All-vanadium redox flow battery fault prediction method and system

Abstract

The invention relates to an all-vanadium redox flow battery fault prediction method and system, and the method comprises the steps: collecting the operation data of an all-vanadium redox flow battery, and carrying out the preprocessing of the operation data of the all-vanadium redox flow battery; constructing a Deepseek-R1 model, wherein the model comprises a convolutional neural network layer, a long and short-term memory network layer and a fusion layer; training and verifying the model; preprocessing operation data, and inputting the preprocessed operation data into the model for fault prediction and fault type diagnosis; fault early warning is triggered according to the output prediction result, and the operation parameters of the battery are automatically adjusted according to needs; and fault data is fed back to the model, and the model is optimized through online learning, so that the accuracy of fault prediction is improved. According to the method, local and time-dependent features can be efficiently extracted, rapid response is realized based on real-time data streams, fault response time is shortened, fault types can be accurately distinguished, early warning and adjustment are realized, and system downtime is shortened.

Description

Technical Field

[0001] The present invention relates to the technical field of liquid flow batteries, and in particular to a fault prediction method and system for an all-vanadium liquid flow battery. Background Art

[0002] All-vanadium redox flow batteries are widely used in large-scale energy storage systems due to their high energy efficiency, long service life, and strong scalability. However, over extended periods of operation, flow batteries can experience gradual performance degradation or even serious failure, impacting their efficiency and service life. Promptly identifying and remediating potential failures is crucial to maintaining stable flow battery system operation.

[0003] Existing fault detection methods for all-vanadium flow batteries primarily rely on regular inspections, manual testing, and simple model-based early warning systems. The operating state and failure modes of flow batteries are typically highly complex and nonlinear. Most existing technologies rely on simple analysis of a small amount of data features and fail to fully process the multidimensional data in flow battery systems, resulting in insufficient data analysis depth. Traditional machine learning algorithms (such as support vector machines and decision trees) often fail to provide sufficiently high prediction accuracy when dealing with the nonlinear failure modes of flow batteries and are easily affected by noise and data fluctuations. Furthermore, existing algorithms often rely on pre-set thresholds and cannot flexibly adapt to battery conditions under different operating conditions.

[0004] Therefore, it is necessary to provide a new fault prediction method for all-vanadium redox flow batteries. Summary of the Invention

[0005] Based on the above-mentioned problems existing in the prior art, the purpose of the embodiments of the present invention is to provide a vanadium liquid flow battery fault prediction method and system that can efficiently extract local and time-dependent features, achieve rapid response based on real-time data streams, shorten fault response time, accurately distinguish fault types, achieve early warning and adjustment, and reduce system downtime.

[0006] To achieve the above-mentioned purpose, the technical solution adopted by the present invention is: a method for predicting faults of all-vanadium redox flow batteries, comprising:

[0007] Collecting and preprocessing the operating data of the all-vanadium redox flow battery;

[0008] Build the Deepseek-R1 model based on deep learning. The Deepseek-R1 model includes a convolutional neural network layer, a long short-term memory network layer, and a fusion layer. The convolutional neural network layer extracts local features from battery data, and the long short-term memory network layer captures the time dependency of the data to analyze long-term trends and temporal changes during battery operation.

[0009] Train and validate the Deepseek-R1 model based on deep learning;

[0010] The real-time collected operating data is pre-processed and then input into the trained Deepseek-R1 model for fault prediction and fault type diagnosis;

[0011] Trigger fault warnings based on predictions from the Deepseek-R1 model and automatically adjust battery operating parameters as needed;

[0012] Fault data is fed back into the Deepseek-R1 model, and the model is optimized through online learning to improve the accuracy of fault prediction.

[0013] Furthermore, the collecting of the operating data of the all-vanadium redox flow battery and preprocessing of the operating data of the all-vanadium redox flow battery include:

[0014] Use multi-dimensional sensors to collect operating data of all-vanadium redox flow batteries;

[0015] Perform data cleaning and standardization on the operating data of all-vanadium redox flow batteries;

[0016] Time series processing is used for the operating data of all-vanadium redox flow battery.

[0017] Furthermore, the convolutional neural network layer is used to extract local features in the battery data. The convolutional neural network layer can effectively identify local patterns in the data and enhance the model's sensitivity to data details.

[0018] The convolutional neural network layer extracts local abnormal features in the data layer by layer by constructing a multi-layer convolutional neural network;

[0019] The output layer receives preprocessed operating data, including voltage, current, temperature, pH value, flow rate, etc., and the operating data is organized as a two-dimensional tensor in the form of time series;

[0020] The convolutional layer uses a 3×3 convolution kernel to adapt to the feature extraction requirements of data of different scales; the step size is set to 1 or 2 to ensure that sufficient feature information is retained while improving computational efficiency; 32 filters are set in the first layer, and the number of filters increases to 64 in subsequent layers, and finally to 128 filters to extract deeper features layer by layer;

[0021] The activation function uses the ReLU function, and the ReLU activation function formula is:

[0022] ReLU(x)=max(0,x)

[0023] The ReLU activation function is used to introduce nonlinear mapping, which makes feature extraction nonlinear and avoids the gradient vanishing problem.

[0024] The pooling layer uses maximum pooling with a pooling window size of 2×2 and a step size of 2 to reduce the dimension and retain the main feature information. Finally, the feature matrix after convolution operation and pooling is input into the long short-term memory network layer for time series analysis.

[0025] Furthermore, the long short-term memory network layer is used to capture the time dependency of data and analyze the long-term trends and time series changes during battery operation. The long short-term memory network layer can process the failure modes that may occur in the battery during long-term operation and identify dynamic changes.

[0026] The long short-term memory network layer controls the flow of information through input gates, forget gates, and output gates;

[0027] The forget gate is used to determine how much past information the memory cell state should retain at the current time step. The forget gate calculation formula is:

[0028] f t =σ(W f ·[h t-1 , x t ]+b f )

[0029] Among them, f t is the output of the forget gate, W f is the weight of the forget gate, b f is the bias of the forget gate, h t-1 is the hidden state of the previous time step, x t is the input data of the current time step, σ is the sigmoid activation function;

[0030] The input gate is used to determine how much new information should be stored in the cell state at the current time step. The input gate calculation formula is:

[0031] i t =σ(W i ·[h t-1 , x t ]+b i )

[0032]

[0033] Among them, i t is the output of the input gate, σ is the sigmoid activation function, and W i The weight matrix of the input gate, h t-1 is the hidden state of the previous time step, x t is the input data of the current time step, bi is the bias vector of the input gate, is the candidate cell state value, tanh is the hyperbolic tangent activation function, W c is the weight matrix used to calculate the candidate cell state, b C is the bias vector used to calculate the candidate cell state;

[0034] State update is the core process of LSTM. Through the joint action of the forget gate and the input gate, the cell state transmits and updates information between time steps. The state update calculation formula is:

[0035]

[0036] Among them, C t is the cell state at the current time step, f t is the output of the forget gate, C t-1 is the cell state at the previous time step, i t is the output of the input gate, is the candidate cell state value;

[0037] The output gate is used to determine the hidden state output of the current time step. The output gate calculation formula is:

[0038] o t =σ(W o ·[h t-1 , x t ]+b o )

[0039] h t =o t *tanh(C t )

[0040] Among them, t is the output of the output gate, σ is the sigmoid activation function, W o is the weight matrix of the output gate, h t-1 is the hidden state of the previous time step, x t is the input data of the current time step, b o is the bias vector of the output gate, h t is the final hidden state of the current time step, tanh is the hyperbolic tangent activation function, C t is the cell state at the current time step.

[0041] Furthermore, the fusion layer performs residual fusion on the features extracted by the convolutional neural network layer and the long short-term memory network layer, comprehensively considering the static and dynamic characteristics of the battery system to form the final fault prediction result;

[0042] Feature splicing directly splices the static features extracted by the convolutional neural network layer and the dynamic features extracted by the long short-term memory network layer, and further processes them through the fully connected layer. It is suitable for data with low feature dimensions;

[0043] The fully connected layer performs further nonlinear transformation and feature fusion on the concatenated feature vectors. By adjusting the number of neurons and activation function of the fully connected layer, the expression ability and output form of the model can be controlled to adapt it to different fault prediction task requirements.

[0044] Furthermore, the training and validation of the Deepseek-R1 model based on deep learning involves training the Deepseek-R1 model using a large amount of historical operating data. During the training process, an optimization algorithm is used to adjust the model parameters to ensure that the model can accurately learn the behavior patterns of the battery system. The training data includes data on the battery in normal operating conditions and various fault conditions. The normal operating data is used to enable the model to learn the normal battery behavior patterns, while the data under fault conditions is used to train the model to identify fault characteristics.

[0045] Furthermore, the fault warning triggered by the prediction results output by the Deepseek-R1 model includes when the fault probability predicted by the model exceeds the set threshold, or when the real-time monitored parameters such as voltage, current, and temperature exceed the normal range, the system triggers a fault warning. The warning information is promptly notified to the operator through sound and light alarms, SMS notifications, system pop-up windows, etc., thereby reminding the operator to pay attention to the operating status of the battery system and take corresponding measures.

[0046] Furthermore, the automatic adjustment of the battery's operating parameters as needed includes automatically reducing the charging power or stopping charging when the battery temperature is too high to avoid overheating damage; automatically adjusting the current output when the battery current is abnormal to prevent battery damage caused by excessive current; for abnormal electrolyte conditions, the system prompts the operator to detect and adjust the electrolyte composition, and automatically adjusts the charge and discharge strategy according to the pH value and concentration of the electrolyte, reducing the charge and discharge depth or limiting the charging current to slow down further deterioration of the electrolyte; for abnormal fluidity, the pump speed is automatically adjusted to optimize the electrolyte flow rate, ensure the uniformity of ion transmission inside the battery, and restore normal operating conditions.

[0047] Furthermore, the method of feeding fault data back to the Deepseek-R1 model and optimizing the model through online learning to improve the accuracy of fault prediction includes adopting an online learning mechanism. When a fault occurs, the system will feed back fault data in real time and incorporate the fault data into the Deepseek-R1 model retraining process to optimize the accuracy and robustness of the model.

[0048] A vanadium redox flow battery fault prediction system is applied to the above-mentioned vanadium redox flow battery fault prediction method, and the system comprises:

[0049] Data acquisition module, used to collect operating data of all-vanadium redox flow batteries;

[0050] Data preprocessing module, used to preprocess the operating data of the all-vanadium liquid flow battery;

[0051] The modeling module is used to build the Deepseek-R1 model based on deep learning. The Deepseek-R1 model includes a convolutional neural network layer, a long short-term memory network layer, and a fusion layer. The convolutional neural network layer extracts local features from battery data, and the long short-term memory network layer captures the time dependency of the data to analyze long-term trends and temporal changes during battery operation.

[0052] Model training and verification module, used to train and verify the Deepseek-R1 model based on deep learning;

[0053] The fault prediction module is used to pre-process the real-time collected operating data and input it into the trained Deepseek-R1 model for fault prediction and fault type diagnosis;

[0054] The parameter adjustment module is used to trigger fault warnings based on the prediction results output by the Deepseek-R1 model and automatically adjust the battery's operating parameters as needed;

[0055] The optimization learning module is used to feed fault data back to the Deepseek-R1 model, optimize the model through online learning, and improve the accuracy of fault prediction.

[0056] The beneficial effects of the present invention are as follows: a method for predicting faults of an all-vanadium liquid flow battery of the present invention comprises: collecting operating data of the all-vanadium liquid flow battery and preprocessing the operating data of the all-vanadium liquid flow battery; constructing a Deepseek-R1 model based on deep learning, the Deepseek-R1 model comprising a convolutional neural network layer, a long short-term memory network layer and a fusion layer, extracting local features in the battery data through the convolutional neural network layer, capturing the time dependency of the data through the long short-term memory network layer, and analyzing the long-term trend and timing changes during the battery operation process; training and verifying the Deepseek-R1 model based on deep learning; inputting the preprocessed operating data collected in real time into the trained Deepseek-R1 model for fault prediction and fault type diagnosis; triggering a fault warning according to the prediction result output by the Deepseek-R1 model, and automatically adjusting the operating parameters of the battery as needed; feeding the fault data back to the Deepseek-R1 model, optimizing the model through online learning, and improving the accuracy of fault prediction. The all-vanadium liquid flow battery fault prediction method of the present invention combines convolutional neural networks and long short-term memory networks to efficiently extract local and time-dependent features, improve the accuracy of fault diagnosis, achieve rapid response based on real-time data streams, shorten fault response time, and can accurately distinguish fault types such as electrode failure. It also automatically adjusts operating parameters in combination with reinforcement learning, extending the average service life of the battery; real-time monitoring reduces manual intervention, reduces maintenance costs, achieves early warning and adjustment, and reduces system downtime. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] The present invention will be further described below with reference to the accompanying drawings and examples.

[0058] In the picture:

[0059] Figure 1 A flowchart of a method for predicting faults of an all-vanadium redox flow battery provided by a first embodiment of the present invention;

[0060] Figure 2 A schematic structural diagram of the Deepseek-R1 model provided for the first embodiment of the present invention;

[0061] Figure 3 A schematic diagram of the modules of the all-vanadium redox flow battery fault prediction system provided by the second embodiment of the present invention;

[0062] Figure 4 It is a structural diagram of a network-side server provided according to a third embodiment of the present invention. DETAILED DESCRIPTION

[0063] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of them. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0064] First embodiment:

[0065] A first embodiment of the present invention provides a method for predicting faults of an all-vanadium liquid flow battery, comprising: collecting operating data of the all-vanadium liquid flow battery and preprocessing the operating data of the all-vanadium liquid flow battery; constructing a Deepseek-R1 model based on deep learning, the Deepseek-R1 model comprising a convolutional neural network layer, a long short-term memory network layer and a fusion layer, extracting local features from the battery data through the convolutional neural network layer, capturing the time dependency of the data through the long short-term memory network layer, and analyzing long-term trends and timing changes during battery operation; training and verifying the Deepseek-R1 model based on deep learning; inputting the preprocessed operating data collected in real time into the trained Deepseek-R1 model for fault prediction and fault type diagnosis; triggering a fault warning based on the prediction result output by the Deepseek-R1 model, and automatically adjusting the operating parameters of the battery as needed; feeding the fault data back to the Deepseek-R1 model, optimizing the model through online learning, and improving the accuracy of fault prediction. The all-vanadium liquid flow battery fault prediction method of the present invention combines convolutional neural networks and long short-term memory networks to efficiently extract local and time-dependent features, improve the accuracy of fault diagnosis, achieve rapid response based on real-time data streams, shorten fault response time, and can accurately distinguish fault types such as electrode failure. It also automatically adjusts operating parameters in combination with reinforcement learning, extending the average service life of the battery; real-time monitoring reduces manual intervention, reduces maintenance costs, achieves early warning and adjustment, and reduces system downtime.

[0066] The following is a detailed description of the implementation details of the all-vanadium liquid flow battery fault prediction method of this embodiment. The following content is only for the convenience of understanding the implementation details, and is not necessary for the implementation of this solution. The specific process of this embodiment is as follows: Figure 1 shown.

[0067] Step S1: collecting and preprocessing the operating data of the all-vanadium redox flow battery.

[0068] Specifically, the steps of collecting the operating data of the all-vanadium redox flow battery and preprocessing the operating data of the all-vanadium redox flow battery include:

[0069] Step S11: using a multi-dimensional sensor to collect operating data of the all-vanadium redox flow battery.

[0070] Specifically, key operating parameters of the all-vanadium redox flow battery are collected in real time through voltage and current sensors, temperature sensors, pH sensors, flow rate sensors, and battery internal resistance sensors to construct a multidimensional data set. Voltage and current sensors are used to collect the battery's voltage and current values ​​for real-time monitoring of changes in the input and output voltage and current of the all-vanadium redox flow battery. A temperature sensor is used to monitor temperature changes during battery operation to prevent failures caused by overheating or temperature fluctuations. A pH sensor is used to monitor the pH value of the battery electrolyte to determine the stability of the battery's chemical reactions. A flow rate sensor is used to monitor the flow rate of the battery electrolyte to determine the flow state of the liquid circulation inside the battery. A battery internal resistance sensor is used to collect changes in the battery's internal resistance in real time to reflect the contact state between the electrode and the electrolyte.

[0071] Step S12: performing data cleaning and standardization on the operating data of the all-vanadium redox flow battery.

[0072] Specifically, the Kalman filter algorithm is used to dynamically filter the sensor noise, remove the noise in the sensor data, and ensure the accuracy of the input data.

[0073] As an example, in voltage and current data, the filtered data can more clearly show the voltage and current curves of the battery's normal charging and discharging, eliminating small fluctuations caused by sensor accuracy limitations and environmental interference, making fault characteristics easier to capture.

[0074] Standardization processing involves normalizing data of different dimensions so that all parameters are within the same range, avoiding excessive influence of certain parameters on the prediction results.

[0075] For example, the Min-Max normalization method is used to normalize the data. All data are linearly transformed to the interval [0, 1], eliminating the influence of dimension and making each parameter equally important in the model training and prediction process.

[0076] As an example, voltage data typically varies within the range of tens of volts, while pH data ranges from 0 to 14. After normalization, both voltage data and pH values ​​can be compared and analyzed on the same scale, which improves the convergence speed and prediction accuracy of the model.

[0077] Furthermore, during the data transmission or collection process, data may be missing. The missing parts of the data can be filled through interpolation or other methods to ensure data integrity.

[0078] Step S13: Time series processing is performed on the operating data of the all-vanadium redox flow battery.

[0079] Specifically, the sliding window method is used to segment the data and divide the data into several time windows. The size of the sliding window is set according to the fault characteristics and operating rules of the battery system. Through sliding window processing, the original long sequence data is divided into multiple short sequence fragments. Each fragment contains the operating status information of the battery over a period of time, which is then used as the input sample of the model. This enables the model to better capture the temporal changes and dynamic characteristics of the battery status, provide a more accurate basis for fault prediction, and facilitate the subsequent dynamic analysis of the model.

[0080] Step S2: Build a Deepseek-R1 model based on deep learning. The Deepseek-R1 model includes a convolutional neural network layer (CNN layer), a long short-term memory network layer (LSTM layer), and a fusion layer. The convolutional neural network layer is used to extract local features in the battery data, and the long short-term memory network layer is used to capture the time dependency of the data to analyze the long-term trends and time series changes during battery operation.

[0081] Specifically, if Figure 2 As shown, the specific steps of constructing the Deepseek-R1 model include:

[0082] In step S21, a convolutional neural network layer (CNN layer) is used to extract local features (such as instantaneous voltage fluctuations, current mutations, etc.) from the battery data. The CNN layer can effectively identify local patterns in the data and enhance the model's sensitivity to data details.

[0083] Specifically, by constructing a multi-layer convolutional neural network, local abnormal features in the data are extracted layer by layer.

[0084] Among them, the output layer receives preprocessed operating data, which includes voltage, current, temperature, pH value, flow rate, etc., and the operating data is organized as a two-dimensional tensor in the form of a time series.

[0085] The convolutional layer uses a 3×3 convolution kernel to adapt to the feature extraction requirements of data of varying sizes. The step size is set to 1 or 2 to ensure sufficient feature information while improving computational efficiency. The first layer uses 32 filters, increasing to 64 filters in subsequent layers and finally to 128 filters to extract deeper features layer by layer.

[0086] The activation function uses the ReLU function, and the ReLU activation function formula is:

[0087] ReLU(x)=max(0,x)

[0088] The nonlinear mapping is introduced through the ReLU activation function, which makes the feature extraction nonlinear and avoids the gradient disappearance problem.

[0089] The pooling layer uses maximum pooling with a pooling window size of 2×2 and a stride of 2 to reduce dimensionality and retain the main feature information.

[0090] Finally, the feature matrix after convolution and pooling is input into the long short-term memory network layer (LSTM layer) for time series analysis.

[0091] In step S22, the long short-term memory network layer (LSTM layer) is used to capture the time dependency of the data and analyze the long-term trends and timing changes during battery operation. The LSTM layer can process possible failure modes of the battery during long-term operation and identify dynamic changes.

[0092] Specifically, the LSTM layer controls the flow of information through the input gate, forget gate, and output gate to solve the gradient vanishing problem of the traditional RNN layer.

[0093] The forget gate is used to determine how much past information the memory cell state should retain at the current time step. The forget gate calculation formula is:

[0094] f t =σ(W f ·[h t-1 , x t ]+b f )

[0095] Among them, f t is the output of the forget gate, W f is the weight of the forget gate, b f is the bias of the forget gate, h t-1 is the hidden state of the previous time step, x t is the input data of the current time step, and σ is the sigmoid activation function.

[0096] The input gate is used to determine how much new information should be stored in the cell state at the current time step. The input gate calculation formula is:

[0097] i t =o(W i ·[h t-1 , x t ]+b i )

[0098]

[0099] Among them, i t is the output of the input gate, σ is the sigmoid activation function, and W i The weight matrix of the input gate, h t-1 is the hidden state of the previous time step, x t is the input data of the current time step, b i is the bias vector of the input gate, is the candidate cell state value, tanh is the hyperbolic tangent activation function, W C is the weight matrix used to calculate the candidate cell state, b C is the bias vector used in calculating the candidate cell state.

[0100] State update is the core process of LSTM. Through the joint action of the forget gate and the input gate, the cell state transmits and updates information between time steps. The state update calculation formula is:

[0101]

[0102] Among them, C t is the cell state at the current time step, f t is the output of the forget gate, C t-1 is the cell state at the previous time step, i t is the output of the input gate, is the candidate cell state value.

[0103] The output gate is used to determine the hidden state output of the current time step. The output gate calculation formula is:

[0104] o t =σ(W o ·[h t-1 , x t ]+b o )

[0105] h t =o t *tanh(C t )

[0106] Among them, t is the output of the output gate, σ is the sigmoid activation function, W o is the weight matrix of the output gate, h t-1 is the hidden state of the previous time step, x t is the input data of the current time step, b o is the bias vector of the output gate, h t is the final hidden state of the current time step, tanh is the hyperbolic tangent activation function, C t is the cell state at the current time step.

[0107] Furthermore, the input and output parameters of the LSTM layer include: The time series features output by the CNN layer serve as the LSTM input. The number of LSTM units is 64 or 128 to enhance the model's memory capacity. Activation functions include tanh as the input activation function and sigmoid as the gated unit activation function to improve time series modeling capabilities. Dropout (0.2-0.5) is used for regularization to prevent overfitting and improve model generalization. The output of the LSTM layer is the extracted time series features, which are input to the fusion layer for final fault prediction.

[0108] In step S23, the fusion layer performs residual fusion on the features extracted by the CNN layer and the LSTM layer, comprehensively considering the static and dynamic characteristics of the battery system to form the final fault prediction result.

[0109] Specifically, feature splicing directly splices the static features extracted by the CNN layer and the dynamic features extracted by the LSTM layer, and further processes them through the fully connected layer. It is suitable for data with lower feature dimensions.

[0110] The fully connected layer, as an important component of the fusion layer, performs further nonlinear transformations and feature fusion on the concatenated feature vectors. By adjusting the number of neurons and activation functions in the fully connected layer, the model's expressive power and output form can be controlled to adapt to different fault prediction tasks.

[0111] As an example, in a binary classification problem (normal / faulty), the concatenated feature vector is input into the fully connected layer, processed by one or two layers of neurons (the number of neurons is adjusted based on the experiment), and finally outputs a fault probability value through the Sigmoid activation function, ranging from [0, 1]. A probability value greater than 0.5 is considered a fault state.

[0112] In the multi-classification problem (distinguishing different fault types), the number of neurons in the fully connected layer is set to the number of fault types N, and the Softmax activation function is used to output the probability distribution of each type of fault. The category with the highest probability is the predicted fault type.

[0113] Step S3: training and verifying the Deepseek-R1 model based on deep learning.

[0114] Specifically, the training and validation of the deep learning-based Deepseek-R1 model involves training the model using a large amount of historical operating data. During the training process, optimization algorithms (such as the Adam optimizer) are used to adjust the model parameters to ensure that the model accurately learns the behavior of the battery system. The training data includes battery status information before and after the fault occurs, ensuring that the model can accurately predict under various operating conditions.

[0115] Furthermore, the training data includes data from the battery under normal operating conditions and various fault conditions. Normal operating data is used by the model to learn the normal battery behavior patterns, while data from fault conditions is used to train the model to identify fault characteristics. Data sources include laboratory simulated fault experiments and historical fault records from actual battery operation. Data is preprocessed and labeled to ensure data quality and label accuracy. The Adam optimizer is used for model training. The Adam optimizer combines the advantages of momentum and adaptive learning rates, enabling rapid convergence and effective optimization of model parameters. The mean squared error function is selected as the loss function to measure the difference between the model's predicted value and the true value. During training, the gradient is calculated using the backpropagation algorithm, and the model parameters are updated using the Adam optimizer to gradually minimize the loss function value and improve the model's prediction accuracy. The ratio of the training set to the validation set is set to 8:2. The model is regularly evaluated using the validation set to prevent overfitting. During training, the optimal model is saved based on the performance indicators on the validation set.

[0116] During long-term operation, all-vanadium flow batteries experience complex local fluctuations and long-term trends in their operating data (such as voltage, current, temperature, pH, flow rate, etc.). Traditional methods often struggle to capture these detailed features and dynamic changes simultaneously, resulting in insufficient fault prediction accuracy. This example combines a convolutional neural network (CNN) with a long short-term memory network (LSTM) to form the Deepseek-R1 model based on deep learning, leveraging the advantages of CNN and LSTM in addressing the complexity and nonlinear characteristics of all-vanadium flow battery data.

[0117] The advantages of CNNs include: During the operation of all-vanadium redox flow batteries, local phenomena such as transient voltage fluctuations and current surges are often early signs of potential failures. CNNs, through convolution operations, can automatically capture these local features in the raw data. Their filters can extract local patterns, thereby increasing sensitivity to subtle abnormal signals, which is critical for accurate fault identification. Furthermore, CNNs utilize a mechanism for sharing convolution kernel parameters, which not only significantly reduces the number of model parameters but also avoids overfitting, enabling the model to generalize better when processing multidimensional sensor data. This characteristic makes CNNs particularly suitable for processing high-dimensional data from all-vanadium redox flow batteries.

[0118] The advantages of LSTM include: the operating status of all-vanadium liquid flow batteries is not only affected by instantaneous factors, but also manifests itself in long-term trends and cyclical changes; LSTM has a "memory" capability that can retain long-term time series information and capture the gradual changes in all-vanadium liquid flow battery performance over time; this is particularly important for predicting the progressive failure and cyclical degradation of all-vanadium liquid flow batteries.

[0119] Therefore, by combining the advantages of CNN and LSTM, the Deepseek-R1 model can simultaneously capture local anomalies and long-term dynamic changes in all-vanadium liquid flow battery data, significantly improving the accuracy and real-time performance of fault prediction, and providing strong support for preventing battery failures and extending battery life. Compared with traditional methods, the Deepseek-R1 model framework not only has obvious advantages in data processing capabilities, but can also continuously optimize prediction performance through online learning, fully meeting the needs of all-vanadium liquid flow batteries in practical applications.

[0120] In step S4, the real-time collected operating data is pre-processed and then input into the trained Deepseek-R1 model for fault prediction and fault type diagnosis.

[0121] Specifically, the battery operation data collected in real time by multi-dimensional sensors is preprocessed, divided into time series segments using a sliding window method, and input into the trained Deepseek-R1 model. The model uses its learned feature extraction and fusion capabilities to dynamically analyze the real-time operation data. When the battery status changes, the model can respond quickly and provide fault predictions.

[0122] Furthermore, an appropriate fault judgment threshold is set based on the battery's normal operating range and historical fault data. When the predicted value exceeds the set threshold, a fault alarm is triggered. Common fault types include: electrode failure, electrolyte abnormalities, and fluidity abnormalities.

[0123] Electrode failure is a condition in all-vanadium redox flow batteries, where electrode materials gradually age over time, leading to localized voltage fluctuations and current spikes. In this embodiment, when battery electrodes become damaged or degraded, causing battery performance to decline or become unchargeable, the model can detect these characteristics and issue a timely warning. In practical applications, when the model detects abnormal fluctuations in voltage and current data, such as significant deviations from normal values, it automatically triggers a fault warning signal, alerting operators to potential electrode degradation.

[0124] When an all-vanadium redox flow battery is operating, electrolyte anomalies can cause deviations in electrolyte concentration and pH due to repeated cycles, affecting reaction efficiency and leading to reduced battery reaction efficiency. This application model can identify electrolyte anomalies by analyzing electrolyte data. By setting a judgment threshold for the all-vanadium redox flow battery's historical operating data, the model can provide early warnings when it detects that the pH value of the all-vanadium redox flow battery deviates from the normal value. It can also provide corresponding diagnostic information and treatment plans, thus providing strong support for the system to take timely maintenance measures and extend the battery life.

[0125] Abnormal fluidity refers to abnormal liquid flow inside the battery during the operation of the all-vanadium liquid flow battery, which can also lead to uneven ion transmission, thereby reducing the overall energy conversion efficiency. The model of this application can determine abnormal fluidity by monitoring the flow rate changes. When the model detects abnormal flow rate in a channel of the all-vanadium liquid flow battery, such as blockage or electrolyte precipitation causing abnormal fluidity, the model can automatically trigger an alert to remind the operator of the current abnormal fluidity problem of the all-vanadium liquid flow battery. In serious cases, the operation of the all-vanadium liquid flow battery should be shut down in time to avoid damage to all all-vanadium liquid flow electrolytes and related all-vanadium liquid flow battery equipment.

[0126] As an example, for a binary classification problem, the model outputs a probability value between 0 and 1, indicating the possibility that the battery is in a faulty state. When the probability value exceeds a preset threshold (such as 0.5), the battery is judged to be faulty. For multi-classification problems, the model outputs the probability distribution of each type of fault, and the category with the highest probability is the predicted fault type. If the model outputs a probability of electrode failure of 0.7, a probability of electrolyte abnormality of 0.2, and a probability of fluidity abnormality of 0.1, it is judged that the battery may currently have an electrode failure. In this way, the model can accurately predict the fault state and type of the battery in real time, providing a basis for subsequent fault response.

[0127] Step S5: triggering a fault warning based on the prediction result output by the Deepseek-R1 model, and automatically adjusting the operating parameters of the battery as needed.

[0128] Specifically, when the model's predicted failure probability exceeds a set threshold, or when real-time monitored parameters such as voltage, current, and temperature exceed normal ranges, the system triggers a fault warning. This warning is promptly communicated to operators through audio and visual alarms, text message notifications, and system pop-up windows, prompting them to pay attention to the battery system's operating status and take appropriate measures. By continuously monitoring the battery's real-time status, the system automatically adjusts the battery's operating parameters when a potential fault is predicted, delaying its occurrence.

[0129] For example, when the battery temperature is too high, the system can automatically reduce the charging power or stop charging to avoid overheating damage. When the battery current is abnormal, the system can automatically adjust the current output to prevent excessive current from causing battery damage. In the event of electrolyte abnormalities, the system prompts the operator to test and adjust the electrolyte composition, and automatically adjusts the charge and discharge strategy based on the pH value and concentration of the electrolyte, such as reducing the charge and discharge depth, limiting the charging current, etc., to slow down further deterioration of the electrolyte. In the event of fluidity abnormalities, the system automatically adjusts the pump speed to optimize the electrolyte flow rate, ensure the uniformity of ion transmission within the battery, and restore normal operation.

[0130] Furthermore, the system combines reinforcement learning strategies to continuously optimize the battery's operating strategy based on historical operating data and real-time fault prediction results, extend the battery life, and ensure that effective measures are taken before the battery fails.

[0131] For example, when the system detects that the battery temperature is too high, it automatically reduces the charging power or suspends charging to reduce heat generation. When a current anomaly is detected, the system automatically adjusts the current output to keep the current value within a safe range to prevent overcurrent from damaging the battery. In the event of electrolyte anomalies, the system prompts the operator to test and adjust the electrolyte composition, and automatically adjusts the charge and discharge strategy based on the pH value and concentration of the electrolyte, such as reducing the charge and discharge depth and limiting the charging current, to slow down further deterioration of the electrolyte. In the event of fluidity anomalies, the system automatically adjusts the pump speed to optimize the electrolyte flow rate, ensure the uniformity of ion transmission within the battery, and restore normal operation.

[0132] In step S6, the fault data is fed back to the Deepseek-R1 model, and the model is optimized through online learning to improve the accuracy of fault prediction.

[0133] Specifically, using an online learning mechanism, the system provides real-time feedback on fault data when a fault occurs. This data is then incorporated into the Deepseek-R1 model retraining process to optimize the model's accuracy and robustness. Through continuous iteration and optimization, the Deepseek-R1 model is able to continuously adapt to the operating characteristics of batteries in different environments, improving the accuracy of long-term predictions.

[0134] Specifically, the relevant data for each fault event is recorded in detail, including operating parameters such as voltage, current, temperature, pH value, flow rate, and so on before and after the fault occurs, as well as information such as the fault type, occurrence time, and treatment measures. This data is sent to the Deepseek-R1 model via feedback and added to the training set as new training samples. The Deepseek-R1 model is retrained using the newly added fault data, and the model parameters are updated through an online learning mechanism. During the retraining process, the Adam optimizer and mean squared error loss function are continued to be used to fine-tune the model, making it more adaptable to the changing fault characteristics of batteries under different operating conditions. With the continuous accumulation of fault data and continuous optimization of the model, the prediction accuracy and robustness of the Deepseek-R1 model continue to improve, enabling more accurate prediction of various fault types and providing stronger guarantees for the stable operation of the battery system.

[0135] A method for predicting faults of an all-vanadium liquid flow battery of the present invention comprises: collecting operating data of the all-vanadium liquid flow battery and preprocessing the operating data of the all-vanadium liquid flow battery; constructing a Deepseek-R1 model based on deep learning, wherein the Deepseek-R1 model comprises a convolutional neural network layer, a long short-term memory network layer and a fusion layer, extracting local features in the battery data through the convolutional neural network layer, capturing the time dependency of the data through the long short-term memory network layer, and analyzing the long-term trend and time series changes during the battery operation; training and verifying the Deepseek-R1 model based on deep learning; inputting the preprocessed operating data collected in real time into the trained Deepseek-R1 model for fault prediction and fault type diagnosis; triggering a fault warning according to the prediction result output by the Deepseek-R1 model, and automatically adjusting the operating parameters of the battery as needed; feeding the fault data back to the Deepseek-R1 model, optimizing the model through online learning, and improving the accuracy of fault prediction. The all-vanadium liquid flow battery fault prediction method of the present invention combines convolutional neural networks and long short-term memory networks to efficiently extract local and time-dependent features, improve the accuracy of fault diagnosis, achieve rapid response based on real-time data streams, shorten fault response time, and can accurately distinguish fault types such as electrode failure. It also automatically adjusts operating parameters in combination with reinforcement learning, extending the average service life of the battery; real-time monitoring reduces manual intervention, reduces maintenance costs, achieves early warning and adjustment, and reduces system downtime.

[0136] Specific examples are:

[0137] Example 1

[0138] The specific steps of the Deepseek-R1 model training and optimization process include:

[0139] Step S1, obtaining the training data of the Deepseek-R1 model, where the training data comes from historical operation data and real-time collected data.

[0140] Specifically, at least one year of operating data is collected from the all-vanadium liquid flow battery system, and the data includes parameters such as voltage, current, temperature, pH value and flow rate under normal operating conditions.

[0141] Fault data includes data on battery failure types such as electrode failure, electrolyte problems, and abnormal fluidity. The dataset for each fault event should include operating data before and after the fault occurs.

[0142] The fault event label includes the fault type (such as electrode failure, electrolyte abnormality, etc.) and the time when the fault occurred.

[0143] Step S2: training the Deepseek-R1 model.

[0144] Specifically, the Deepseek-R1 model uses a combination of CNN and LSTM layers. The CNN layer extracts local features from the data, while the LSTM layer captures temporal dependencies. The model structure is shown in the framework diagram of the vanadium redox flow battery fault prediction method based on the Deepseek-R1 model. The input layer contains multiple input nodes, corresponding to various battery monitoring data. The CNN layer uses a convolution kernel size of 3, a convolution stride of 1, and 64 filters to extract local features. The LSTM layer uses 128 neurons to capture time series changes. The fusion layer combines the outputs of the CNN and LSTM layers to produce the fault prediction results.

[0145] The Adam optimizer was used for model training, and the mean squared error (MSE) loss function was used. The training set and validation set ratio was 8:2. A batch size of 32 was used for training, and 500 iterations were performed per batch.

[0146] During the training process, the model's loss value gradually decreased, the prediction accuracy reached more than 85%, and the fault diagnosis precision and recall rate both exceeded 80%.

[0147] Step S3: Online learning and optimization of the Deepseek-R1 model.

[0148] Specifically, when an actual fault occurs, the system feeds the fault data back into the model and uses this data for retraining. The system records detailed data for each fault event, including battery status parameters at the time of the fault, fault type, and treatment measures. This fault data is then added to the training set, and the Deepseek-R1 model is retrained. This model is updated through online learning mechanisms, improving its adaptability and accuracy.

[0149] Second embodiment:

[0150] like Figure 3 As shown, the second embodiment of the present invention provides a vanadium liquid flow battery fault prediction system, which includes: a data acquisition module 201, a data preprocessing module 202, a modeling module 203, a model training and verification module 204, a fault prediction module 205, a parameter adjustment module 206, and an optimization learning module 207.

[0151] Specifically, the data acquisition module 201 is used to collect the operating data of the all-vanadium redox flow battery; the data preprocessing module 202 is used to preprocess the operating data of the all-vanadium redox flow battery; the modeling module 203 is used to build the Deepseek-R1 model based on deep learning. The Deepseek-R1 model includes a convolutional neural network layer, a long short-term memory network layer and a fusion layer. The convolutional neural network layer is used to extract local features in the battery data, and the long short-term memory network layer is used to capture the time dependency of the data to analyze the long-term trend and time series changes during the battery operation process; the model training and verification module 204 is used to Used to train and verify the deep learning-based Deepseek-R1 model; the fault prediction module 205 is used to input the real-time collected operating data into the trained Deepseek-R1 model after preprocessing for fault prediction and fault type diagnosis; the parameter adjustment module 206 is used to trigger fault warnings based on the prediction results output by the Deepseek-R1 model and automatically adjust the battery operating parameters as needed; the optimization learning module 207 is used to feed fault data back to the Deepseek-R1 model and optimize the model through online learning to improve the accuracy of fault prediction.

[0152] Specific examples are:

[0153] Example 3

[0154] The data acquisition module 201 in the vanadium redox flow battery fault prediction system includes a voltage sensor, a current sensor, a temperature sensor, a pH sensor, and a flow rate sensor. Each sensor is connected to a data acquisition device via serial communication or an I2C bus, collecting real-time data and sending it to a central control system.

[0155] The voltage sensor has a range of 0-100V and an accuracy of ±0.1V. The current sensor has a range of 0-100A and an accuracy of ±0.5A. The temperature sensor has a range of -40°C to 100°C and an accuracy of ±0.5°C. The pH sensor has a range of 0-14 and an accuracy of ±0.01pH. The flow rate sensor has a range of 0-10L / min and an accuracy of ±0.1L / min.

[0156] The data preprocessing module 202 includes preprocessing algorithm modules such as data cleaning, denoising, and normalization to process the raw data received from the data acquisition module. The data preprocessing code is written in Python, using Kalman filtering to remove data noise and Min-Max normalization to standardize the data.

[0157] Kalman filtering reduces noise through two steps: prediction and update. The calculation formula of the prediction step is:

[0158]

[0159] in, is the predicted state at the current moment, A is the state transfer matrix, is the predicted state of the previous moment, B is the control input matrix, u k is the control vector, is the predicted covariance matrix, P k-1 is the covariance matrix predicted at the previous moment, A T is the transposed matrix of the state transfer matrix A, and Q is the process noise covariance matrix.

[0160] The update step is calculated as:

[0161]

[0162] Among them, K k is the Kalman gain, H is the observation matrix, H T is the transposed matrix of the observation matrix H, z k is the observed value, R is the measurement noise covariance matrix, P k is the updated covariance matrix.

[0163] Min-Max normalization is a linear transformation of data so that its value range is within a specified range, usually [0,1]. The normalization calculation formula is:

[0164]

[0165] Among them, X is the original data, X min is the minimum value in the data, X max is the maximum value in the data, X norm The data are normalized.

[0166] Standardization (Z-Score) is to transform data into a standard normal distribution with a mean of 0 and a variance of 1. It is often used to deal with data scale issues. The standardization calculation formula is:

[0167]

[0168] Among them, X is the original data, μ is the mean of the data, σ is the standard deviation of the data, X standard The data are standardized.

[0169] Modeling module 203 is developed using a Python framework (such as TensorFlow or PyTorch) and primarily consists of CNN and LSTM layers, with a fusion layer used for final fault prediction. The model input is real-time data processed by the data preprocessing module, and the output is fault prediction results and fault type diagnosis.

[0170] The fault prediction module 205 and parameter adjustment module 206 include the functions of fault warning and automatic adjustment of battery operating parameters. When the fault prediction value exceeds the preset threshold, the system will trigger an alarm and take measures, such as automatically adjusting the battery charging power, flow rate or stopping charging.

[0171] The optimization learning module 207 can feed back each fault data to the Deepseek-R1 model for retraining and improving the model's prediction accuracy. Online learning updates are performed using the Keras-based deep learning framework.

[0172] The vanadium redox flow battery fault prediction system works by collecting real-time battery operating status data (such as voltage, current, temperature, pH, flow rate, etc.) through multiple sensors. This data is then processed through a data preprocessing module, including denoising and normalization. The processed data is then fed into the Deepseek-R1 model, which extracts local features through a CNN layer and time series features through an LSTM layer. The outputs of these two layers are then combined to predict faults. When the model identifies a potential fault, the system issues an alert through the fault warning module and adjusts battery parameters based on the fault type to delay the occurrence and ensure system stability.

[0173] It is not difficult to find that this embodiment is a system embodiment corresponding to the first embodiment, and this embodiment can be implemented in conjunction with the first embodiment. The relevant technical details mentioned in the first embodiment are still valid in this embodiment, and to reduce repetition, they are not repeated here. Accordingly, the relevant technical details mentioned in this embodiment can also be applied to the first embodiment.

[0174] It is worth noting that all modules involved in this embodiment are logical modules. In actual applications, a logical unit can be a physical unit, a part of a physical unit, or a combination of multiple physical units. In addition, to highlight the innovations of the present invention, this embodiment does not include units that are not closely related to solving the technical problems proposed by the present invention. However, this does not mean that other units do not exist in this embodiment.

[0175] The third embodiment of the present invention relates to a network side server, such as Figure 4 As shown, it includes at least one processor 302; and a memory 301 that is communicatively connected to the at least one processor 302; wherein the memory 301 stores instructions that can be executed by the at least one processor 302, and the instructions are executed by the at least one processor 302 to enable the at least one processor 302 to execute the above-mentioned data processing method.

[0176] Memory 301 and processor 302 are connected using a bus. The bus can include any number of interconnected buses and bridges, connecting various circuits of one or more processors 302 and memory 301. The bus can also connect various other circuits such as peripheral devices, voltage regulators, and power management circuits. These are all well known in the art and are therefore not described further herein. The bus interface provides an interface between the bus and the transceiver. The transceiver can be a single component or multiple components, such as multiple receivers and transmitters, providing a unit for communicating with various other devices over a transmission medium. Data processed by processor 302 is transmitted over a wireless medium via an antenna. Furthermore, the antenna receives data and transmits it to processor 302.

[0177] The processor 302 is responsible for managing the bus and general processing, and can also provide various functions, including timing, peripheral interfaces, voltage regulation, power management, and other control functions. The memory 301 can be used to store data used by the processor 302 when performing operations.

[0178] A fourth embodiment of the present invention relates to a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the all-vanadium redox flow battery fault prediction method of the first embodiment.

[0179] That is, those skilled in the art will understand that all or part of the steps in the above-mentioned embodiments can be implemented by instructing the relevant hardware through a program, which is stored in a storage medium and includes a number of instructions for causing a device (which may be a single-chip microcomputer, chip, etc.) or a processor to execute all or part of the steps of the methods described in the various embodiments of the present application. The aforementioned storage medium includes: a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk, etc., various media that can store program code.

[0180] The above is only an embodiment of the present invention. Common knowledge such as the known specific structures and characteristics in the scheme is not described in detail here. Ordinary technicians in the field are aware of all common technical knowledge in the technical field of the invention before the application date or priority date, can obtain all existing technologies in the field, and have the ability to apply conventional experimental means before that date. Ordinary technicians in the field can improve and implement this scheme in combination with their own abilities under the inspiration given by this application. Some typical known structures or known methods should not become obstacles for ordinary technicians in the field to implement this application. It should be pointed out that for those skilled in the art, without departing from the structure of the present invention, several variations and improvements can be made, which should also be regarded as the scope of protection of the present invention. These will not affect the effect of the implementation of the present invention and the practicality of the patent. The scope of protection required by this application shall be based on the content of its claims, and the specific implementation methods and other records in the specification can be used to interpret the content of the claims.

[0181] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for predicting failure of an all-vanadium redox flow battery, characterized in that: include: Collecting and preprocessing the operating data of the all-vanadium redox flow battery; Build the Deepseek-R1 model based on deep learning. The Deepseek-R1 model includes a convolutional neural network layer, a long short-term memory network layer, and a fusion layer. The convolutional neural network layer extracts local features from battery data, and the long short-term memory network layer captures the time dependency of the data to analyze long-term trends and temporal changes during battery operation. Train and validate the Deepseek-R1 model based on deep learning; The real-time collected operating data is pre-processed and then input into the trained Deepseek-R1 model for fault prediction and fault type diagnosis; Trigger fault warnings based on predictions from the Deepseek-R1 model and automatically adjust battery operating parameters as needed; Fault data is fed back into the Deepseek-R1 model, and the model is optimized through online learning to improve the accuracy of fault prediction.

2. The method for predicting failure of an all-vanadium redox flow battery according to claim 1, wherein: The collecting of the operating data of the all-vanadium redox flow battery and preprocessing of the operating data of the all-vanadium redox flow battery include: Use multi-dimensional sensors to collect operating data of all-vanadium redox flow batteries; Perform data cleaning and standardization on the operating data of all-vanadium redox flow batteries; Time series processing is used for the operating data of all-vanadium redox flow battery.

3. The all-vanadium redox flow battery fault prediction method according to claim 1, characterized in that: The convolutional neural network layer is used to extract local features in the battery data. The convolutional neural network layer can effectively identify local patterns in the data and enhance the model's sensitivity to data details. The convolutional neural network layer extracts local abnormal features in the data layer by layer by constructing a multi-layer convolutional neural network; The output layer receives preprocessed operating data, including voltage, current, temperature, pH value, flow rate, etc., and the operating data is organized as a two-dimensional tensor in the form of time series; The convolutional layer uses a 3×3 convolution kernel to adapt to the feature extraction requirements of data of different scales; the step size is set to 1 or 2 to ensure that sufficient feature information is retained while improving computational efficiency; 32 filters are set in the first layer, and the number of filters increases to 64 in subsequent layers, and finally to 128 filters to extract deeper features layer by layer; The activation function uses the ReLU function, and the ReLU activation function formula is: ReLU(x)=max(0,x) The ReLU activation function is used to introduce nonlinear mapping, which makes feature extraction nonlinear and avoids the gradient vanishing problem. The pooling layer uses maximum pooling with a pooling window size of 2×2 and a step size of 2 to reduce the dimension and retain the main feature information. Finally, the feature matrix after convolution operation and pooling is input into the long short-term memory network layer for time series analysis.

4. The all-vanadium redox flow battery fault prediction method according to claim 1, characterized in that: The LSTM layer is used to capture the time dependency of data and analyze long-term trends and timing changes during battery operation. The LSTM layer can process possible failure modes of batteries during long-term operation and identify dynamic changes. The long short-term memory network layer controls the flow of information through input gates, forget gates, and output gates; The forget gate is used to determine how much past information the memory cell state should retain at the current time step. The forget gate calculation formula is: f t =σ(W f ·[h t-1 ,x t ]+b f ) Among them, f t is the output of the forget gate, W f is the weight of the forget gate, b f is the bias of the forget gate, h t-1 is the hidden state of the previous time step, χ t is the input data of the current time step, σ is the sigmoid activation function; The input gate is used to determine how much new information should be stored in the cell state at the current time step. The input gate calculation formula is: I t =σ(W i ·[h t-1 ,x t ]+b i ) Among them, i t is the output of the input gate, σ is the sigmoid activation function, and W i The weight matrix of the input gate, h t-1 is the hidden state of the previous time step, χ t is the input data of the current time step, b i is the bias vector of the input gate, is the candidate cell state value, tanh is the hyperbolic tangent activation function, W C is the weight matrix used to calculate the candidate cell state, b C is the bias vector used to calculate the candidate cell state; State update is the core process of LSTM. Through the joint action of the forget gate and the input gate, the cell state transmits and updates information between time steps. The state update calculation formula is: Among them, C t is the cell state at the current time step, f t is the output of the forget gate, C t-1 is the cell state at the previous time step, i t is the output of the input gate, is the candidate cell state value; The output gate is used to determine the hidden state output of the current time step. The output gate calculation formula is: the t =σ(W o ·[h t-1 ,x t ]+b o ) h t =o t *tanh(C t ) Among them, t is the output of the output gate, σ is the sigmoid activation function, W o is the weight matrix of the output gate, h t-1 is the hidden state of the previous time step, χ t is the input data of the current time step, b o is the bias vector of the output gate, h t is the final hidden state of the current time step, tanh is the hyperbolic tangent activation function, C t is the cell state at the current time step.

5. The method for predicting failure of an all-vanadium redox flow battery according to claim 1, wherein: The fusion layer performs residual fusion on the features extracted by the convolutional neural network layer and the long short-term memory network layer, comprehensively considering the static and dynamic characteristics of the battery system to form the final fault prediction result; Feature splicing directly splices the static features extracted by the convolutional neural network layer and the dynamic features extracted by the long short-term memory network layer, and further processes them through the fully connected layer. It is suitable for data with low feature dimensions; The fully connected layer performs further nonlinear transformation and feature fusion on the concatenated feature vectors. By adjusting the number of neurons and activation function of the fully connected layer, the expression ability and output form of the model can be controlled to adapt it to different fault prediction task requirements.

6. The all-vanadium redox flow battery fault prediction method according to claim 1, characterized in that: The training and verification of the Deepseek-R1 model based on deep learning includes training the Deepseek-R1 model using a large amount of historical operating data. During the training process, an optimization algorithm is used to adjust the model parameters to ensure that the model can accurately learn the behavior pattern of the battery system. The training data includes data from the battery under normal operating conditions and various fault conditions. The normal operating data is used by the model to learn the normal behavior patterns of the battery, while the data under fault conditions is used to train the model to identify fault characteristics.

7. The all-vanadium redox flow battery fault prediction method according to claim 1, characterized in that: The fault warning triggered by the prediction results output by the Deepseek-R1 model includes triggering a fault warning when the fault probability predicted by the model exceeds a set threshold, or when real-time monitored parameters such as voltage, current, and temperature exceed the normal range. The warning information is promptly notified to the operator through sound and light alarms, SMS notifications, system pop-up windows, etc., thereby reminding the operator to pay attention to the operating status of the battery system and take corresponding measures.

8. The method for predicting failure of an all-vanadium redox flow battery according to claim 1, characterized in that: The automatic adjustment of the battery's operating parameters as needed includes automatically reducing the charging power or stopping charging when the battery temperature is too high to avoid overheating damage; automatically adjusting the current output when the battery current is abnormal to prevent battery damage caused by excessive current; for abnormal electrolyte conditions, the system prompts the operator to detect and adjust the electrolyte composition, and automatically adjusts the charge and discharge strategy according to the pH value and concentration of the electrolyte, reducing the charge and discharge depth or limiting the charging current to slow down further deterioration of the electrolyte; for abnormal fluidity, the pump speed is automatically adjusted to optimize the electrolyte flow rate, ensure the uniformity of ion transmission within the battery, and restore normal operating conditions.

9. The method for predicting failure of an all-vanadium redox flow battery according to claim 1, characterized in that: Feeding back fault data to the Deepseek-R1 model and optimizing the model through online learning to improve the accuracy of fault prediction includes adopting an online learning mechanism. When a fault occurs, the system will feedback the fault data in real time and incorporate the fault data into the Deepseek-R1 model retraining process to optimize the accuracy and robustness of the model.

10. A vanadium redox flow battery fault prediction system, characterized in that: The method for predicting faults of all-vanadium redox flow batteries according to claims 1 to 9 comprises: Data acquisition module, used to collect operating data of all-vanadium redox flow batteries; Data preprocessing module, used to preprocess the operating data of the all-vanadium liquid flow battery; The modeling module is used to build the Deepseek-R1 model based on deep learning. The Deepseek-R1 model includes a convolutional neural network layer, a long short-term memory network layer, and a fusion layer. The convolutional neural network layer extracts local features from battery data, and the long short-term memory network layer captures the time dependency of the data to analyze long-term trends and temporal changes during battery operation. Model training and verification module, used to train and verify the Deepseek-R1 model based on deep learning; The fault prediction module is used to pre-process the real-time collected operating data and input it into the trained Deepseek-R1 model for fault prediction and fault type diagnosis; The parameter adjustment module is used to trigger fault warnings based on the prediction results output by the Deepseek-R1 model and automatically adjust the battery's operating parameters as needed; The optimization learning module is used to feed fault data back to the Deepseek-R1 model, optimize the model through online learning, and improve the accuracy of fault prediction.

Images