efficiency is improved through quantum parallel computing to quickly find the optimal diagnosis strategy combination; combined with the extraction of high-order causal statistics, the weights of battery health degradation indicators are dynamically updated, the influence of each indicator on health degradation is accurately measured, and an optimized health degradation indicator combination is provided; the quantum representation of the causal diagnosis map is learned through the quantum graph neural network (QGNN) to capture the structural characteristics and semantic information of the map, and to achieve efficient encoding and learning of causal relationships; the learned quantum representation of the causal diagnosis map is used to quantify the contribution of different indicators to the diagnosis results, and the probability distribution of the diagnosis results is determined through quantum measurement to provide accurate battery health status assessment.

[0122] Figure 2 FIG. 1 is a schematic diagram of a battery health status diagnosis system for an energy storage power station based on big data according to an embodiment of the present invention. Figure 2 As shown, the system comprises:

[0123] The first unit is used to collect multimodal operation data at the PACK level and the cell level of the energy storage power station battery, homomorphically encrypt the multimodal operation data to obtain ciphertext data, analyze and calculate the ciphertext data based on the self-attention data fusion mechanism of Transformer, adaptively extract explicit and implicit correlation features, and obtain fused time series data;

[0124] The second unit is used to build a battery health diagnosis model based on the fused time series data, extract the quantum entanglement features of multiple paths through the convolution operation of the quantum walk convolution kernel, dynamically adjust the phase and amplitude of the quantum bit in combination with the self-evolving quantum attention mechanism, explore the combination weights of different feature subspaces, generate diagnostic feature enhanced representation, based on the diagnostic feature enhanced representation, realize the entanglement storage of long and short-term memory through quantum circuits, and combine with the gate control unit to predict the health state sequence;

[0125] The third unit is used to determine the battery health prediction result based on the predicted health status sequence, construct a quantum reinforcement learning algorithm, and search the diagnostic strategy combination space online by combining quantum state superposition and quantum parallelism. In the loop of the quantum reinforcement learning algorithm, a multi-order causal relationship tensor model is embedded to model the causal relationship strength between battery performance indicators, determine the battery health degradation indicator, generate a causal diagnostic map, and obtain the battery health status diagnosis result.

[0126] According to a third aspect of the embodiments of the present invention,

[0127] An electronic device is provided, comprising:

[0128] processor;

[0129] a memory for storing processor-executable instructions;

[0130] The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0131] A fourth aspect of the embodiments of the present invention is:

[0132] A computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the aforementioned method is implemented.

[0133] The present invention may be a method, an apparatus, a system and / or a computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for executing various aspects of the present invention.

[0134] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein by equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for diagnosing the health status of batteries in energy storage power stations based on big data, characterized in that: include: Collect multimodal operation data at the PACK and cell levels of the energy storage power station battery, homomorphically encrypt the multimodal operation data to obtain ciphertext data, analyze and calculate the ciphertext data based on the Transformer self-attention data fusion mechanism, adaptively extract explicit and implicit correlation features, and obtain fused time series data; Based on the fused time series data, a battery health diagnosis model is constructed. Through the convolution operation of the quantum walk convolution kernel, the quantum entanglement features of multiple paths are extracted. Combined with the self-evolving quantum attention mechanism, the phase and amplitude of the quantum bit are dynamically adjusted, and the combination weights of different feature subspaces are explored to generate diagnostic feature enhanced representations. Based on the diagnostic feature enhanced representations, the entangled storage of long-term and short-term memory is realized through quantum circuits, and combined with gated units, the health state sequence is predicted. Based on the predicted health state sequence, the battery health prediction result is determined, and a quantum reinforcement learning algorithm is constructed. By combining quantum state superposition and quantum parallelism, the diagnosis strategy combination space is searched online. In the loop of the quantum reinforcement learning algorithm, a multi-order causal relationship tensor model is embedded to model the causal relationship strength between battery performance indicators, determine the battery health degradation indicator, generate a causal diagnosis map, and obtain the battery health state diagnosis result; Based on the fused time series data, a battery health diagnosis model is constructed. Through the convolution operation of the quantum walk convolution kernel, the quantum entanglement features of multiple paths are extracted. Combined with the self-evolving quantum attention mechanism, the phase and amplitude of the quantum bit are dynamically adjusted, and the combination weights of different feature subspaces are explored to generate diagnostic feature enhanced representation. Based on the diagnostic feature enhanced representation, the entangled storage of long-term and short-term memory is realized through quantum circuits. Combined with the gate control unit, the health state sequence is predicted, including: Construct a quantum walk convolution kernel, map the fused time series data to the quantum bit space, each time step corresponds to a quantum bit, create a translation operator and a coin operator, generate a quantum walk convolution kernel, and perform a convolution operation in the quantum bit space through the quantum walk convolution kernel. The translation operator controls the movement of the quantum bit in space, and the coin operator controls the superposition state of the quantum bit. By adjusting the movement parameters corresponding to the translation operator and the superposition parameters corresponding to the coin operator, the step size and direction of the quantum walk are controlled, and the quantum entanglement characteristics are extracted; The quantum entangled features are input into a self-evolving quantum attention module based on the attention mechanism. Through quantum gate operations, the phase and amplitude of the quantum bit are dynamically adjusted to explore the combined weights of different feature subspaces. For each quantum entangled feature, quantum gate operations are performed according to a set of preset parameters to obtain a primary feature representation. The parameters corresponding to the quantum gate operations are adjusted in combination with the self-evolution algorithm, and the diagnostic feature enhanced representation is generated according to the self-attention allocation. Input the diagnostic feature enhancement representation into a pre-built quantum long short-term memory circuit, and perform entanglement storage and update on the timing information corresponding to the diagnostic feature enhancement representation. The quantum long short-term memory circuit is composed of an input gate, a forget gate, an output gate, and a state update gate. The input gate controls the new diagnostic feature enhancement representation to enter the memory unit, the forget gate controls the forgetting of historical information, the output gate controls the influence of the memory unit on the current state, and the state update gate updates the state of the memory unit; The output of the quantum long short-term memory circuit is entangled with the diagnostic feature enhancement representation through a parameterized quantum circuit to obtain a predicted state. The predicted state is continuously generated by recursively calling the quantum long short-term memory circuit. Each predicted state is measured in combination with a quantum measurement operation, and the predicted state quantum state is converted into a classical state to generate a health state prediction sequence. The method further comprises: The quantum walk convolution kernel extracts quantum entanglement features through multi-step quantum walks. The formula is as follows: Among them, U walk represents the quantum walk convolution kernel, l represents the quantum walk step number, L represents the total number of quantum walk steps, S l () represents the translation operator, θ l represents the movement parameter, C l () represents the hidden operator, α l represents the superposition parameter, I represents the identity matrix, Represents a tensor product operation; The output of the quantum long short-term memory circuit is entangled with the diagnostic feature enhancement representation through a parameterized quantum circuit to obtain the predicted states including: The predicted state is formulated as follows: Among them, |y t > represents the predicted quantum state of the health state at the tth time step, p represents the quantum gate operation sequence, N p represents the total number of quantum gate operations, γ p,t represents the weight of the p-th quantum gate operation at the t-th time step, V p represents the pth parameterized quantum gate operation at the tth time step, λ p,t represents the parameter controlling the pth quantum gate operation at the tth time step, |h t > represents the hidden quantum state at the tth time step, |ψ′ t > represents the quantum entanglement characteristic representation after adjustment at the t-th time step; In the loop of the quantum reinforcement learning algorithm, a multi-order causal relationship tensor model is embedded to model the causal relationship strength between battery performance indicators, determine the battery health degradation indicators, generate a causal diagnosis map, and obtain the battery health status diagnosis results including: Perform high-order singular value decomposition on the battery performance indicators to generate multi-order tensor causal relationships between battery performance indicators. Based on the causal connection strength between the performance indicators, determine the causal strength tensor. Based on the causal discovery algorithm, set the loss function and combine the causal intervention regularization term to construct the causal graph structure between the performance indicators, generate a multi-order causal relationship tensor model, and embed the multi-order causal relationship tensor model in the loop of the quantum reinforcement learning algorithm. A quantum-classical hybrid optimization algorithm is used to search for battery health degradation indicators online, and a data-driven quantum causal enhancement method is used to extract high-order causal statistics, dynamically update the weights corresponding to battery health degradation indicators, and generate a combination of performance degradation indicators. Based on the combination of multi-order tensor causal relationships and performance degradation indicators, a causal diagnosis graph is constructed. The quantum representation of the causal diagnosis graph is learned using a pre-trained quantum graph neural network, and the semantic consistency of the graph is maintained through graph embedding regularization constraints. Based on the causal diagnostic map, the contribution of different indicators to the diagnostic results is quantified, and the probability distribution of the diagnostic results is determined based on the quantum density matrix. The current health diagnostic result is determined by selecting the diagnostic result corresponding to the maximum probability.

2. The method according to claim 1, characterized in that Collect multimodal operation data of the battery pack and cell level of the energy storage power station, perform homomorphic encryption on the multimodal operation data to obtain ciphertext data, analyze and calculate the ciphertext data based on the self-attention data fusion mechanism of Transformer, adaptively extract explicit and implicit correlation features, and obtain fused time series data including: Collect multimodal operation data at the PACK and cell levels of the energy storage power station battery, including voltage, current, temperature and state of charge. During the collection process, introduce edge computing devices based on the MQTT protocol to collect and preprocess the multimodal operation data in real time, and perform hash operations on the multimodal operation data through a pre-built blockchain data authentication mechanism, and store the hash value in the blockchain; Perform fully homomorphic encryption based on the lattice cryptographic system on multimodal operation data, convert the original data into encrypted ciphertext data, introduce a noise mechanism in the encryption process, encrypt the original data one by one based on the preset public key, and divide the preset private key into multiple shares through the key management protocol based on secure multi-party computing, and distribute them to multiple energy storage power stations, where the public key and secret key are distributed and updated through secure multi-party computing; The ciphertext data is input into the Transformer-based self-attention data fusion model, and the explicit and implicit correlation features between different modal data are calculated through the self-attention mechanism, wherein the self-attention data fusion model adopts an encoder-decoder architecture, the encoder introduces a position encoding mechanism to capture the position information of the ciphertext data, the encoder and the decoder adopt a multi-head attention mechanism to capture the correlation features between the ciphertext data from different subspaces, and the corresponding features are extracted from the ciphertext data and fused through a forward propagation algorithm based on fully homomorphic encryption to obtain fused time series ciphertext data; Based on the homomorphic decryption algorithm corresponding to the fully homomorphic encryption algorithm, the fused time series ciphertext data is homomorphically decrypted, and the ciphertext data is decrypted one by one using the private key. Combined with the GPU-based parallel decryption framework, the ciphertext data is transferred to the GPU in batches, and the decryption algorithm is executed in parallel on the GPU to obtain the fused time series decrypted data.

3. The method according to claim 1, characterized in that Based on the predicted health status sequence, the battery health prediction results are determined, and a quantum reinforcement learning algorithm is constructed. By combining quantum state superposition and quantum parallelism, the online search diagnostic strategy combination space includes: Map the performance indicators of the battery under different health states to the probability amplitude of the quantum state, construct a quantum representation of the battery health state, add the quantum phase as a hidden variable to the quantum state, and capture the nonlinear dynamic characteristics of the battery health degradation process; Based on quantum circuits, value iteration and strategy iteration algorithms are constructed. Strategy search is accelerated through quantum parallel computing. The optimal diagnostic strategy combination is located using quantum amplitude amplification. The diagnostic strategy combination space is sampled through quantum gate sequences. Based on the diagnostic strategy combination space, a search tree of diagnostic strategy combinations is dynamically generated, branches of the search tree are pruned based on heuristic knowledge, and adaptive step length exploration and utilization of the diagnostic strategy combination space is performed based on a quantum walk algorithm.

4. A big data-based energy storage power station battery health status diagnosis system, used to implement the method described in any one of claims 1 to 3, characterized in that: include: The first unit is used to collect multimodal operation data at the PACK level and the cell level of the energy storage power station battery, homomorphically encrypt the multimodal operation data to obtain ciphertext data, analyze and calculate the ciphertext data based on the self-attention data fusion mechanism of Transformer, adaptively extract explicit and implicit correlation features, and obtain fused time series data; The second unit is used to build a battery health diagnosis model based on the fused time series data, extract the quantum entanglement features of multiple paths through the convolution operation of the quantum walk convolution kernel, dynamically adjust the phase and amplitude of the quantum bit in combination with the self-evolving quantum attention mechanism, explore the combination weights of different feature subspaces, generate diagnostic feature enhanced representation, based on the diagnostic feature enhanced representation, realize the entanglement storage of long and short-term memory through quantum circuits, and combine with the gate control unit to predict the health state sequence; The third unit is used to determine the battery health prediction result based on the predicted health status sequence, construct a quantum reinforcement learning algorithm, and search the diagnostic strategy combination space online by combining quantum state superposition and quantum parallelism. In the loop of the quantum reinforcement learning algorithm, a multi-order causal relationship tensor model is embedded to model the causal relationship strength between battery performance indicators, determine the battery health degradation indicator, generate a causal diagnostic map, and obtain the battery health status diagnosis result.

5. An electronic device, characterized in that: include: processor; a memory for storing processor-executable instructions; The processor is configured to call the instructions stored in the memory to execute the method described in any one of claims 1 to 3.

6. A computer-readable storage medium having computer program instructions stored thereon, characterized in that: When the computer program instructions are executed by a processor, the method according to any one of claims 1 to 3 is implemented.

Images