Pipeline inversion lining repair operation process monitoring and evaluation method, device and medium

Abstract

The application provides a pipeline inversion inner lining repair operation process monitoring and evaluation method and device and a computer readable storage medium. By introducing a PELT variable point detection algorithm, the time series data is divided into different time periods based on the detected variable point quantity, and then different time period feature vectors are extracted. Based on the multi-dimensional feature vectors, a Transformer model is used to predict future features of process parameters, improving the prediction accuracy of the Transformer model, thereby realizing accurate process parameter monitoring. A multi-index decision CoCoSo method is introduced, and through multi-aggregation strategy fusion, both the optimal solution and the consistency of each strategy are considered, thereby improving the reliability of the decision result, obtaining a reliable process state evaluation result, realizing dynamic monitoring and adjustment of the inversion inner lining repair operation process parameters, improving the repair quality, and helping to ensure the safe and stable operation of the gas pipeline system.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of pipeline repair, in particular to a pipeline inversion lining repair operation process monitoring and evaluation method and device and a computer readable storage medium. BACKGROUND

[0002] With the large-scale use of gas, the aging of the pipe network and the increase of hidden dangers have brought problems. Therefore, the gas pipeline elimination reconstruction has become a trend. The gas pipeline elimination reconstruction currently mainly adopts two ways of open replacement and trenchless construction. Trenchless construction is a construction method of placing new lining in the old pipe after pretreatment of the inner wall of the old pipe to solve the defects of pipe cracks, corrosion and leakage and at least increase the service life by 30 years. It is the best solution for underground pipeline repair and reinforcement. Among them, the inversion lining repair method is the most typical method for repairing damaged gas pipelines by trenchless construction. Its core technology includes tubular textile composite material for inner wall repair, pressure inversion bin equipment and automatic control system. However, the key to the inversion lining repair quality lies in the monitoring and adjustment of the process during the repair.

[0003] For the problem of inversion lining repair operation process monitoring of small-diameter gas pipelines, the process parameters are mainly adjusted by the experience of construction personnel. The general process parameter monitoring process is to first give the set value of key parameters such as pressure and speed according to construction experience; secondly, the inversion lining repair progress and condition are observed by digging multiple operation pits and feedback is given; finally, the construction personnel adjust the key parameters of the inversion lining repair operation based on the feedback and previous experience to realize the monitoring and adjustment of the process parameters during the inversion lining repair process. However, the inversion lining repair process parameter monitoring mainly relies on manual experience judgment and adjustment, and there are problems such as short repair length and poor bending capacity, which cannot meet the needs of inversion lining repair of small-diameter gas pipelines with multiple bends and long distances. SUMMARY

[0004] The present application aims to provide a pipeline inversion lining repair operation process monitoring and evaluation method, device and computer readable storage medium which overcome the above problems or at least partially solve the above problems.

[0005] To achieve the above purpose, the technical solution of the present application is as follows:

[0006] The first aspect of the present application provides a pipeline inversion lining repair operation process monitoring and evaluation method, comprising:

[0007] segmenting the process parameter time series data by PELT change point detection algorithm;

[0008] extracting parameter feature vectors at different time periods;

[0009] The Transformer model is used to predict future process parameter characteristics;

[0010] The CoCoSo multi-index decision-making method is used to dynamically evaluate the process status.

[0011] Optionally, the segmentation of process parameter time series data using the PELT change point detection algorithm includes:

[0012] Collect the timing data of the process parameters during the inversion lining repair process;

[0013] Construct the PELT change point detection objective function:

[0014]

[0015] in, For the number of variable points, For the change point, The penalty factor is used to control the number of variable points. Let the piecewise mean squared error cost function be... Let i be the data subsequence of the i-th segment; the segmented mean square error cost function is:

[0016]

[0017] in, Let t be a subsequence of the time series from index a to b, where t is the time-th moment. Let be the observation value at time t. , where a is the index of the start time point of the segment and b is the index of the end time point of the segment.

[0018] Optionally, the extracted feature vectors for different time periods include:

[0019] Extract the feature vector for each time period. The feature vector includes: mean, standard deviation, upper quartile, FFT main frequency amplitude, and trend slope.

[0020] The feature matrix obtained by concatenating the sliding windows is used as input to construct the input sequence.

[0021] Optionally, the extraction of the feature vector for each time period includes:

[0022]

[0023] in, Let d be the feature vector, and d be the dimension. The mean, Standard deviation It is the upper quartile. This represents the FFT main frequency amplitude. The trend slope For segmented data, For length, For Discrete Fourier Transform, Time-centered coordinates;

[0024] The step of concatenating the sliding windows to obtain the feature matrix as input and constructing the input sequence includes:

[0025] The feature matrix is ​​obtained based on the feature vector. ;

[0026] The feature matrix is ​​used as input to construct a feature vector input sequence. :

[0027]

[0028] Where M is the total number of segments, R is the real matrix space, representing a matrix of dimension a×b, and N is the number of input sequence samples. The size of the sliding window. The number of training samples that can be generated; Concat is a vector concatenation operation.

[0029] Optionally, the use of the Transformer model to predict future process parameter characteristics includes:

[0030] The feature vector input sequence is mapped to a high-dimensional space and position information is injected.

[0031] Capture long-term dependencies through self-attention mechanisms;

[0032] The features of the self-attention output at each time step are nonlinearly transformed using a feedforward network;

[0033] Output the predicted eigenvector values ​​of future process parameters.

[0034] Optionally,

[0035] The step of mapping the feature vector input sequence to a high-dimensional space and injecting position information includes:

[0036] Set to hidden state for:

[0037]

[0038] in, Position encoding for embedding matrix The formula is as follows:

[0039]

[0040] Among them, time step position Dimension index d model For the hidden layer dimension of the model, This is a position-encoded vector;

[0041] The method of capturing long-term dependencies through self-attention mechanisms includes:

[0042] Set the first layer Attention output:

[0043]

[0044] in, These are the query, key, and value projection matrices, respectively. For the previous step, calculate the first step that carries mixed information of temporal sequence and features. The hidden state of the layer Dimensions for each attention head;

[0045] The nonlinear transformation of the self-attention output at each time step using a feedforward network includes:

[0046] Nonlinear transformation is performed using the following formula:

[0047]

[0048] Where x is the input feature and ReLU is the activation function. , These are the weights for the first and second layers, respectively. For bias terms, For output bias;

[0049] The predicted feature vector values ​​of the output future process parameters include:

[0050] Take the hidden state at the last time step Decoded into feature vector predictions for future process segments:

[0051]

[0052] The predicted characteristics of the next process segment are: The predicted feature vector Convert to a sequence of process parameters for future time periods The formula is as follows:

[0053]

[0054] Among them, L new The length of a future time period. It is Gaussian noise. The predicted process parameter value at time point k. The mean of the generated sequence, The slope represents the linear trend.

[0055] Optionally, the dynamic evaluation of process status using the CoCoSo multi-index decision-making method includes:

[0056] Normalize the decision matrix;

[0057] Calculate the sum of the weighted comparable sequences of process states for each time period;

[0058] Calculate the sum of geometrically comparable sequences of process states for each time period;

[0059] Calculate the relative score of the process status for each time period;

[0060] Calculate the final score for the process status at each time period.

[0061] Optionally,

[0062] The normalization of the decision matrix includes:

[0063] Set n parameters Predicted using the Transformer model Parameters Time period Parameter values, construct decision matrix .

[0064] For decision matrix Normalize:

[0065] Among them, C ij Let j be the value of parameter j at time i. and maximum value and minimum value ;

[0066] The calculation of the sum of the weighted comparable sequences of process states for each time period includes:

[0067] The sum of the weighted comparable sequences is calculated using the following formula:

[0068]

[0069] in, For the first The weights of each parameter;

[0070] The calculation of the sum of geometrically comparable sequences of process states for each time period includes:

[0071] The sum of geometrically comparable sequences is calculated using the following formula:

[0072] ;

[0073] The calculation of the relative score of the process status for each time period includes:

[0074] The relative score is calculated using the following formula:

[0075]

[0076]

[0077]

[0078] in, This represents the arithmetic mean of the scores from the weighted sum method and the weighted product method. This represents the sum of scores from the weighted sum method and the weighted product method relative to the optimal score. This represents a trade-off between the weighted sum method and the weighted product method scores. This is the balance coefficient;

[0079] The calculation of the final score for the process status in each time period includes:

[0080]

[0081] in, This represents the score for the process status at each time period.

[0082] A second aspect of the present invention provides a device for monitoring and evaluating the process of pipeline overturning lining repair, comprising:

[0083] The segmentation module is used to segment the timing data of process parameters using the PELT change point detection algorithm.

[0084] The extraction module is used to extract feature vectors of parameters at different time periods;

[0085] The prediction module is used to predict future process parameter characteristics using the Transformer model;

[0086] The evaluation module is used to dynamically evaluate the process status by combining the CoCoSo multi-index decision-making method.

[0087] A third aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon, characterized in that, when the program is executed by a processor, it implements the above-described method for monitoring and evaluating the pipeline overturning lining repair process.

[0088] Therefore, the pipeline overturning lining repair operation process monitoring and evaluation method, device, and computer-readable storage medium provided by this invention, by introducing the PELT variable point detection algorithm, divides the time series data into different time periods based on the number of detected variable points, and then extracts feature vectors for different time periods. Based on the multi-dimensional feature vectors, the Transformer model is used to predict the future features of process parameters, which improves the prediction accuracy of the Transformer model and thus achieves accurate process parameter monitoring. The multi-index decision-making CoCoSo method is introduced, which pursues the optimal solution while taking into account the consistency of each strategy through the fusion of multiple aggregation strategies, thereby improving the reliability of the decision results and obtaining reliable process status evaluation results. This enables dynamic monitoring and adjustment of process parameters for overturning lining repair operations, improves repair quality, and helps ensure the safe and stable operation of gas pipeline systems. Attached Figure Description

[0089] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0090] Figure 1 A flowchart of the pipeline overturning lining repair operation process monitoring and evaluation method provided in the embodiments of the present invention;

[0091] Figure 2 This is a schematic diagram of the structure of the pipeline overturning lining repair operation process monitoring and evaluation device provided in an embodiment of the present invention. Detailed Implementation

[0092] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.

[0093] Figure 1 A flowchart of the pipeline overturning lining repair operation process monitoring and evaluation method provided in an embodiment of the present invention is shown. See also: Figure 1 The pipeline overturning lining repair operation monitoring and evaluation method provided in this embodiment of the invention includes:

[0094] S1, the process parameter timing data is segmented using the PELT change point detection algorithm.

[0095] As an optional implementation of this invention, the segmentation of process parameter time series data using the PELT change point detection algorithm includes:

[0096] Collect the timing data of the process parameters during the inversion lining repair process;

[0097] Construct the PELT change point detection objective function:

[0098]

[0099] in, For the number of variable points, For the change point, The penalty factor is used to control the number of variable points. Let the piecewise mean squared error cost function be... Let i be the data subsequence of the i-th segment; the segmented mean square error cost function is:

[0100]

[0101] in, Let t be a subsequence of the time series from index a to b, where t is the time-th moment. Let be the observation value at time t. , where a is the index of the start time point of the segment and b is the index of the end time point of the segment.

[0102] In practice, time-series data of process parameters are collected. Output variable point Construct the PELT change point detection objective function:

[0103]

[0104] in, For the number of variable points, The penalty factor, c, is a constant that controls the number of variable points and is adjusted according to the actual situation. Let cost function be This represents the data subsequence of the i-th segment. When i=2, The data sequence representing the second segment is the subsequence of data between the time 1 moment after the first change point and the time of the second change point, ensuring that the segments do not overlap. The statistical difference between the segmented data is calculated using the mean squared error, a statistic that measures the degree of deviation between the predicted and actual values. The formula is as follows:

[0105]

[0106] in, This is a complete time series data set containing T observation points; yt is the observation value at time t. Let be a subsequence (closed interval) of the time series from index a to b; t is the time t. , where a is the index of the start time point of the segment and b is the index of the end time point of the segment.

[0107] S2, extract feature vectors of parameters at different time periods.

[0108] As an optional implementation of this invention, the extraction of feature vectors for different time periods includes:

[0109] Extract the feature vector for each time period. The feature vector includes: mean, standard deviation, upper quartile, FFT main frequency amplitude, and trend slope.

[0110] The feature matrix obtained by concatenating the sliding windows is used as input to construct the input sequence.

[0111] The extraction of feature vectors for each time period includes:

[0112]

[0113] in, For feature vectors, The mean, Standard deviation It is the upper quartile. This represents the FFT main frequency amplitude. The trend slope For segmented data, For length, For Discrete Fourier Transform, Time-centered coordinates;

[0114] The step of concatenating the sliding windows to obtain the feature matrix as input and constructing the input sequence includes:

[0115] The feature matrix is ​​obtained based on the feature vector. ;

[0116] The feature matrix is ​​used as input to construct a feature vector input sequence. :

[0117]

[0118] Where M is the total number of segments, R is the real matrix space, representing a matrix of dimension a×b, and N is the number of input sequence samples. The size of the sliding window. The number of training samples that can be generated; Concat is a vector concatenation operation.

[0119] In practice, on the one hand, for each segment Feature vector extraction and selection should consider the following dimensions. :

[0120] mean This reflects the baseline level of the segmented data. Sudden changes in the mean may indicate a change in operating conditions (such as a heating phase transition) or equipment failure (such as a pressure leak).

[0121] Standard deviation Characterizing the intensity of data fluctuations and assessing process stability. Solidification stage. A sudden increase may indicate that the resin flow is obstructed (such as a partial blockage in the pipeline).

[0122] Upper quartiles This reflects the distribution characteristics of high-value areas and is more resistant to extreme value interference than the mean. Temperatures are higher than average, indicating a risk of localized overheating.

[0123] FFT main frequency amplitude : Reveals the intensity of periodic fluctuations in data. Detects oscillations in process parameters caused by mechanical vibration;

[0124] Trend slope Quantify the direction and rate of change in data. Predict whether parameters will exceed safe limits (e.g., if the pressure slope remains positive, early intervention is necessary).

[0125] Feature vector The calculation formula is as follows:

[0126]

[0127] Among them, the segmented data is ,length , This is the Discrete Fourier Transform, used to calculate frequency domain characteristics. For time-centered coordinates.

[0128] On the other hand, by stitching together the sliding windows, the feature matrix obtained above is... As input, construct the input sequence :

[0129]

[0130] Where R is the real matrix space, representing a matrix of dimension a×b; N is the number of input sequence samples, the total number of training samples generated by the sliding window. The size of the sliding window. The number of training samples that can be generated; Concat is a vector concatenation operation that horizontally concatenates multiple feature vectors into a higher-dimensional vector. Then the concatenated vector.

[0131] S3 uses the Transformer model to predict future process parameter characteristics.

[0132] As an optional implementation of this invention, the step of using the Transformer model to predict future process parameter features includes:

[0133] The feature vector input sequence is mapped to a high-dimensional space and position information is injected.

[0134] Capture long-term dependencies through self-attention mechanisms;

[0135] The features of the self-attention output at each time step are nonlinearly transformed using a feedforward network;

[0136] Output the predicted eigenvector values ​​of future process parameters.

[0137] in,

[0138] The step of mapping the feature vector input sequence to a high-dimensional space and injecting position information includes:

[0139] Set to hidden state for:

[0140]

[0141] in, Position encoding for embedding matrix The formula is as follows:

[0142]

[0143] Among them, time step position Dimension index d model For the hidden layer dimension of the model, This is a position-encoded vector;

[0144] The method of capturing long-term dependencies through self-attention mechanisms includes:

[0145] Set the first layer Attention output:

[0146]

[0147] in, These are the query, key, and value projection matrices, respectively. For the previous step, calculate the first step that carries mixed information of temporal sequence and features. The hidden state of the layer Dimensions for each attention head;

[0148] The nonlinear transformation of the self-attention output at each time step using a feedforward network includes:

[0149] Nonlinear transformation is performed using the following formula:

[0150]

[0151] Where x is the input feature and ReLU is the activation function. , These are the weights for the first and second layers, respectively. For bias terms, For output bias;

[0152] The predicted feature vector values ​​of the output future process parameters include:

[0153] Take the hidden state at the last time step Decoded into feature vector predictions for future process segments:

[0154]

[0155] The predicted characteristics of the next process segment are: The predicted feature vector Convert to a sequence of process parameters for future time periods The formula is as follows:

[0156]

[0157] Among them, L new The length of a future time period. It is Gaussian noise. The predicted process parameter value at time point k. The mean of the generated sequence, The slope represents the linear trend.

[0158] In practice, firstly, the segmented feature vector sequence is... Mapping to a high-dimensional space and injecting positional information allows the model to perceive the temporal order of features. Hidden states. for:

[0159]

[0160] in, Position encoding for embedding matrix The formula is as follows:

[0161]

[0162] Among them, time step position Dimension index Pos represents the time step position, indicating the absolute position index in the input sequence. If the input sequence length is 10, then... ;j represents the dimension index, controlling the alternating sine / cosine pattern of the feature dimension;d model This represents the dimension of the hidden layer of the model and determines the length of the position encoding vector. If dmodel=512, then the position encoding vector is 512-dimensional. The position encoding vector is a dmodel-dimensional vector generated for each position. Even positions use sine and odd positions use cosine to form an alternating pattern.

[0163] Secondly, the self-attention mechanism dynamically allocates attention weights by calculating the correlation between each pair of time steps in the feature sequence, thereby capturing long-term dependencies. Attention output:

[0164]

[0165] in, These are the query, key, and value projection matrices, respectively. For the previous step, calculate the first step that carries mixed information of temporal sequence and features. The hidden state of the layer Dimensions for each attention head.

[0166] Then, a feedforward network is used to perform a nonlinear transformation on the features of the self-attention output at each time step to enhance the model's expressive power, as shown in the following formula:

[0167]

[0168] in, , These are the weights for the first and second layers, respectively. For bias terms, is the output bias; x represents the input feature, the output of the self-attention layer (feature vector at a single time step). ReLU stands for activation function, defined as follows: Nonlinearity is introduced to filter out negative features.

[0169] Finally, output the predicted feature vector, taking the hidden state at the last time step. (Including global timing information), decoded into feature vector predictions for future process segments:

[0170]

[0171] The predicted characteristics of the next process segment are: The predicted feature vector Convert to a sequence of process parameters for future time periods The formula is as follows:

[0172]

[0173] Among them, L new The length of a future time period. It is Gaussian noise. The predicted process parameter value at time point k. The mean of the generated sequence, The slope represents the linear trend.

[0174] S4, combined with the CoCoSo multi-index decision-making method, dynamically evaluates the process status.

[0175] As an optional implementation of this invention, the dynamic evaluation of process status using the CoCoSo multi-index decision-making method includes:

[0176] Normalize the decision matrix;

[0177] Calculate the sum of the weighted comparable sequences of process states for each time period;

[0178] Calculate the sum of geometrically comparable sequences of process states for each time period;

[0179] Calculate the relative score of the process status for each time period;

[0180] Calculate the final score for the process status at each time period.

[0181] in,

[0182] The normalization of the decision matrix includes:

[0183] Set n parameters Predicted using the Transformer model Parameters Time period Parameter values, construct decision matrix .

[0184] For decision matrix Normalize:

[0185] Among them, C ij Let j be the value of parameter j at time i. and maximum value and minimum value ;

[0186] The calculation of the sum of the weighted comparable sequences of process states for each time period includes: calculating the sum of the weighted comparable sequences using the following formula:

[0187]

[0188] in, For the first The weights of each parameter;

[0189] The calculation of the sum of geometrically comparable sequences of process states for each time period includes:

[0190] The sum of geometrically comparable sequences is calculated using the following formula:

[0191] ;

[0192] The calculation of the relative score of the process status for each time period includes: calculating the relative score using the following formula:

[0193]

[0194]

[0195]

[0196] in, This represents the arithmetic mean of the scores from the weighted sum method and the weighted product method. This represents the sum of scores from the weighted sum method and the weighted product method relative to the optimal score. This represents a trade-off between the weighted sum method and the weighted product method scores. This is the balance coefficient;

[0197] The calculation of the final score for the process status in each time period includes:

[0198]

[0199] in, This represents the score for the process status at each time period.

[0200] In practical implementation, assume there are n parameters. The above Transformer model predicts Parameters Time period Parameter values, construct decision matrix .

[0201] First, the decision matrix Normalize:

[0202]

[0203] Among them, C ij Let j be the value of parameter j at time i. and maximum value and minimum value .

[0204] Secondly, calculate the sum of the weighted comparable sequences of process states for each time period. :

[0205] in, For the first The weights of each parameter.

[0206] Next, calculate the sum of geometrically comparable sequences of process states for each time period. :

[0207] Then, the relative score of the process state for each time period is calculated. and Three scoring strategies are used to generate a relative score for the process status at each time period, as shown in the following formula:

[0208]

[0209]

[0210]

[0211] in, This represents the arithmetic mean of the scores from the weighted sum method and the weighted product method. This represents the sum of scores from the weighted sum method and the weighted product method relative to the optimal score. This represents the trade-off between the scores obtained using the weighted sum method and the weighted product method. The balance coefficient. The CoCoSo algorithm's safety and adaptability are generally set to 0.5.

[0212] Finally, calculate the final score for the process status at each time period:

[0213] in, This represents the score of the process status at each time period; the higher the score, the better the process status.

[0214] Therefore, the pipeline overturning lining repair operation process monitoring and evaluation method provided by the embodiments of the present invention proposes a PELT-based Transformer prediction model. By introducing the PELT change point detection algorithm, it accurately captures the feature vectors of parameters at different time periods, improves the prediction accuracy of process parameters, and avoids the problem of low reliability of judging process parameters solely based on human experience. Furthermore, it proposes a Transformer-CoCoSo process state time sequence prediction method. Based on accurate parameter prediction values, it introduces the multi-index decision-making CoCoSo method, which solves the problem that construction personnel cannot make accurate multi-index decisions based solely on subjective experience.

[0215] Figure 2 This diagram illustrates the structure of a pipeline overturning lining repair process monitoring and evaluation device provided in an embodiment of the present invention. This device applies the aforementioned method. The following is only a brief description of the structure of the device; for other matters not covered herein, please refer to the relevant descriptions in the above-described pipeline overturning lining repair process monitoring and evaluation method. Figure 2 The pipeline overturning lining repair operation process monitoring and evaluation device provided in this embodiment of the invention includes:

[0216] The segmentation module is used to segment the timing data of process parameters using the PELT change point detection algorithm.

[0217] The extraction module is used to extract feature vectors of parameters at different time periods;

[0218] The prediction module is used to predict future process parameter characteristics using the Transformer model;

[0219] The evaluation module is used to dynamically evaluate the process status by combining the CoCoSo multi-index decision-making method.

[0220] Therefore, the pipeline overturning lining repair operation process monitoring and evaluation device provided in this embodiment of the invention proposes a PELT-based Transformer prediction model. By introducing the PELT change point detection algorithm, it accurately captures the feature vectors of parameters at different time periods, improving the prediction accuracy of process parameters and avoiding the problem of low reliability in judging process parameters solely based on human experience. Furthermore, it proposes a Transformer-CoCoSo process state time-series prediction method. Based on accurate parameter prediction values, it introduces the multi-index decision-making CoCoSo method, solving the problem that construction personnel cannot make accurate multi-index decisions based solely on subjective experience.

[0221] The present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-described method for monitoring and evaluating the pipeline overturning lining repair process.

[0222] Therefore, the computer-readable storage medium provided in this invention provides a PELT-based Transformer prediction model. By introducing the PELT change point detection algorithm, it accurately captures the feature vectors of parameters at different time periods, improving the prediction accuracy of process parameters and avoiding the problem of low reliability in judging process parameters solely based on human experience. Furthermore, a Transformer-CoCoSo process state timing prediction method is proposed. Based on accurate parameter prediction values, it introduces the multi-index decision-making CoCoSo method, solving the problem that construction personnel cannot make accurate multi-index decisions based solely on subjective experience.

[0223] The above are merely embodiments of this application and are not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of the claims of this application.

Claims

1. A method for monitoring and evaluating the process of pipeline overturning lining repair, characterized in that, include: The time-series data of process parameters are segmented using the PELT change point detection algorithm; Extract feature vectors of parameters at different time periods; The Transformer model is used to predict future process parameter characteristics; Dynamically evaluate process status by combining the CoCoSo multi-index decision-making method; in: The segmentation of process parameter time-series data using the PELT change point detection algorithm includes: Collect the timing data of the process parameters during the inversion lining repair process; Construct the PELT change point detection objective function: in, For the number of variable points, Let's define an auxiliary variable as the variable point. This is the position before the start of the first time period. , The penalty factor is used to control the number of variable points. Let the piecewise mean squared error cost function be... Let i be the data subsequence of the i-th segment; the segmented mean square error cost function is: in, Let t be a subsequence of the time series from index a to b, where t is the time t. Let be the observation value at time t. , where a is the index of the start time point of the segment, and b is the index of the end time point of the segment; The extracted parameter feature vectors for different time periods include: Extract the feature vector for each time period. The feature vector includes: mean, standard deviation, upper quartile, FFT main frequency amplitude, and trend slope. The feature matrix obtained by concatenating the sliding windows is used as input to construct the input sequence; The dynamic evaluation of process status using the CoCoSo multi-index decision-making method includes: Normalize the decision matrix; Calculate the sum of the weighted comparable sequences of process states for each time period; Calculate the sum of geometrically comparable sequences of process states for each time period; Calculate the relative score of the process status for each time period; Calculate the final score for the process status in each time period; The normalization of the decision matrix includes: Set n parameters Predicted using the Transformer model One parameter, Time period Parameter values, construct decision matrix L new The length of a future time period; For decision matrix Normalize: Among them, C ij Let j be the value of parameter j at time i. and maximum value and minimum value .

2. The method according to claim 1, characterized in that, The extraction of feature vectors for each time period includes: in, Let d be the feature vector of the k-th time period, where d is the dimension. The mean, For segmented data, For length, Standard deviation, It is the upper quartile. This represents the FFT main frequency amplitude. This indicates that for this segment of length... The data is subjected to a discrete Fourier transform, [1] representing the first frequency component. The trend slope Time-centered coordinates; The step of concatenating the sliding windows to obtain the feature matrix as input and constructing the input sequence includes: The feature matrix is ​​obtained based on the feature vector. ; The feature matrix is ​​used as input to construct a feature vector input sequence. : Where M is the total number of segments, R is the real matrix space, representing a matrix of dimension a×b, and N is the number of input sequence samples. The size of the sliding window. The number of training samples that can be generated; Concat is a vector concatenation operation.

3. The method according to claim 2, characterized in that, The features used to predict future process parameters using the Transformer model include: The feature vector input sequence is mapped to a high-dimensional space and position information is injected. Capture long-term dependencies through self-attention mechanisms; The features of the self-attention output at each time step are nonlinearly transformed using a feedforward network; Output the predicted eigenvector values ​​of future process parameters.

4. The method according to claim 3, characterized in that, The step of mapping the feature vector input sequence to a high-dimensional space and injecting position information includes: Set to hidden state for: in, Position encoding for embedding matrix The formula is as follows: Among them, the time step position Dimension index d model For the hidden layer dimension of the model, This is a position-encoded vector; The method of capturing long-term dependencies through self-attention mechanisms includes: Set the first Layer attention output: in, These are the query, key, and value projection matrices, respectively. For the previous step, calculate the first step that carries mixed information of temporal sequence and features. The hidden state of the layer Dimensions for each attention head; The nonlinear transformation of the self-attention output at each time step using a feedforward network includes: Nonlinear transformation is performed using the following formula: Where x is the input feature and ReLU is the activation function. , These are the weights for the first and second layers, respectively. For the first-level bias term, For output bias; The predicted feature vector values ​​of the output future process parameters include: Take the hidden state at the last time step Decoded into feature vector predictions for future process segments: The predicted characteristics of the next process segment are: The predicted feature vector Convert to a sequence of process parameters for future time periods The formula is as follows: in, It is Gaussian noise. The predicted process parameter value at time point k. The mean of the generated sequence, The slope of the linear trend. Standard deviation, It follows a pattern with a mean of 0 and a variance of . The Gaussian distribution.

5. The method according to claim 4, characterized in that, The calculation of the sum of the weighted comparable sequences of process states for each time period includes: calculating the sum of the weighted comparable sequences using the following formula: in, For the first The weights of each parameter, The weighted arithmetic sum of comparable sequences in the i-th time period; The calculation of the sum of geometrically comparable sequences of process states for each time period includes: The sum of geometrically comparable sequences is calculated using the following formula: ； in, Let be the weighted geometrically comparable sequence sum for the i-th time period; The calculation of the relative score of the process status for each time period includes: calculating the relative score using the following formula: in, This represents the arithmetic mean of the scores from the weighted sum method and the weighted product method. This represents the sum of scores from the weighted sum method and the weighted product method relative to the optimal score. This represents a trade-off between the weighted sum method and the weighted product method scores. This is the balance coefficient; The calculation of the final score for the process status in each time period includes: in, This represents the score for the process status at each time period.

6. A device for monitoring and evaluating the process of pipeline overturning lining repair, characterized in that, include: The segmentation module is used to segment the timing data of process parameters using the PELT change point detection algorithm. The extraction module is used to extract feature vectors of parameters at different time periods; The prediction module is used to predict future process parameter characteristics using the Transformer model; The evaluation module is used to dynamically evaluate the process status by combining the CoCoSo multi-index decision-making method; in: The segmentation of process parameter time-series data using the PELT change point detection algorithm includes: Collect the timing data of the process parameters during the inversion lining repair process; Construct the PELT change point detection objective function: in, For the number of variable points, Let's define an auxiliary variable as the variable point. This is the position before the start of the first time period. , The penalty factor is used to control the number of variable points. Let the piecewise mean squared error cost function be... Let i be the data subsequence of the i-th segment; the segmented mean square error cost function is: in, Let t be a subsequence of the time series from index a to b, where t is the time t. Let be the observation value at time t. , where a is the index of the start time point of the segment, and b is the index of the end time point of the segment; The extracted parameter feature vectors for different time periods include: Extract the feature vector for each time period. The feature vector includes: mean, standard deviation, upper quartile, FFT main frequency amplitude, and trend slope. The feature matrix obtained by concatenating the sliding windows is used as input to construct the input sequence; The dynamic evaluation of process status using the CoCoSo multi-index decision-making method includes: Normalize the decision matrix; Calculate the sum of the weighted comparable sequences of process states for each time period; Calculate the sum of geometrically comparable sequences of process states for each time period; Calculate the relative score of the process status for each time period; Calculate the final score for the process status in each time period; The normalization of the decision matrix includes: Set n parameters Predicted using the Transformer model One parameter, Time period Parameter values, construct decision matrix L new The length of a future time period; For decision matrix Normalize: Among them, C ij Let j be the value of parameter j at time i. and maximum value and minimum value .

7. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by the processor, the program implements the pipeline overturning lining repair operation process monitoring and evaluation method as described in any one of claims 1 to 5.

Images