Method and system for predicting fatigue life of compressed air energy storage cavern and storage equipment

Abstract

The invention discloses a compressed air energy storage cavern fatigue life prediction method and system and storage equipment, and belongs to the technical field of compressed air energy storage. The method comprises the following steps: preprocessing compressed air storage hole data to be analyzed to obtain a sample data set; and inputting the sample data set into a pre-trained prediction model of the key parameters of the gas storage cavern structure, performing analysis to obtain an analysis result, and predicting the fatigue life of the cavern structure of the compressed air storage cavern according to the analysis result. The method can accurately predict the fatigue life of the compressed air energy storage cavern, discover potential safety and efficiency problems in advance, effectively reduce the fatigue damage risk of long-term operation of the underground cavern, and guarantee the stability of a power grid.

Description

Technical Field

[0001] The present invention relates to the technical field of compressed air energy storage, and in particular to a method, system and storage device for predicting the fatigue life of a compressed air energy storage chamber. Background Art

[0002] With the continuous expansion of the installed capacity of renewable energy sources such as solar energy and wind energy, the proportion of new energy power generation in the energy supply system is increasing day by day. However, due to the significant volatility and intermittency characteristics of renewable energy sources such as solar energy and wind energy, their power generation is greatly affected by natural factors such as weather and seasons, and it is difficult to achieve stable and continuous power output like traditional fossil energy power generation. This instability of the power generation characteristics brings huge challenges to the stable operation of the power grid. When the power generation of new energy fluctuates greatly, the supply-demand balance of the power grid is broken, which may lead to a series of problems such as frequency deviation and voltage fluctuation, and even cause the collapse of the power grid in severe cases, threatening the safe and reliable operation of the entire power system.

[0003] Under this background, compressed air energy storage technology, as a highly potential energy storage solution, has received extensive attention and entered a golden period of rapid development. This technology makes ingenious use of the compressibility and storability of air. During the low-demand period of electricity, the excess electricity in the power grid is used to compress air, and the high-pressure air is stored in specific spaces such as underground chambers; when the peak electricity demand comes, the stored compressed air is released, and the internal energy of the air is converted into mechanical energy through equipment such as expanders, and then drives the generator to generate electricity, realizing the storage and release of electric energy, effectively balancing the supply-demand contradiction of the power grid at different times, and providing an important technical support for solving the intermittency and volatility problems of renewable energy power generation.

[0004] However, in the application process of compressed air energy storage technology in underground chambers, a key and intractable problem is faced - the fatigue damage of the underground chamber structure. During the long-term inflation, gas storage, and deflation cycles, the underground chamber structure bears periodic pressure changes, and this repeated stress action will cause fatigue damage to the surrounding rock and lining structure of the chamber. With the continuous accumulation of damage, phenomena such as crack expansion, deterioration of sealing performance, and even air leakage may occur in the chamber, which will not only significantly reduce the efficiency of the energy storage system, increase the operation cost, but also may cause serious safety accidents, posing a threat to the surrounding environment and personnel safety. Summary of the Invention

[0005] In view of the problems in the existing technology, such as the instability of the power grid caused by the fluctuations of new energy power generation, and the fatigue damage prone to occur during the long-term operation of compressed air energy storage underground caverns, which will lead to problems such as efficiency and safety. The present invention provides a method for predicting the fatigue life of compressed air energy storage caverns, which can accurately predict the fatigue life of compressed air energy storage caverns, discover potential safety and efficiency problems in advance, effectively reduce the risk of fatigue damage during the long-term operation of underground caverns, and ensure the stability of the power grid.

[0006] To achieve the above object, the present invention provides the following technical solutions.

[0007] In a first aspect, the present invention provides a method for predicting the fatigue life of a compressed air energy storage cavern, including: Preprocessing the compressed air storage cavern data to be analyzed to obtain a sample data set; Inputting the sample data set into a prediction model of the key parameters of the structure of the air storage cavern for pre-training, performing analysis to obtain an analysis result, and predicting the fatigue life of the cavern structure of the compressed air storage cavern according to the analysis result; The training method of the prediction model of the key parameters of the structure of the pre-trained air storage cavern includes: Obtaining compressed air storage cavern data and performing preprocessing to obtain a key parameter data set; Using a CNN spatial feature extraction model to extract spatial features from the key parameter data set to obtain spatial feature data; Inputting the spatial feature data into a prediction model based on LSTM for training processing to obtain a prediction model of the key parameters of the structure of the pre-trained air storage cavern.

[0008] As a further improvement of the present invention, the obtaining of the compressed air storage cavern data and performing preprocessing to obtain a key parameter data set includes: Based on the engineering geological data, design parameters and operation parameters of the air storage cavern in the compressed air storage cavern data, using a numerical simulation method to simulate the charging and discharging process of the compressed air energy storage underground cavern to obtain the key parameters of the steel lining, concrete lining and surrounding rock of the cavern; The key parameters of the steel lining, concrete lining and surrounding rock of the cavern include the maximum tensile strain of the steel lining , the minimum tensile strain of the steel lining , the maximum tensile stress of the concrete lining , the first principal stress of the surrounding rock , the second principal stress of the surrounding rock and the third principal stress of the surrounding rock .

[0009] As a further improvement of the present invention, the using of a CNN spatial feature extraction model to extract spatial features from the key parameter data set to obtain spatial feature data includes: The key parameter dataset is processed layer by layer using a CNN spatial feature extraction model to extract feature maps; The mean square error loss function is used to calculate the loss function between the output features and the true features in the feature map ; where, is the loss value; is the number of samples; is the th true value; is the th predicted value, ; According to the gradient of the loss function, the parameters of the CNN are updated. The CNN is optimized through the backpropagation algorithm to adjust the convolution kernel and weights; then, the loss function is recalculated until the loss value L of the loss function meets the set conditions, and the feature sequence is output to obtain the spatial feature data.

[0010] As a further improvement of the present invention, the spatial feature data is input into a prediction model based on LSTM for training to obtain a pre-trained prediction model of the key parameters of the gas storage cavern structure, including: The training dataset in the spatial feature data is input into a prediction model based on LSTM for training to obtain the training results and the trained prediction model based on LSTM; The test dataset in the spatial feature data is input into the trained prediction model based on LSTM to obtain the test results; The training results and the test results are compared to determine whether the prediction model based on LSTM converges. If it converges, the pre-trained prediction model of the key parameters of the gas storage cavern structure is obtained.

[0011] As a further improvement of the present invention, inputting the training dataset in the spatial feature data into a prediction model based on LSTM for training to obtain the training results includes: A prediction model based on LSTM is constructed according to the LSTM network, and an input layer, an LSTM layer, and an output layer are set in the prediction model based on LSTM; The input layer receives the training dataset in the spatial feature data. The input layer extracts the feature sequence in the training dataset of the spatial feature data and sends the extracted feature sequence in the training dataset to the LSTM layer; The LSTM layer receives the feature sequence in the training dataset transmitted by the input layer, trains the LSTM layer. After the training is completed, the changing trend of the key parameters in several cycles is captured, and the LSTM layer sends the changing trend to the output layer; The output layer receives the changing trend transmitted by the LSTM layer and generates the training results.

[0012] As a further improvement of the present invention, the LSTM layer receives the feature sequence in the training set transmitted from the input layer, and trains the LSTM layer, including: The LSTM layer receives the feature sequence in the training set transmitted from the input layer, calculates the output of the feature sequence in the training set at each time step in the order of time steps, obtains the predicted chamber structure state parameters, compares the predicted chamber structure state parameters with the true values, and calculates the loss function; Update the LSTM layer according to the gradient of the loss function, calculate the gradient of the loss function in each time step by the BPTT algorithm in reverse, and use the gradient descent method to update the parameters of the LSTM network until the parameters of the LSTM network meet the set conditions, and complete the training of the LSTM layer.

[0013] As a further improvement of the present invention, input the sample data set into the prediction model of the key parameters of the pre-trained compressed air storage chamber structure, perform analysis, obtain the analysis result, and predict the fatigue life of the chamber structure of the compressed air storage cavern according to the analysis result, including: Input the sample data set into the prediction model of the key parameters of the trained compressed air storage chamber structure to obtain the prediction result; Analyze the prediction result according to the fatigue criterion to obtain the prediction result of the fatigue life of the compressed air energy storage chamber; Optimize the chamber structure parameters according to the prediction result of the fatigue life of the compressed air energy storage chamber and the engineering design service life.

[0014] In a second aspect, the present invention provides a compressed air energy storage chamber fatigue life prediction system, including: A data set acquisition module: used to preprocess the compressed air storage cavern data to be analyzed to obtain a sample data set; An analysis result module: used to input the sample data set into the prediction model of the key parameters of the pre-trained compressed air storage chamber structure, perform analysis, obtain the analysis result, and predict the fatigue life of the chamber structure of the compressed air storage cavern according to the analysis result; The training method of the pre-trained prediction model of the key parameters of the compressed air storage chamber structure includes: Obtain the compressed air storage cavern data and perform preprocessing to obtain a key parameter data set; Use the CNN spatial feature extraction model to extract spatial features from the key parameter data set to obtain spatial feature data; Input the spatial feature data into the prediction model based on LSTM for training processing to obtain the prediction model of the key parameters of the pre-trained compressed air storage chamber structure.

[0015] In a third aspect, the present invention provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the method for predicting the fatigue life of a compressed air energy storage chamber is implemented.

[0016] In a fourth aspect, the present invention provides a computer-readable storage medium storing a computer program, and when the computer program is executed by a processor, the method for predicting the fatigue life of a compressed air energy storage chamber is implemented.

[0017] In a fifth aspect, the present invention provides a computer program product including computer instructions, and when the computer instructions are executed by a processor, the method for predicting the fatigue life of a compressed air energy storage chamber is implemented.

[0018] Compared with the prior art, the present invention has the following beneficial effects: By accurately predicting the fatigue life of the compressed air energy storage chamber, this application can pre-judge the possible fatigue damage conditions of the chamber structure during long-term operation, and thus provide a scientific basis for power grid dispatching, enabling the power grid to reasonably allocate power resources according to the health status of the energy storage chamber, effectively buffering the impact of new energy power generation fluctuations on the power grid, greatly enhancing the stability and reliability of the power grid, and ensuring the continuity of power supply. Moreover, through the organic combination of the preprocessing technology, the CNN spatial feature extraction model, and the LSTM algorithm, this application can deeply mine the key information in the chamber operation data, accurately analyze the fatigue evolution law of the chamber structure, and discover potential safety and efficiency problems in advance. Based on the accurate fatigue life prediction results, the operation and maintenance personnel can timely take targeted maintenance and repair measures to avoid the further deterioration of the chamber structure, effectively reduce the risk of long-term operation fatigue damage of the underground chamber, extend the service life of the chamber, and at the same time ensure that the energy storage system is always in an efficient operation state, improving the energy utilization efficiency and economic benefits. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] The drawings described herein are for illustrative purposes only and are not intended to limit the scope of the disclosure of the present invention in any way. In the drawings: Figure 1 is a schematic flow chart of a method for predicting the fatigue life of a compressed air energy storage chamber according to the present invention; Figure 2 is a schematic flow chart of a training method for a prediction model of key parameters of a pre-trained gas storage chamber structure according to the present invention; Figure 3 is a schematic flow chart of the specific process of a method for predicting the fatigue life of a compressed air energy storage chamber according to the present invention; Figure 4Schematic diagram of numerical simulation of compressed air energy storage cavern operation in a fatigue life prediction method for a compressed air energy storage cavern according to the present invention; Figure 5 Schematic diagram of the variation law of key parameters of the steel lining, concrete lining and surrounding rock of the cavern in a fatigue life prediction method for a compressed air energy storage cavern according to the present invention; Figure 6 Schematic diagram of the structure of a fatigue life prediction system for a compressed air energy storage cavern according to the present invention; Figure 7 Schematic diagram of the structure of a training system for a prediction model of key parameters of a pre-trained gas storage cavern structure according to the present invention; Figure 8 Schematic diagram of an electronic device in an embodiment of the present invention. Detailed implementation manners

[0020] In order to enable those skilled in the art of the present technology to better understand the technical solutions in the present invention, the following will clearly and completely describe the technical solutions in the present invention in conjunction with the accompanying drawings in the present invention. The described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments in the present invention without making creative efforts shall fall within the protection scope of the present invention.

[0021] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the technical field of the present invention. The terms used in the description of the present invention in this specification are only for the purpose of describing specific embodiments and are not intended to limit the present invention. The term "and / or" used herein includes any and all combinations of one or more of the related listed items.

[0022] In view of the problems in the prior art that the instability of the power grid is caused by the fluctuation of new energy power generation, and the long-term operation of underground caverns for compressed air energy storage is prone to fatigue damage, which will cause problems such as efficiency and safety. The present invention provides a method for predicting the fatigue life of a compressed air energy storage cavern, as Figure 1 shown, including: S100: Preprocess the data of the compressed air storage cavern to be analyzed to obtain a sample data set; S200: Input the sample data set into the prediction model of the key parameters of the pre-trained gas storage cavern structure for analysis to obtain an analysis result, and predict the fatigue life of the cavern structure of the compressed air storage cavern according to the analysis result.

[0023] As Figure 2 shown, a training method for a prediction model of key parameters of a pre-trained gas storage cavern structure includes: S201: Obtain the data of the compressed air storage cavity, and perform preprocessing to obtain a key parameter data set; S202: Use the CNN spatial feature extraction model to extract spatial features from the key parameter data set to obtain spatial feature data; S203: Input the spatial feature data into the prediction model based on LSTM for training to obtain a prediction model for the key parameters of the pre-trained gas storage cavity structure.

[0024] This method can accurately predict the fatigue life of the compressed air energy storage cavity, discover potential safety and efficiency problems in advance, effectively reduce the risk of long-term operation fatigue damage of underground cavities, and ensure the stability of the power grid.

[0025] The following further explains and illustrates the present application in conjunction with specific drawings.

[0026] As Figure 3 shown, a method for predicting the fatigue life of a compressed air energy storage cavity includes: S1: Obtain the engineering geological data, design parameters, and operation parameters of the gas storage cavity.

[0027] Specifically, obtain the engineering geological data, design parameters, and operation parameters of the gas storage cavity. Among them, the engineering geological data of the gas storage cavity includes the rock density , cohesion , internal friction angle , rock elastic modulus , rock Poisson's ratio . Among them, the rock density is measured by the wax-sealing method, the cohesion , internal friction angle are measured by the rock triaxial compression test, and the rock elastic modulus , rock Poisson's ratio are measured by the uniaxial compression test; the design parameters include the cavity volume , cavity surface area , cavity radius , steel lining thickness , steel lining density , steel lining elastic modulus , steel lining Poisson's ratio , concrete lining thickness , concrete lining density , concrete lining elastic modulus , concrete lining Poisson's ratio ; the operation parameters include the initial temperature inside the cavity , the initial air pressure inside the cavity , the gas injection temperature , the gas injection time for one cycle 、Storage time after gas injection 、Gas production time 、Storage time after gas production 、Gas injection flow rate 。

[0028] S2: Conduct numerical simulations of cyclic charging and discharging of compressed air energy storage underground caverns to obtain key parameters of the steel lining, concrete lining and surrounding rock of the caverns.

[0029] Among them, examples of engineering geology, design parameters and operation parameters for conducting numerical simulations of cyclic charging and discharging of compressed air energy storage underground caverns are shown in Tables 1 - 3 respectively.

[0030] Table 1 Examples of engineering geology parameters for numerical simulations of cyclic charging and discharging of compressed air energy storage underground caverns

[0031] Table 2 Examples of cavern design parameters for numerical simulations of cyclic charging and discharging of compressed air energy storage underground caverns

[0032] Table 3 Examples of cavern operation parameters for numerical simulations of cyclic charging and discharging of compressed air energy storage underground caverns

[0033] Conduct numerical simulations of cyclic charging and discharging of compressed air energy storage underground caverns in COMSOL Multiphysics software. As Figure 4 shown, Figure 4 where the vertical axis is the depth of the compressed air energy storage underground cavern, and the horizontal axis is the width on both sides of the center point of the cross-section of the compressed air energy storage underground cavern. Obtain the maximum tensile strain and minimum tensile strain of the steel lining, the maximum tensile stress of the concrete lining, and the first principal stress , second principal stress and third principal stress of the surrounding rock in each cycle.

[0034] In COMSOL Multiphysics (multiphysics simulation platform) software, the computational domain is numerically discretized through a discretized grid model, and a local grid refinement strategy is adopted to densify the elements in the area around the cavern wall to accurately capture the stress concentration effect. The rock mechanical behavior adopts the Mohr - Coulomb elastoplastic constitutive model, which is used to simulate the yield, plastic deformation and shear failure characteristics of rock and soil under cyclic loading. This model applies a gravity field, and the initial displacement is set as a zero - value boundary; the boundary conditions are set as follows: the upper boundary is free, the lower boundary is fixed and constrained, the left and right boundaries are applied with roller supports, and the initial air pressure in the cavern , which is applied through the air pressure boundary condition of the inner boundary of the steel lining; the initial value of the temperature field is . A convective heat transfer boundary is set on the inner wall of the steel lining to simulate the gas-structure heat exchange. In the COMSOL Multiphysics software, first, the in-situ stress balance is carried out, and then the simulation of 30 cycles of cyclic charging and discharging is carried out. The key parameters of the cavern steel lining, concrete lining, and surrounding rock obtained are as Figure 5 shown.

[0035] Therefore, the key parameters of the cavern steel lining, concrete lining, and surrounding rock are respectively the maximum tensile strain and the minimum tensile strain of the steel lining in each cycle, the maximum tensile stress of the concrete lining, the first principal stress of the surrounding rock, the second principal stress and the third principal stress .

[0036] S3: Process the obtained key parameters, use the CNN (Convolutional Neural Network) algorithm for spatial feature extraction, and obtain the distribution law of the key parameters during each charging and discharging cycle.

[0037] S31: CNN feature sequence extraction Organize the stress and strain values obtained from the simulation in each cycle into a format suitable for CNN input. Assume that the simulated spatial region is divided into grid points, where represents the number of grid points in the first spatial dimension; represents the number of grid points in the second spatial dimension; represents the number of grid points in the third spatial dimension, and there are corresponding stress and strain values at each grid point. For each cycle , organize the key parameters of each cavern structure into a three-dimensional matrix , where represents the stress or strain value at the grid position in the th cycle. To meet the input requirements of CNN, further process the three-dimensional matrix into a suitable tensor form to form a tensor, where is the number of channels, and each channel corresponds to a physical quantity. For each cycle , the input data can be expressed as:

[0038] where is the The input data of one cycle; is the three-dimensional matrix of the th physical quantity in the

[0039] Construct a convolutional layer. Slide the convolutional kernel on the input data to extract local features and capture the concentrated areas of stress and strain. The mathematical expression of the convolutional layer is:

[0040] where, is the value of the output feature map at the grid position and channel ; is the input data; is the value of the input feature map at the grid position and channel ; is the current position coordinate of the output feature map; is the offset inside the convolutional kernel. The value range of is from 0 to , The value range of is from 0 to and the value range of is a positive integer; is a fixed channel index, indicating that this operation is only for the th channel of the input; is the convolutional kernel; is the weight value of the convolutional kernel at the grid position and channel is the bias term; is the bias value added to the output channel to adjust the numerical distribution of the output.

[0041] Select an activation function. After the convolutional operation, use the activation function to introduce non-linearity, alleviate the vanishing gradient problem, and accelerate the training speed. The activation function adopted is , and the expression is:

[0042] where, is the activated feature value, that is, the value passed to the next layer of the convolutional layer; is the original input value before neuron activation; means taking the larger value between 0 and . When the original input value before neuron activation is positive, the output is . When the original input value before neuron activation When it is negative or zero, the output is 0.

[0043] Construct a pooling layer, using max pooling to reduce the spatial size of the feature map, reduce the computational amount and retain important features. The mathematical expression of max pooling is:

[0044] Where, is the value of the feature map after pooling at the grid position , channel ; is the size of the pooling window; is the value of the output feature map at the grid position , channel ; is to traverse all positions within the pooling window and calculate the maximum value of the local area.

[0045] Feature map output. After multiple layers of convolution and pooling, the obtained feature map contains the spatial features of the input data. These feature maps will be used as the input of LSTM (Long Short-Term Memory, a time recurrent neural network) to predict the state of the chamber structure under long-term charging and discharging cycles.

[0046] S32: CNN Training The training process of CNN is divided into two stages: forward propagation and backward propagation.

[0047] In the forward propagation stage, the input data undergoes operations such as convolution layers, activation functions, pooling layers, etc., and finally obtains the output feature map. The mean squared error loss function is used to calculate the loss between the CNN output feature and the true feature. The expression of the loss function is:

[0048] Where, is the loss value; is the number of samples; is the th true value; is the th predicted value, .

[0049] In the backward propagation stage, the parameters of CNN are updated according to the gradient of the loss function to minimize the prediction error. The values of the convolution kernel and bias term are adjusted by calculating the gradients of the output layer, fully connected layer, convolution layer, and pooling layer. The specific steps and formulas are: Calculate the output layer gradient:

[0050] Calculate the fully connected layer gradient:

[0051]

[0052] Among them, is the weight matrix of the fully connected layer; is the activation output of the previous layer; is the transpose of the activation output of the previous layer; is a full vector; is the partial differential symbol.

[0053] Calculate the gradient of the convolutional layer: Assume that the input of the convolutional layer is , the output is , the convolutional kernel is , the bias term is , then:

[0054]

[0055] Among them, represents the convolution operation; is the gradient of the loss function with respect to the output of the convolutional layer; is the gradient of the output feature map at position ; is the gradient of the convolutional layer; is the gradient of the bias term.

[0056] Calculate the gradient of the pooling layer:

[0057] Among them, is an indicator function, which is 1 only when is the maximum value, and 0 otherwise.

[0058] According to the calculated gradient, use the gradient descent method to update the values of the convolutional kernel and the bias term:

[0059] According to the information of the convolutional layer gradient , adjust the convolutional kernel in the direction of the decrease of the loss function until the loss converges to a satisfactory degree, and complete the update iteration of the convolutional kernel .

[0060]

[0061] According to the information of the bias term gradient , adjust the bias term Adjust in the direction of the loss function decrease until the loss converges to a satisfactory level, and complete the update iteration of the bias term .

[0062] Among them, is the learning rate, which controls the step size of parameter update; represents the assignment operation.

[0063] Repeatedly execute the forward propagation and backward propagation processes until the loss function reaches a small value or no longer decreases significantly, and output the feature sequence.

[0064] S4: Input the spatial feature data into the prediction model of the key parameters of the gas storage cavern structure, and use the LSTM algorithm to train the extracted spatial feature data to obtain the variation law of the key parameters with the number of charging and discharging cycles, and establish the prediction model of the key parameters of the gas storage cavern structure; The LSTM algorithm trains the extracted spatial feature data, including the following steps: Use the feature sequence extracted by CNN as the input data of LSTM. The input feature at each time step t can be the feature vector of a certain cycle extracted by CNN. At the same time, directly obtain the chamber structure state parameters at the next time step t + 1 through S2, and prepare the corresponding target value, which is the true value Construct a prediction model based on LSTM according to the LSTM network, including an input layer, an LSTM layer, an output layer, etc. Among them, the input layer receives the feature sequence extracted by CNN, including the maximum tensile strain and minimum tensile strain of the steel lining in each cycle, the maximum tensile stress of the concrete lining, the first principal stress of the surrounding rock, the second principal stress and the third principal stress and other parameter spatial features.

[0065] The LSTM layer receives the feature sequence transmitted from the input layer and models the time series data through its internal gating mechanism to capture the variation trend of the key parameters in multiple cycles.

[0066] The output layer converts the output of the LSTM layer into specific predicted values.

[0067] Initialize the weight and bias parameters of the LSTM network, and adopt the Glorot initialization method. The specific formula is: Uniform distribution initialization:

[0068] Normal distribution initialization:

[0069] Among them, U is a uniform distribution; N is a normal distribution; is the number of input neurons; is the number of output neurons.

[0070] The training process of the LSTM is divided into two stages: forward propagation and backward propagation.

[0071] In the forward propagation stage, at each time step, the input feature vector is passed into the LSTM cell to calculate the activation values of the forget gate, input gate, and output gate, as well as the updates of the cell state and hidden state. The specific formulas are as follows: Forget gate:

[0072] In the formula, is the output of the forget gate; is the weight matrix of the forget gate; is the bias term of the forget gate; is the sigmoid function; is the hidden state at the current time step; is the hidden state at the previous time step; is the input vector at the current time step.

[0073] Input gate:

[0074] In the formula, is the output of the input gate; is the weight matrix of the input gate; is the bias term of the input gate.

[0075]

[0076] In the formula, is the candidate cell state; is the cell state weight matrix; is the cell state bias term.

[0077] Cell state update:

[0078] In the formula, is the cell state at the current time step; is the candidate cell state at the previous time step.

[0079] Output gate:

[0080] In the formula, is the output of the output gate; is the weight matrix of the output gate; is the bias term of the output gate.

[0081]

[0082] In the formula, is the hidden state at the current moment, and ⊙ represents element-wise multiplication; is the activation function; ( ) compresses each element in to the range [-1, 1].

[0083] The hidden state at the current moment After being processed by the output layer, it is transformed into the predicted chamber structure state parameters.

[0084] At each time step, the predicted chamber structure state parameters are compared with the true values to calculate the loss function.

[0085] In the backpropagation stage, the parameters of the LSTM network are updated according to the gradient of the loss function with the aim of minimizing the prediction error. Backpropagation calculates the gradient of the loss function with respect to each parameter in each time step through the BPTT (Backpropagation Through Time) algorithm. The specific process is as follows: Calculate the hidden state gradient:

[0086] Among them, is the loss value at the current moment; is the hidden state at the next time step.

[0087] Calculate the cell state gradient:

[0088] Among them, is the candidate cell state at the next time step; is the output of the forget gate at the next time step; is the derivative of tanh; is to independently calculate the derivative for each element in ; is the multiplication operation in the chain rule.

[0089] When , , at this time the gradient disappears.

[0090] When , , at this time the gradient is completely transmitted.

[0091] Calculate the gating gradients: Output gradient of the forget gate:

[0092] Output gradient of the input gate:

[0093] Candidate cell state gradient:

[0094] Calculate the weight and bias gradients: Weight matrix gradient of the forget gate:

[0095] Bias term gradient of the forget gate:

[0096] Weight matrix gradient of the input gate:

[0097] Bias term gradient of the input gate:

[0098] Weight matrix gradient of the cell state:

[0099] Bias term gradient of the cell state:

[0100] Weight matrix gradient of the output gate:

[0101] Bias term gradient of the output gate:

[0102] According to the calculated gradients, use gradient descent to update the parameters of the LSTM network:

[0103] According to the weight matrix gradient of the forget gate information, adjust the weight matrix of the forget gate in the direction of the loss function decrease until the loss converges to a satisfactory degree, and complete the update iteration of the weight matrix of the forget gate

[0104]

[0105] According to the weight matrix gradient of the forget gate information, adjust the bias term of the forget gate in the direction of the loss function decrease until the loss converges to a satisfactory degree, and complete the update iteration of the bias term of the forget gate

[0106] ​​

[0107] According to the gradient of the weight matrix of the input gate information, adjust the weight matrix of the input gate in the direction of the loss function decline until the loss converges to a satisfactory level, completing the update iteration of the weight matrix of the input gate .

[0108]

[0109] According to the gradient of the bias term of the input gate information, adjust the bias term of the input gate in the direction of the loss function decline until the loss converges to a satisfactory level, completing the update iteration of the bias term of the input gate .

[0110]

[0111] According to the gradient of the weight matrix of the cell state information, adjust the weight matrix of the cell state in the direction of the loss function decline until the loss converges to a satisfactory level, completing the update iteration of the weight matrix of the cell state .

[0112]

[0113] According to the gradient of the bias term of the cell state information, adjust the bias term of the cell state in the direction of the loss function decline until the loss converges to a satisfactory level, completing the update iteration of the bias term of the cell state .

[0114]

[0115] According to the gradient of the weight matrix of the output gate information, adjust the weight matrix of the output gate in the direction of the loss function decline until the loss converges to a satisfactory level, completing the update iteration of the weight matrix of the output gate .

[0116]

[0117] According to the gradient of the bias term of the output gate information, adjust the bias term of the output gate in the direction of the loss function decline until the loss converges to a satisfactory level, completing the update iteration of the bias term of the output gate .

[0118] Use 70% of the simulated data for training and 30% for testing, and repeatedly execute the forward propagation and backpropagation processes until the model converges.

[0119] S5: Use the trained model to predict the key parameters, and predict the fatigue life of the cavern structure according to the fatigue criterion; The method for predicting the fatigue life of the cavern structure according to the fatigue criterion is as follows: (1) Steel lining fatigue criterion:

[0120] In the formula, is the maximum tensile stress of the steel lining in each stress cycle; is the minimum tensile stress of the steel lining in each stress cycle, is the allowable strain, which can be calculated by the following formula:

[0121] In the formula, is the elongation at break of the material; is the breaking strength; is the elastic modulus; is the fatigue life, that is, the number of failure cycles.

[0122] (2) Concrete lining fatigue criterion:

[0123] In the formula, is the maximum tensile stress of the concrete lining, is the tensile strength of the concrete.

[0124] (3) Surrounding rock fatigue criterion:

[0125] In the formula, is the first principal stress of the surrounding rock; is the second principal stress of the surrounding rock; is the third principal stress of the surrounding rock; is the tensile strength of the rock.

[0126] When predicting the life of the cavern structure, if one of the above three criteria is met, it is considered that the cavern fails, and the corresponding number of cycles n is the limit cycle number of the cavern. The calculation formula for the operation life of the cavern is:

[0127] Among them, in one cycle, is the gas injection time, is the storage time after gas injection, is the gas production time, is the storage time after gas production; is the operating life of the chamber.

[0128] S6: Optimize the chamber structure parameters according to the prediction results and the engineering design service life until the predicted life meets the design requirements.

[0129] The second object of the present invention is to propose a fatigue life prediction system for a compressed air energy storage chamber, as Figure 6 shown, including: Data set acquisition module 100: used to preprocess the compressed air storage chamber data to be analyzed to obtain a sample data set; Analysis result module 200: used to input the sample data set into the prediction model of the key parameters of the pre-trained gas storage chamber structure for analysis to obtain an analysis result, and predict the fatigue life of the chamber structure of the compressed air storage chamber according to the analysis result.

[0130] As Figure 7 shown, a training system for a prediction model of the key parameters of a pre-trained gas storage chamber structure includes: Key parameter data module 201: used to obtain compressed air storage chamber data and preprocess it to obtain a key parameter data set; Spatial feature data module 202: used to extract spatial features from the key parameter data set by using a CNN spatial feature extraction model to obtain spatial feature data; Training prediction model module 203: used to input the spatial feature data into a prediction model based on LSTM for training processing to obtain a prediction model of the key parameters of the pre-trained gas storage chamber structure.

[0131] As Figure 8 shown, the third object of the present invention is to provide an electronic device, which includes: a processor 301, a memory 302 and a display screen 303. Among them, the memory 302 and the display screen 303 are both connected to the processor 301, such as through a bus 304. Optionally, the electronic device may further include a transceiver 305. It should be noted that in practical applications, the transceiver 305 is not limited to one, and the structure of this electronic device does not constitute a limitation to the embodiments of the present application.

[0132] The processor 301 may be a CPU (Central Processing Unit), a general-purpose processor, a DSP (Digital Signal Processor), an ASIC (Application Specific Integrated Circuit), an FPGA (Field Programmable Gate Array), or other programmable logic devices, transistor logic devices, hardware components, or any combination thereof. It can implement or execute various exemplary logical blocks, modules, and circuits described in connection with the disclosure of this application. The processor 301 may also be a combination that implements computing functions, such as a combination of one or more microprocessors, a combination of a DSP and a microprocessor, etc.

[0133] The bus 304 may include a path for transmitting information between the above components. The bus 304 may be a PCI (Peripheral Component Interconnect) bus or an EISA (Extended Industry Standard Architecture) bus, etc. The bus 304 may be divided into an address bus, a data bus, a control bus, etc.

[0134] The memory 302 may be a ROM (Read Only Memory) or other types of static storage devices that can store static information and instructions, a RAM (Random Access Memory) or other types of dynamic storage devices that can store information and instructions, or it may also be an EEPROM (Electrically Erasable Programmable Read Only Memory), a CD-ROM (Compact Disc Read Only Memory), or other optical disc storage, optical disc storage (including compact discs, laser discs, optical discs, digital versatile discs, Blu-ray discs, etc.), magnetic disk storage media, or any other medium that can be used to carry or store the desired program code in the form of instructions or data structures and can be accessed by a computer, but is not limited thereto.

[0135] The memory 302 is used to store the application program code for executing the solution of this application and is controlled by the processor 301 for execution. The processor 301 is used to execute the application program code stored in the memory 302 to implement the content shown in the foregoing method embodiments.

[0136] Figure 8 The illustrated electronic device is merely an example and should not impose any limitations on the functions and scope of use of the embodiments of this application.

[0137] The fourth objective of the present invention is to provide a computer-readable storage medium storing a computer program, on which a computer program is stored. When the program is executed by a processor, it implements each process of the method embodiment as described above. Figures 1 to 3 For example, a memory including instructions, and the above instructions can be executed by a processor of an electronic device to complete the above method.

[0138] A computer-readable storage medium can be a tangible device that holds and stores instructions used by an instruction execution device. A computer-readable storage medium can be, but is not limited to, an electrical storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any combination of the above. Specifically, a computer-readable storage medium can be a portable computer disk, a hard disk, a USB flash drive, a random access memory, a read-only memory, an erasable programmable read-only memory, a podium random access memory, a portable compact disk read-only memory, a digital versatile disk, a memory stick, a floppy disk, an optical disk, a magnetic disk, a mechanical encoding device, and any combination of the above.

[0139] The fifth objective of the present invention is to provide a computer program product including computer instructions, which implement each process of the method embodiment as described above when executed by a processor, and can achieve the same technical effects. To avoid repetition, it will not be elaborated here. Figures 1 to 3 When executed by a processor, the computer instructions can achieve each process of the method embodiment as described above and can achieve the same technical effects. To avoid repetition, it will not be elaborated here.

[0140] Upon reading the above description, many embodiments and many applications beyond the provided examples will be obvious to those skilled in the art. Therefore, the scope of this teaching should not be determined with reference to the above description, but should be determined with reference to the full scope of the foregoing claims and the equivalents of these claims. For the sake of comprehensiveness, all articles and references, including patent applications and published announcements, are incorporated herein by reference. The omission of any aspect of the subject matter disclosed herein in the foregoing claims is not intended to abandon such subject matter, nor should it be considered that the applicant has not considered such subject matter as part of the disclosed inventive subject matter.

[0141] The above content is a further detailed description of the present invention. It cannot be determined that the specific implementation of the present invention is limited to this. For those of ordinary skill in the technical field to which the present invention pertains, without departing from the concept of the present invention, several simple deductions or substitutions can still be made, which should all be regarded as belonging to the protection scope of the present invention.

Claims

1. A method for predicting the fatigue life of a compressed air energy storage chamber, characterized in that Including: Preprocess the compressed air storage cavern data to be analyzed to obtain a sample data set; Input the sample data set into the prediction model of the key parameters of the pre-trained gas storage cavern structure for analysis to obtain an analysis result, and predict the fatigue life of the cavern structure of the compressed air storage cavern according to the analysis result; The training method of the pre-trained prediction model of the key parameters of the gas storage cavern structure includes: Obtain the compressed air storage cavern data and preprocess it to obtain a key parameter data set; Use the CNN spatial feature extraction model to extract spatial features from the key parameter data set to obtain spatial feature data; Input the spatial feature data into the prediction model based on LSTM for training to obtain the pre-trained prediction model of the key parameters of the gas storage cavern structure.

2. A method for predicting the fatigue life of a compressed air energy storage chamber according to claim 1, characterized in that, The obtaining of the compressed air storage cavern data and preprocessing to obtain a key parameter data set includes: Based on the engineering geological data, design parameters and operation parameters of the gas storage cavern in the compressed air storage cavern data, use the numerical simulation method to simulate the charging and discharging process of the underground cavern of the compressed air energy storage to obtain the key parameters of the cavern steel lining, concrete lining and surrounding rock; The key parameters of the cavern steel lining, concrete lining and surrounding rock include the maximum tensile strain of the steel lining , the minimum tensile strain of the steel lining , the maximum tensile stress of the concrete lining , the first principal stress of the surrounding rock , the second principal stress of the surrounding rock , and the third principal stress of the surrounding rock .

3. A method for predicting the fatigue life of a compressed air energy storage chamber according to claim 1, characterized in that, The using of the CNN spatial feature extraction model to extract spatial features from the key parameter data set to obtain spatial feature data includes: Use the CNN spatial feature extraction model to process the key parameter data set layer by layer to extract feature maps; The mean square error loss function is used to calculate the loss function between the output features and the true features in the feature map ; wherein, is the loss value; is the number of samples; is the th true value; is the th predicted value, ; Update the parameters of the CNN according to the gradient of the loss function. The CNN is optimized through the backpropagation algorithm to adjust the convolution kernel and weights; then, recalculate the loss function until the loss value L of the loss function meets the set conditions, output the feature sequence, and obtain the spatial feature data.

4. A method for predicting the fatigue life of a compressed air energy storage chamber according to claim 1, characterized in that The inputting of the spatial feature data into the prediction model based on LSTM for training to obtain the pre-trained prediction model of the key parameters of the gas storage cavern structure includes: Input the training data set in the spatial feature data into the prediction model based on LSTM for training to obtain a training result and the trained prediction model based on LSTM; Input the test data set in the spatial feature data into the trained prediction model based on LSTM to obtain a test result; Compare the training result with the test result to determine whether the prediction model based on LSTM converges. If it converges, obtain the pre-trained prediction model of the key parameters of the gas storage cavern structure.

5. A method for predicting the fatigue life of a compressed air energy storage chamber according to claim 4, characterized in that, The inputting of the training data set in the spatial feature data into the prediction model based on LSTM for training to obtain a training result includes: Construct a prediction model based on LSTM according to the LSTM network, and set an input layer, an LSTM layer and an output layer in the prediction model based on LSTM; The input layer receives the training data set in the spatial feature data, extracts the feature sequence in the training data set of the spatial feature data, and sends the extracted feature sequence in the training data set to the LSTM layer; The LSTM layer receives the feature sequence in the training data set transmitted by the input layer, trains the LSTM layer, and after the training is completed, captures the change trend of the key parameters in several cycles. The LSTM layer sends the change trend to the output layer; The output layer receives the trend passed from the LSTM layer and generates the training result.

6. A method for predicting the fatigue life of a compressed air energy storage chamber according to claim 5, characterized in that The LSTM layer receives the feature sequence in the training set passed from the input layer and trains the LSTM layer, including: The LSTM layer receives the feature sequence in the training set passed from the input layer, calculates the output of the feature sequence in the training set at each time step in the order of time steps, obtains the predicted chamber structure state parameters, compares the predicted chamber structure state parameters with the true values, and calculates the loss function; Update the LSTM layer according to the gradient of the loss function, calculate the gradient of the loss function in each time step by the BPTT algorithm in reverse, and use the gradient descent method to update the parameters of the LSTM network until the parameters of the LSTM network meet the set conditions, and complete the training of the LSTM layer.

7. A method for predicting the fatigue life of a compressed air energy storage chamber according to claim 1, characterized in that, Input the sample data set into the prediction model of the key parameters of the pre-trained compressed air storage chamber structure, analyze it to obtain the analysis result, and predict the fatigue life of the chamber structure of the compressed air storage chamber according to the analysis result, including: Input the sample data set into the prediction model of the key parameters of the trained compressed air storage chamber structure to obtain the prediction result; Analyze the prediction result according to the fatigue criterion to obtain the prediction result of the fatigue life of the compressed air energy storage chamber; Optimize the chamber structure parameters according to the prediction result of the fatigue life of the compressed air energy storage chamber and the engineering design service life.

8. A prediction system for the fatigue life of a compressed air energy storage chamber, characterized in that, Including: Data set acquisition module: used to preprocess the compressed air storage chamber data to be analyzed to obtain the sample data set; Analysis result module: used to input the sample data set into the prediction model of the key parameters of the pre-trained compressed air storage chamber structure, analyze it to obtain the analysis result, and predict the fatigue life of the chamber structure of the compressed air storage chamber according to the analysis result; The training method of the prediction model of the key parameters of the pre-trained compressed air storage chamber structure includes: Obtain the compressed air storage chamber data and preprocess it to obtain the key parameter data set; Use the CNN spatial feature extraction model to extract the spatial features of the key parameter data set to obtain the spatial feature data; Input the spatial feature data into the prediction model based on LSTM for training processing to obtain the prediction model of the key parameters of the pre-trained compressed air storage chamber structure.

9. An electronic device, characterized in that, It includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the method for predicting the fatigue life of a compressed air energy storage chamber according to any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, and when the computer program is executed by the processor, it implements the method for predicting the fatigue life of a compressed air energy storage chamber according to any one of claims 1-7.

Images