A neural network-based method for predicting substation settlement in collapsible loess areas

Abstract

This invention provides a neural network-based method for predicting substation settlement in collapsible loess areas, belonging to the field of geological engineering monitoring technology. The invention obtains soil mechanics data by conducting collapsibility experiments on each sub-region, and then analyzes settlement characteristic indicators such as collapsibility, bearing capacity influence, and deformation stability. Using soil mechanics data as input and settlement characteristic indicators as output, a prediction model is established. The model performance is optimized using particle swarm optimization and a loss function. Finally, a fitness function is constructed based on the settlement characteristic indicators to calculate the settlement fitness value of each sub-region. The settlement amount is assessed by combining load time-series data, and a settlement threshold is set. Sub-regions exceeding the threshold are marked as having risk. This invention, by establishing a settlement characteristic indicator prediction model and a dynamic monitoring mechanism, achieves effective assessment and control of settlement risk in loess areas where substations are located.

Description

Technical Field

[0001] This invention relates to the field of geological engineering monitoring technology, specifically to a method for predicting substation settlement in collapsible loess areas based on neural networks. Background Technology

[0002] With the continuous progress of society and the rapid development of power construction, the continuous adoption of various new technologies and equipment has brought more and more new challenges and opportunities to power infrastructure projects. Collapsible loess collapse has become a typical and prominent problem restricting the deformation and stability of buildings and structures in loess areas due to its qualitative suddenness and quantitative discontinuity and non-negligibility. Traditional settlement prediction methods mainly rely on empirical formulas and simplified theoretical models. These methods have certain limitations in dealing with complex conditions, such as settlement prediction of collapsible soil and load changes. Due to the non-uniformity and variability of soil, traditional methods often fail to accurately reflect the actual situation, resulting in insufficient reliability of prediction results.

[0003] In existing technologies, when measuring substation settlement in collapsible loess areas, the location of the measurement benchmark is often unstable, which can easily lead to deviations in the measured settlement. Using unstable benchmarks for data collection affects the accuracy of prediction. Furthermore, traditional methods only consider a limited set of soil mechanical parameters, ignoring the comprehensive characteristics of the soil. They also rely on simple empirical formulas, which cannot fully utilize complex data features. In addition, existing technologies often ignore the characteristics of collapsible loess areas, overlook measurement errors caused by environmental factors, and fail to predict settlement for a specific area of ​​the substation, thus lacking risk management capabilities.

[0004] Therefore, it is necessary to provide a neural network-based method for predicting substation settlement in collapsible loess areas to solve the aforementioned problem.

[0005] The information disclosed in the background section is only intended to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0006] The purpose of this invention is to provide a method for predicting substation settlement in collapsible loess areas based on neural networks, so as to solve the problems mentioned in the background art.

[0007] To achieve the above objectives, the present invention provides the following technical solution:

[0008] A method for predicting substation settlement in collapsible loess areas based on neural networks, comprising the following steps:

[0009] Step 1: Divide the loess area where the target substation is located into regions and set up multiple initial measurement benchmarks in each sub-region. Perform stability analysis on each initial measurement benchmark and retain the benchmarks that meet the stability standards as the final measurement benchmarks. Collect substation settlement time series data and load time series data for each sub-region at the final measurement benchmarks.

[0010] Step 2: Conduct collapsibility tests on each sub-region to obtain relevant soil mechanics data, and perform cross-parameter correlation based on the soil mechanics data obtained from the experiments to obtain multiple settlement characteristic indicators of the substation. The settlement characteristic indicators include collapsibility indicators, bearing capacity influence indicators, and deformation stability indicators.

[0011] Step 3: Using the soil mechanics data of each sub-region as input and the substation settlement characteristic index of each sub-region as output, establish a substation settlement characteristic index prediction model. Based on the particle swarm optimization algorithm, search and adjust the output results of the prediction model, and optimize the model based on the loss function.

[0012] Step 4: Construct a settlement fitness function based on the settlement characteristic index output by the substation settlement prediction model to calculate the settlement fitness value of each sub-region. Combine the load time series data of each sub-region to obtain the specific settlement amount of each sub-region, and set a settlement threshold. Mark the risk of sub-regions of the substation that exceed the settlement threshold.

[0013] Furthermore, the loess area where the target substation is located is divided into regions, and the initial measurement benchmark points in each sub-region are subjected to stability analysis using both the limit error test method and the average gap method to determine the final measurement benchmark points. The method used is as follows:

[0014] Based on the design and function of the substation, sub-regions related to equipment layout were divided. Based on the equipment layout and load conditions of each sub-region, the sub-regions were divided into heavy-load areas and light-load areas. The initial reference measurement points were arranged using the uniform distribution method in both types of areas, and a square measurement grid was used as the initial reference measurement points for measurement. The number of measurement grids was increased in the heavy-load area to better monitor the impact of load changes on settlement.

[0015] Set the observation period for the initial measurement benchmark, use the limit error test method to perform independence analysis on each initial measurement benchmark, record the initial elevation of the initial measurement benchmark and the elevation of each observation, calculate the elevation change for each observation, compare the absolute value of the elevation change for each observation with the preset limit error value, if there is a case where the absolute value of the elevation change is greater than the preset limit error value within the observation period, then the initial measurement benchmark is determined to be a moving point and is removed.

[0016] The average gap method is used to compare the two observation data of the initial measurement benchmark point. The elevation correction value of each benchmark point in the two observations is calculated, and the gap deviation value between these observation values ​​is obtained. By calculating the variance of the gap deviation value, the ratio of the variance of the gap deviation value to the variance of the overall observation is defined as the F value. If the calculated F value is less than the preset significance level threshold, the benchmark point is considered to have not changed significantly, indicating that its stability is high; otherwise, there is instability or a moving point, and it is removed.

[0017] The initial measurement reference point that simultaneously meets the standards of the limit test method and the average gap method is used as the final measurement reference point.

[0018] Furthermore, collapsibility experiments were conducted using the acquired settlement time series data to obtain relevant soil mechanics data. This soil mechanics data included collapsibility fitness values, compression modulus, internal friction angle, cohesion, and moisture content. The method used was as follows:

[0019] Settlement data were collected at the established final measurement benchmark point, and a collapsibility test was conducted. At the benchmark point, the initial height of the soil sample was first measured. During the experiment, a predetermined pressure was applied to the soil sample, and while maintaining this pressure, it was immersed in water. The settlement of the soil sample after immersion was recorded. The ratio of the settlement after immersion to the initial height of the soil sample is the collapsibility coefficient of the soil. ;

[0020] Different levels of vertical pressure were applied to the soil sample at the final measurement benchmark. Under each level of pressure, the strain of the soil sample was measured. The ratio of the stress increment to the strain increment is the compressibility modulus of the soil. ;

[0021] Direct shear tests were conducted on soil samples at the final measurement benchmark to determine the shear strength under different confining pressures. The experimental data were then fitted with the Mohr-Coulomb failure criterion curve, and the relationship was derived from... The internal friction angle is obtained by extracting shear stress and normal stress. ,in This represents the shear stress of the soil at the final measurement benchmark point. To finally measure the cohesion of the soil at the benchmark point, It is the normal stress of the soil at the final measurement benchmark point;

[0022] Similarly, by conducting direct shear tests on soil samples at the final measurement benchmark, cohesion is calculated in combination with soil shear strength data, and normal stress is calculated using strength data measured under different confining pressures and shear conditions.

[0023] The moisture content of a soil sample is measured using a humidity sensor; the ratio of water weight to dry soil weight is the moisture content. .

[0024] Furthermore, the soil mechanical parameters obtained from the experiment were combined across channels to obtain substation settlement characteristic indices. These indices include collapsibility indices, bearing capacity influence indices, and deformation stability indices. The method used is as follows:

[0025] A spatial-feature correlation matrix is ​​established, where the spatial dimension is based on the three-dimensional deployment coordinates of the final measurement reference point. Original parameter channels and cross-parameter correlation channels are defined. The original parameter channels include collapsibility fitness values, compression modulus, internal friction angle, cohesion, and moisture content. The cross-parameter correlation channels construct substation settlement characteristic indicators reflecting the substation settlement situation. The cross-parameter correlation feature set is defined as follows. Each element This represents a cross-parameter correlation pair, which corresponds to a substation settlement characteristic index, including collapsibility index. : Bearing capacity influencing indicators : Deformation stability index : .

[0026] Furthermore, a prediction model for substation settlement characteristic indicators was established and optimized using the following method:

[0027] The obtained collapsibility coefficient, compression modulus, internal friction angle, cohesion, and moisture content are normalized and formed into feature vectors. A substation settlement prediction model is constructed based on a neural network structure. The substation settlement prediction model includes an input layer, a hidden layer, and an output layer. The input layer is responsible for receiving feature vectors. The hidden layer is an intermediate layer in the neural network between the input layer and the output layer. It is responsible for feature extraction and information processing of the input feature vectors. In addition, a nonlinear activation function is used in the hidden layer to capture complex patterns. The output layer generates the final prediction result of the model. The output layer has three nodes to output three substation settlement characteristic indicators.

[0028] The particle swarm optimization algorithm is used to search and adjust the substation settlement characteristic indicators output by the prediction model. A particle swarm is initialized, with each particle representing a substation settlement characteristic indicator in a sub-region. The position and velocity of the particles are set, where position represents a combination of weights and thresholds in the prediction model, and velocity defines the particle's direction and stride in the search space. At the beginning of each iteration, a set of weights and thresholds is randomly generated as the initial positions of the particles. The initial positions of the particles are generated based on a normal distribution to ensure that the particles are uniformly distributed throughout the search space. The mean squared error is used as the loss function to calculate the loss of the prediction model, based on the following formula:

[0029] ;

[0030] in, This represents the calculated mean square error. This represents the number of sub-regions of the substation. Indicates the first The subregion containing the nth particle The actual values ​​of the settlement characteristic indicators of each substation Indicates the first The subregion containing the nth particle Predicted values ​​of substation settlement characteristic indicators The index is for the sub-region, and , It is an index of power station settlement characteristic indicators in the sub-region, and hour, This refers to the collapsibility index, which is a characteristic index of substation settlement. hour, Indicates the load-bearing capacity impact index, When, it represents the deformation stability index ;

[0031] The formula used to update the particle's velocity and position is:

[0032] ;

[0033] ;

[0034] in, Indicates the first Individual particles The speed of time, Indicates the first Individual particles Location at any given moment For inertial weights, , It is a learning factor. , It is a random number. Indicates the first The optimal position of each particle Indicates the globally optimal position. For the first Individual particles Location at any moment For the first Individual particles The speed of time;

[0035] The process of repeatedly evaluating the loss and updating the position and velocity of the particles is repeated until the maximum number of iterations is reached. The optimal particle position found will be used to update the weights and thresholds of the neural network model. By retraining the model, its performance in predicting substation settlement characteristics is ensured to be at its best.

[0036] Furthermore, a settlement fitness function is constructed to calculate the settlement fitness value of each sub-region. The method used is as follows:

[0037] A settlement fitness function is constructed based on the settlement formula of the compression index, and the settlement characteristic indices of each sub-region predicted by the substation settlement characteristic index prediction model are substituted into it. The formula used is as follows:

[0038] ;

[0039] in, This represents the settlement fitness value of the sub-region. It is the initial void ratio of the compressible soil layer. To compress the thickness of the soil layer, The compression index is... The initial effective stress of the soil layer, It is the additional stress of the soil layer.

[0040] Furthermore, the specific settlement amounts for each sub-region are obtained, and settlement thresholds are set. Sub-regions of substations exceeding these settlement thresholds are then risk-marked. The method used is as follows:

[0041] The substation settlement fitness values ​​for each sub-region are obtained based on the settlement fitness function. Combined with the corresponding load time series data for each sub-region, the substation settlement is calculated within each time period, using the following formula:

[0042] ;

[0043] in, Indicates the first In each sub-region Settlement within a given time period For the first Substation settlement fitness values ​​in each sub-region For the first In each sub-region The load experienced at any given time;

[0044] The settlement time series data of each sub-region of the substation over the past year were summarized, and the average settlement of all sub-regions was calculated as the settlement threshold. ,when If the risk is not specified, the sub-region is marked with a risk marker, and the settlement amount of the sub-region at that time is recorded.

[0045] Compared with the prior art, the beneficial effects of the present invention are:

[0046] This invention significantly improves the accuracy and reliability of substation settlement prediction in collapsible loess areas. By establishing stable measurement benchmarks, accurate relevant data is obtained. Through a neural network-based model combined with particle swarm optimization, the shortcomings of traditional methods in data processing and model parameter optimization are solved, ensuring high-quality data input and enhanced model performance. Furthermore, by dividing the substation into regions, establishing a settlement fitness function, and setting a settlement threshold, this invention can effectively conduct regional risk assessments, promptly identify potential safety hazards, and thus improve the safety and management efficiency of substations. Attached Figure Description

[0047] Figure 1 This is a schematic diagram of the overall method flow of the present invention. Detailed Implementation

[0048] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments.

[0049] It should be noted that, unless otherwise defined, the technical or scientific terms used in this invention should have the ordinary meaning understood by one of ordinary skill in the art to which this invention pertains. The terms "first," "second," and similar terms used in this invention do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Terms such as "comprising" or "including" mean that the element or object preceding the word encompasses the elements or objects listed following the word and their equivalents, without excluding other elements or objects. Terms such as "connected" or "linked" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. Terms such as "upper," "lower," "left," and "right" are used only to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0050] Example:

[0051] Please see Figure 1 A method for predicting substation settlement in collapsible loess areas based on neural networks, comprising the following steps:

[0052] Step 1: Divide the loess area where the target substation is located into regions and set multiple initial measurement benchmarks in each sub-region. Perform stability analysis on each initial measurement benchmark and retain the benchmarks that meet the stability standards as the final measurement benchmarks. Collect substation settlement time series data and load time series data for each sub-region at the final measurement benchmarks.

[0053] In a specific embodiment of the present invention, we take into account the influence of factors such as the structure and water content of collapsible loess areas. The settlement behavior of each region may be significantly different. After dividing the region, the characteristics of each sub-region can be analyzed independently, thereby improving the accuracy and reliability of settlement prediction. Furthermore, based on the regional division, collecting time series data of each sub-region can provide richer and more detailed information for subsequent data analysis and training of neural network models, thereby improving the performance of the prediction model.

[0054] The reason we set up initial measurement points and conducted stability analysis in each sub-region is that the measurement points set up in collapsible loess areas are easily affected by the settlement of the soil itself, which can lead to errors in the measured settlement of the substation. Ensuring that the measurement benchmark is not affected by soil settlement is the key to obtaining accurate settlement data and avoiding data errors caused by the movement of the benchmark itself.

[0055] It should be noted that the core idea of ​​the limit deviation test method is to quantitatively compare the absolute value of the elevation change in each observation by using a preset limit deviation value to assess the stability of the measurement point. Setting the limit deviation value involves statistically analyzing historical observation data to calculate the standard deviation or average change in elevation. Based on the calculation results, the limit deviation value is set as the average value. The standard deviation is set to a multiple of the standard deviation, and a specific limit range is defined as follows: In this embodiment, the preset limit value is actually a specific threshold range that we have calculated. In subsequent observations, if the absolute value of a certain elevation change exceeds this limit range, the measurement point is determined to be an unstable measurement point and needs to be removed.

[0056] The core idea of ​​the average gap method is to compare the elevation correction values ​​of the initial measurement benchmark point in two observation periods, calculate its gap deviation value, and analyze the variance of these deviation values ​​to assess the stability of the measurement point. From a statistical perspective, this method compares the variance of the gap deviation value with the variance of the overall observation and uses the F test to determine whether the benchmark point has changed significantly. If the calculated F value is less than the preset significance level threshold, the benchmark point is considered to have not changed significantly, indicating that its stability is high; otherwise, there is instability or a moving point that needs to be removed.

[0057] Furthermore, the loess area where the target substation is located is divided into regions, and the initial measurement benchmark points in each sub-region are subjected to stability analysis using both the limit error test method and the average gap method to determine the final measurement benchmark points. The method used is as follows:

[0058] Based on the design and function of the substation, sub-regions related to equipment layout were divided. Based on the equipment layout and load conditions of each sub-region, the sub-regions were divided into heavy-load areas and light-load areas. The initial reference measurement points were arranged using the uniform distribution method in both types of areas, and a square measurement grid was used as the initial reference measurement points for measurement. The number of measurement grids was increased in the heavy-load area to better monitor the impact of load changes on settlement.

[0059] Set the observation period for the initial measurement benchmark, use the limit error test method to perform independence analysis on each initial measurement benchmark, record the initial elevation of the initial measurement benchmark and the elevation of each observation, calculate the elevation change for each observation, compare the absolute value of the elevation change for each observation with the preset limit error value, if there is a case where the absolute value of the elevation change is greater than the preset limit error value within the observation period, then the initial measurement benchmark is determined to be a moving point and is removed.

[0060] The average gap method is used to compare the two observation data of the initial measurement benchmark point. The elevation correction value of each benchmark point in the two observations is calculated, and the gap deviation value between these observation values ​​is obtained. By calculating the variance of the gap deviation value, the ratio of the variance of the gap deviation value to the variance of the overall observation is defined as the F value. If the calculated F value is less than the preset significance level threshold, the benchmark point is considered to have not changed significantly, indicating that its stability is high; otherwise, there is instability or a moving point, and it is removed.

[0061] The initial measurement reference point that simultaneously meets the standards of the limit test method and the average gap method is used as the final measurement reference point.

[0062] Step 2: Conduct collapsibility tests on each sub-region to obtain relevant soil mechanics data, and perform cross-parameter correlation based on the soil mechanics data obtained from the experiments to obtain multiple settlement characteristic indicators of the substation. The settlement characteristic indicators include collapsibility indicators, bearing capacity influence indicators, and deformation stability indicators.

[0063] In a specific embodiment of the present invention, by combining settlement time series data with soil mechanics data obtained from collapsibility experiments, the soil characteristics of collapsible loess areas can be comprehensively quantified and analyzed, thereby accurately predicting the settlement behavior of substations. These soil parameters, such as collapsibility coefficient, compression modulus, internal friction angle, cohesion and water content, provide key input data for establishing a neural network model, making the training and verification of the model more accurate, and enhancing the understanding and prediction ability of complex soil behavior.

[0064] Furthermore, collapsibility experiments were conducted using the acquired settlement time series data to obtain relevant soil mechanics data. This soil mechanics data included collapsibility fitness values, compression modulus, internal friction angle, cohesion, and moisture content. The method used was as follows:

[0065] Settlement data were collected at the established final measurement benchmark point, and a collapsibility test was conducted. At the benchmark point, the initial height of the soil sample was first measured. During the experiment, a predetermined pressure was applied to the soil sample, and while maintaining this pressure, it was immersed in water. The settlement of the soil sample after immersion was recorded. The ratio of the settlement after immersion to the initial height of the soil sample is the collapsibility coefficient of the soil. ;

[0066] Different levels of vertical pressure were applied to the soil sample at the final measurement benchmark. Under each level of pressure, the strain of the soil sample was measured. The ratio of the stress increment to the strain increment is the compressibility modulus of the soil. ;

[0067] Direct shear tests were conducted on soil samples at the final measurement benchmark to determine the shear strength under different confining pressures. The experimental data were then fitted with the Mohr-Coulomb failure criterion curve, and the relationship was derived from... The internal friction angle is obtained by extracting shear stress and normal stress. ,in This represents the shear stress of the soil at the final measurement benchmark point. To finally measure the cohesion of the soil at the benchmark point, It is the normal stress of the soil at the final measurement benchmark point;

[0068] Similarly, by conducting direct shear tests on soil samples at the final measurement benchmark, cohesion is calculated in combination with soil shear strength data, and normal stress is calculated using strength data measured under different confining pressures and shear conditions.

[0069] The moisture content of a soil sample is measured using a humidity sensor; the ratio of water weight to dry soil weight is the moisture content. .

[0070] It should be noted that by establishing a spatial-feature correlation matrix and combining cross-channel features, the settlement characteristic indicators of substations can be comprehensively analyzed and evaluated. The importance of this method lies in integrating the correlations between different soil mechanical parameters to more accurately predict settlement behavior, ensuring the design and operational safety of the project. Among these, the collapsibility index... Combining the collapsibility coefficient and moisture content, this reflects the soil's settlement potential when submerged in water. A larger value indicates that the soil in that area is sensitive to changes in moisture and may experience significant settlement. A higher value also indicates stronger collapsibility, potentially requiring improved foundation design or drainage measures. (Bearing capacity influence index) Combining the internal friction angle and cohesion, the bearing capacity and shear strength of the soil are evaluated. Higher values ​​indicate higher bearing capacity and shear strength, which is beneficial for supporting the substation structure. Deformation stability index... Combining compression modulus, cohesion, internal friction angle, and collapsibility coefficient reflects the overall deformation stability of soil under external forces. The larger the value, the stronger the soil deformation stability, which means that the structure has a stronger ability to maintain stability under soil deformation. Since analyzing a single soil mechanical parameter may not fully reflect the overall performance of the soil, and the uncertainty of soil performance comes from various factors, including soil sampling and experimental errors, combining features can measure the settlement characteristics of soil with higher confidence and reduce the impact of individual parameter errors on the overall assessment.

[0071] Therefore, it is necessary to perform cross-channel characteristic combination of the soil mechanical parameters obtained from the experiment to obtain the substation settlement characteristic index. The substation settlement characteristic index includes collapsibility index, bearing capacity influence index, and deformation stability index. The method used is as follows:

[0072] A spatial-feature correlation matrix is ​​established, where the spatial dimension is based on the three-dimensional deployment coordinates of the final measurement reference point. Original parameter channels and cross-parameter correlation channels are defined. The original parameter channels include collapsibility fitness values, compression modulus, internal friction angle, cohesion, and moisture content. The cross-parameter correlation channels construct substation settlement characteristic indicators reflecting the substation settlement situation. The cross-parameter correlation feature set is defined as follows. Each element This represents a cross-parameter correlation pair, which corresponds to a substation settlement characteristic index, including collapsibility index. : Bearing capacity influencing indicators : Deformation stability index : .

[0073] Step 3: Using the soil mechanics data of each sub-region as input and the substation settlement characteristic index of each sub-region as output, establish a substation settlement characteristic index prediction model. Based on the particle swarm optimization algorithm, search and adjust the output results of the prediction model, and optimize the model based on the loss function.

[0074] In a specific embodiment of the present invention, the soil mechanics data of each sub-region we input here changes over time. Since the settlement of the substation is not completed at once, but is a slow process, we normalize the soil mechanics data of each sub-region to form a feature vector. This feature vector can change over time. By dynamically adjusting the model weights and thresholds through particle swarm optimization, the prediction accuracy and model stability are improved, so that the substation settlement feature index prediction model can predict the settlement feature index of each sub-region in each time period.

[0075] Furthermore, a prediction model for substation settlement characteristic indicators was established and optimized using the following method:

[0076] The obtained collapsibility coefficient, compression modulus, internal friction angle, cohesion, and moisture content are normalized and formed into feature vectors. A substation settlement prediction model is constructed based on a neural network structure. The substation settlement prediction model includes an input layer, a hidden layer, and an output layer. The input layer is responsible for receiving feature vectors. The hidden layer is an intermediate layer in the neural network between the input layer and the output layer. It is responsible for feature extraction and information processing of the input feature vectors. In addition, a nonlinear activation function is used in the hidden layer to capture complex patterns. The output layer generates the final prediction result of the model. The output layer has three nodes to output three substation settlement characteristic indicators.

[0077] The particle swarm optimization algorithm is used to search and adjust the substation settlement characteristic indicators output by the prediction model. A particle swarm is initialized, with each particle representing a substation settlement characteristic indicator in a sub-region. The position and velocity of the particles are set, where position represents a combination of weights and thresholds in the prediction model, and velocity defines the particle's direction and stride in the search space. At the beginning of each iteration, a set of weights and thresholds is randomly generated as the initial positions of the particles. The initial positions of the particles are generated based on a normal distribution to ensure that the particles are uniformly distributed throughout the search space. The mean squared error is used as the loss function to calculate the loss of the prediction model, based on the following formula:

[0078] ;

[0079] in, This represents the calculated mean square error. This represents the number of sub-regions of the substation. Indicates the first The subregion containing the nth particle The actual values ​​of the settlement characteristic indicators of each substation Indicates the first The subregion containing the nth particle Predicted values ​​of substation settlement characteristic indicators The index is for the sub-region, and , It is an index of power station settlement characteristic indicators in the sub-region, and hour, This refers to the collapsibility index, which is a characteristic index of substation settlement. hour, Indicates the load-bearing capacity impact index, When, it represents the deformation stability index ;

[0080] The formula used to update the particle's velocity and position is:

[0081] ;

[0082] ;

[0083] in, Indicates the first Individual particles The speed of time, Indicates the first Individual particles The position of time - For inertial weights, , It is a learning factor. , It is a random number. Indicates the first The optimal position of each particle Indicates the globally optimal position. For the first Individual particles Location at any moment For the first Individual particles The speed of time;

[0084] The process of repeatedly evaluating the loss and updating the position and velocity of the particles is repeated until the maximum number of iterations is reached. The optimal particle position found will be used to update the weights and thresholds of the neural network model. By retraining the model, its performance in predicting substation settlement characteristics is ensured to be at its best.

[0085] Step 4: Construct a settlement fitness function based on the settlement characteristic index output by the substation settlement prediction model to calculate the settlement fitness value of each sub-region. Combine the load time series data of each sub-region to obtain the specific settlement amount of each sub-region, and set a settlement threshold. Mark the risk of sub-regions of the substation that exceed the settlement threshold.

[0086] In a specific embodiment of this invention, we construct a settlement fitness function based on a settlement formula using the compression index to calculate the settlement fitness value of each sub-region. This method can combine the output of the substation settlement characteristic index prediction model to quantify the settlement performance of different sub-regions. By incorporating key parameters such as the initial void ratio, thickness, compression index, and effective stress of the compressible soil layer into the calculation, we can accurately assess the potential risks of settlement, helping engineers identify and optimize weak points in the design, thereby ensuring the safety and stability of the substation in collapsible loess areas and greatly improving the scientific accuracy and reliability of the prediction.

[0087] Furthermore, a settlement fitness function is constructed to calculate the settlement fitness value of each sub-region. The method used is as follows:

[0088] A settlement fitness function is constructed based on the settlement formula of the compression index, and the settlement characteristic indices of each sub-region predicted by the substation settlement characteristic index prediction model are substituted into it. The formula used is as follows:

[0089] ;

[0090] in, This represents the settlement fitness value of the sub-region. It is the initial void ratio of the compressible soil layer. To compress the thickness of the soil layer, The compression index is... The initial effective stress of the soil layer, It is the additional stress of the soil layer.

[0091] It should be noted that in the formulas for calculating the settlement fitness values ​​of each sub-region, the collapsibility index... It is a settlement characteristic index related to the initial void ratio of compressible soil layers. As the value increases, the settlement fitness value It increases, here Indicates the initial void ratio of the compressible soil layer The sensitivity coefficient of the initial void ratio indicates that the soil layer has higher porosity, which will increase the settlement fitness value. Increased because more pores may compress under stress; load-bearing capacity affects indicators. It is the additional stress of the soil layer Relevant settlement characteristic indicators, The value increases, It will increase. It is the sensitivity coefficient to the additional stress of the soil layer. Get bigger It becomes larger, and as the additional stress in the soil layer in the denominator increases, The overall size decreases, thus making An increase indicates an increase in the additional stress on the soil layer, which will increase the settlement tolerance value, meaning the soil is bearing a greater external load; deformation stability index It is the thickness of the compressed soil layer. The sensitivity coefficient, The value increases, It will decrease because the thicker the soil layer, the lower the settlement fitness value usually is. A thicker soil layer can disperse the applied pressure and reduce the settlement per unit area.

[0092] Furthermore, the specific settlement amounts for each sub-region are obtained, and settlement thresholds are set. Sub-regions of substations exceeding these settlement thresholds are then risk-marked. The method used is as follows:

[0093] The substation settlement fitness values ​​for each sub-region are obtained based on the settlement fitness function. Combined with the corresponding load time series data for each sub-region, the substation settlement is calculated within each time period, using the following formula:

[0094] ;

[0095] in, Indicates the first In each sub-region Settlement within a given time period For the first Substation settlement fitness values ​​in each sub-region For the first In each sub-region The load experienced at any given time;

[0096] The settlement time series data of each sub-region of the substation over the past year were summarized, and the average settlement of all sub-regions was calculated as the settlement threshold. ,when If the risk is detected, the sub-area is marked and the settlement amount of the sub-area is recorded at that time.

[0097] It should be noted that setting the settlement calculation formula directly links the settlement amount with the settlement fitness value and load changes, which can reflect the settlement status of each sub-area in real time at different time periods. Continuous monitoring and analysis of settlement data can, on the one hand, promptly identify and repair problems that may cause structural safety hazards, ensure the safe operation of the substation, and avoid equipment damage or failure caused by settlement. On the other hand, it can also predict the overall settlement of the substation, providing data basis for the maintenance and management of the substation, allowing for timely adjustment of maintenance plans, reducing operating costs, and improving the service life and operating efficiency of facilities.

[0098] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0099] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented in software, the above embodiments can be implemented, in whole or in part, as a computer program product. Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented by electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution.

[0100] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment, depending on actual needs.

[0101] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

Claims

1. A neural network-based substation settlement prediction method in collapsible loess areas, characterized by, The specific steps include: Step 1: The loess region where the target substation is located is divided into sub-regions, and a plurality of initial measurement reference points are set in each sub-region. The stability of each initial measurement reference point is analyzed, and the reference points that meet the stability standard are reserved as final measurement reference points. The settlement time series data and load time series data of each sub-region are collected at the final measurement reference points; Step 2: A collapsibility experiment is performed on each sub-region to obtain related soil mechanical data, and cross-parameter correlation is performed based on the soil mechanical data obtained by the experiment to obtain a plurality of settlement characteristic indexes of the substation, including a collapsibility index, a bearing capacity influence index, and a deformation stability index; Step 3: The soil mechanical data of each sub-region is input, and the settlement characteristic indexes of the substation in each sub-region are output. A substation settlement characteristic index prediction model is established. The output results of the prediction model are searched and adjusted based on a particle swarm optimization algorithm, and the model is optimized based on a loss function; Step 4: A settlement fitness function is constructed based on the settlement characteristic indexes output by the substation settlement prediction model to calculate the settlement fitness values of each sub-region. The specific settlement amount of each sub-region is obtained in combination with the load time series data of each sub-region, and a settlement threshold is set. The sub-regions of the substation that exceed the settlement threshold are marked as risky; The soil mechanical parameters obtained by the experiment are combined across channels to obtain substation settlement characteristic indexes, including a collapsibility index, a bearing capacity influence index, and a deformation stability index. The method used is: A space-feature correlation matrix is established, wherein a space dimension is established based on three-dimensional deployment coordinates of final measurement reference points, original parameter channels and cross-parameter correlation channels are set, wherein the original parameter channels include collapsibility, compression modulus, internal friction angle, cohesion and water content, the cross-parameter correlation channels construct substation settlement feature indexes reflecting substation settlement, and a cross-parameter correlation feature set is set as , each element represents a cross-parameter correlation pair, and sequentially corresponds to a substation settlement feature index, which includes a collapsibility index : , a bearing capacity influence index : , and a deformation stability index : , wherein is a collapsibility coefficient of the soil, is a water content of the soil, is an internal friction angle of the soil, is a cohesion of the soil at the final measurement reference point, and is a compression modulus of the soil.

2. The neural network-based substation settlement prediction method in a collapsible loess area according to claim 1, characterized in that, The loess region where the target substation is located is divided into sub-regions, and the stability of the initial measurement reference points in each sub-region is analyzed using the limit difference test method and the average gap method. The final measurement reference points are determined. The method used is: According to the design and function of the substation, sub-regions related to equipment layout are divided. Based on the equipment layout and load of each sub-region, the sub-regions are divided into heavy load areas and light load areas. Both types of areas use uniform distribution method to arrange initial reference measurement points, and square measurement grids are used as initial reference measurement points for measurement. The number of measurement grids is increased in heavy load areas to better monitor the impact of load changes on settlement; The observation period of the initial measurement reference points is set. The independence of each initial measurement reference point is analyzed using the limit difference test method. The initial elevation of the initial measurement reference point and the elevation of each observation are recorded. The elevation change of each observation is calculated. The absolute value of the elevation change of each observation is compared with the preset limit difference value. If the absolute value of the elevation change is greater than the preset limit difference value within the observation period, the initial measurement reference point is determined as a moving point and is removed. The two-period observation data of the initial measurement reference points are compared using the average interval method, the height correction values of each reference point in the two observations are calculated, and the interval deviation values between the observation values are calculated. By calculating the variance of the interval deviation values, the ratio of the variance of the interval deviation values to the overall observation is defined as the F value. If the calculated F value is less than the preset critical value of the significance level, it is considered that the reference point has not changed significantly, indicating that its stability is higher; On the contrary, there is instability or a moving point, which is removed; The initial measurement reference points that meet the limit difference test method and the average interval method standards are used as the final measurement reference points.

3. The neural network-based substation settlement prediction method in a collapsible loess area according to claim 2, characterized in that, The obtained settlement time series data are used for collapsibility experiments to obtain relevant soil mechanical data, including collapsibility fitness values, compression modulus, internal friction angle, cohesion, and water content, according to the following method: The settlement data is collected and the collapsibility experiment is conducted at the determined final measuring datum point. The initial height of the soil sample is measured at the final measuring datum point first, and the predetermined pressure is applied to the soil sample in the experiment. The settlement of the soil sample after immersion is recorded under the condition of keeping the pressure, and the ratio of the settlement after immersion to the initial height of the soil sample is the collapsibility coefficient of the soil body ; The soil sample at the final measuring reference point is subjected to different levels of vertical pressure, and at each level of pressure, the strain of the soil sample is measured, wherein the ratio of the stress increment to the strain increment is the compression modulus of the soil body ; Direct shear tests were conducted on soil samples at the final measurement benchmark to determine the shear strength under different confining pressures. The experimental data were then fitted with the Mohr-Coulomb failure criterion curve, and the relationship was derived from... The internal friction angle is obtained by extracting shear stress and normal stress. ,in This represents the shear stress of the soil at the final measurement benchmark point. To finally measure the cohesion of the soil at the benchmark point, It is the normal stress of the soil at the final measurement benchmark point; Similarly, the soil sample at the final measurement reference point is subjected to a direct shear test, and the cohesion is calculated based on the shear strength data of the soil. The strength data measured under different confining pressures and shear conditions are used to calculate the normal stress; The moisture content of the soil sample is measured by the humidity sensor, and the ratio of the water weight to the dry soil weight is the water content .

4. The neural network-based substation settlement prediction method in a collapsible loess area according to claim 3, characterized in that, A substation settlement characteristic index prediction model is established, and the model is optimized according to the following method: The obtained collapsibility coefficient, compression modulus, internal friction angle, cohesion, and water content are normalized and formed into a feature vector. A substation settlement prediction model is constructed based on a neural network structure, which includes an input layer, a hidden layer, and an output layer. The input layer is responsible for receiving the feature vector. The hidden layer is an intermediate layer between the input layer and the output layer in the neural network, responsible for feature extraction and information processing of the input feature vector. Nonlinear activation functions are used in the hidden layer to capture complex patterns. The output layer generates the final prediction results of the model, and the output layer has three nodes for outputting three substation settlement characteristic indexes. The particle swarm optimization algorithm is used to adjust the substation settlement characteristic indexes output by the prediction model. A particle swarm is initialized, and each particle represents a substation settlement characteristic index in a subregion. The position and velocity of the particle are set, where the position represents a combination of the weights and thresholds of the prediction model, and the velocity defines the moving direction and step of the particle in the search space. At the beginning of each iteration, a set of weights and thresholds are randomly generated as the initial position of the particle. The initial position of the particle is generated based on a normal distribution to ensure uniform distribution of the particle in the entire search space. The mean square error is used as the loss function to calculate the loss of the prediction model according to the following formula: in, This represents the calculated mean square error. This represents the number of sub-regions of the substation. Indicates the first The subregion containing the nth particle The actual values ​​of the settlement characteristic indicators of each substation Indicates the first The subregion containing the nth particle Predicted values ​​of substation settlement characteristic indicators The index is for the sub-region, and , It is an index of power station settlement characteristic indicators in the sub-region, and hour, This refers to the collapsibility index, which is a characteristic index of substation settlement. hour, Indicates the load-bearing capacity impact index, When, it represents the deformation stability index ; The velocity and position of the particle are updated according to the following formula: in, Indicates the first Individual particles The speed of time Indicates the first Individual particles Location at any given moment For inertial weights, , It is a learning factor. , It is a random number. Indicates the first The optimal position of each particle Indicates the globally optimal position. For the first Individual particles Location at any moment For the first Individual particles The speed of time; The process of evaluating the loss and updating the position and velocity of the particle is repeated until the maximum number of iterations is reached. The optimal particle position found is used to update the weights and thresholds of the neural network model. By training the model again, its performance in predicting substation settlement characteristic indexes is ensured to be optimal.

5. The neural network-based substation settlement prediction method in a collapsible loess area according to claim 4, characterized in that, A settlement fitness function is constructed to calculate the settlement fitness values of each subregion according to the following method: The settlement fitness function is constructed based on the settlement formula of the compression index, and the settlement characteristic indexes of each sub-region predicted by the substation settlement characteristic index prediction model are substituted into the formula: wherein, represents a settlement fitness value of a sub-area, is an initial void ratio of the compressed soil layer, is a thickness of the compressed soil layer, is a compression index, is an initial effective stress of the soil layer, is an additional stress of the soil layer.

6. The neural network-based substation settlement prediction method in a collapsible loess area according to claim 5, characterized in that, The specific settlement of each sub-region is obtained, and the settlement threshold is set. The sub-regions of the substation that exceed the settlement threshold are marked as risks. The method is: The substation settlement fitness values of each sub-region of the substation are obtained based on the settlement fitness function, combined with the load time series data of each sub-region, and divided according to the time period. The settlement of each sub-region of the substation in each time period is calculated. The formula is: wherein, denotes the settling amount in the th sub-area within the time interval , is the substation settling fitness value in the th sub-area, is the load received in the th sub-area within the time interval . The settlement time series data of each sub-area of the substation in the past year is summarized, and the mean value of the settlement amount of all sub-areas is calculated as the settlement threshold When , the sub-area is marked as a risk, and the settlement amount of the sub-area at this time is recorded.

Images