Photovoltaic slope safety evaluation method considering spatial-temporal variability of multi-source parameters

Abstract

The invention provides a photovoltaic slope safety evaluation method considering spatial-temporal variability of multi-source parameters, and the method comprises the steps: collecting geological exploration data and multi-source monitoring data of a photovoltaic slope, determining the random distribution type of initial multi-source parameters, and selecting key monitoring data; performing data matching between the multi-source parameters and the monitoring data, and screening the multi-source parameters and the monitoring data by using an MIC algorithm; the method comprises the following steps: constructing a three-dimensional simulation model considering photovoltaic slope multi-source parameter spatio-temporal variability based on field geological exploration data, and obtaining a data set; training a data set based on a 1D-CNN-MCMC deep learning model framework, and dynamically updating posterior probability distribution of a photovoltaic slope safety factor FS in combination with field monitoring data; and calculating a photovoltaic slope failure probability Pf, and carrying out safety evaluation and early warning. According to the method, uncertain factors such as slope geological parameter variability, environment and photovoltaic structure are comprehensively considered, and on-site multi-source monitoring data is utilized, so that the photovoltaic slope failure probability more conforming to the actual situation is obtained for safety evaluation and early warning.

Description

Technical Field

[0001] The present invention relates to the technical field of geotechnical engineering safety monitoring, and in particular to a photovoltaic slope safety evaluation method integrating the temporal and spatial variability of multi-source monitoring data. Background Art

[0002] Photovoltaic power generation, as a clean, renewable energy source, requires significant land. However, significant amounts of vacant land and space exist along highway slopes. Therefore, highway slopes could be a new application for photovoltaic power generation, reducing carbon emissions from transportation. Currently, the photovoltaic slope industry is nascent, and existing slope stability assessment methods do not incorporate the load effects of photovoltaic structures. Directly applying traditional slope safety methods would result in inaccurate and unreliable assessment results, posing significant safety risks.

[0003] The safety of photovoltaic slopes is influenced by numerous parameters, including geotechnical parameters, environmental parameters, and photovoltaic structures. Effectively selecting these parameters is a key issue in photovoltaic slope safety assessment. Furthermore, previous slope safety assessment methods typically only consider the spatial variability of geotechnical parameters, with limited consideration of the impact of time on slopes. These methods typically account for this effect through theoretical equations or numerical simulations, lacking practical guidance and making it difficult to ensure long-term prediction accuracy. Summary of the Invention

[0004] The present invention aims to solve the problem of selecting multi-source parameters and monitoring data for photovoltaic slopes, as well as how to consider the spatiotemporal variability of multi-source parameters for safety evaluation and early warning of photovoltaic slopes, so as to ensure the stable operation of photovoltaic slopes.

[0005] To achieve the above objectives, the present invention proposes a photovoltaic slope safety assessment method that considers the spatiotemporal variability of multi-source parameters, which is characterized by comprising the following steps:

[0006] Step 1: Collect geological survey data and multi-source monitoring data of the photovoltaic slope, determine the random distribution type of the initial multi-source parameters, and select key monitoring data;

[0007] Step 2: Perform data matching between multi-source parameters and monitoring data, and use the MIC algorithm to screen multi-source parameters and monitoring data;

[0008] Step 3: Based on the on-site geological survey data, a three-dimensional simulation model considering the spatiotemporal variability of multi-source parameters of the photovoltaic slope is constructed to obtain a data set;

[0009] Step 4: Based on the 1D-CNN-MCMC deep learning model framework, the dataset is trained and the posterior probability distribution of the photovoltaic slope safety factor FS is dynamically updated in combination with the field monitoring data;

[0010] Step 5: Calculate the probability of failure of the photovoltaic slope P f , conduct safety assessment and early warning.

[0011] Preferably, the step 1 further comprises the following steps:

[0012] Step 1.1: Collected PV slope survey data includes, but is not limited to, slope topography, slope geology, drilling results, and PV structure design drawings. Initial multi-source parameter information includes: mean and standard values of the geotechnical parameters (cohesion C, internal friction angle α, elastic modulus E, Poisson's ratio v); mean and standard deviation of rainfall RF; mean and standard deviation of wind speed AV; and mean and standard deviation of the PV structure's position P on the slope. Determine the random distribution type of the initial multi-source parameters, i.e., the prior distribution of the multi-source parameters, which can be expressed as the following formula.

[0013] θ prio =C,α,E,v,…,RF,AV,P

[0014] Where C is the cohesion, α is the internal friction angle, E is the elastic modulus, v is the Poisson's ratio, RF is the rainfall, AV is the wind speed, and P is the photovoltaic structure position parameter.

[0015] Step 1.2: Collect multi-source monitoring data for the photovoltaic slope, including but not limited to surface displacement (SD), deep displacement (DD), soil moisture (SM), photovoltaic column relative displacement (PD), photovoltaic column inclination (PA), and photovoltaic bracing and column strain (PS). This multi-source monitoring data is divided into key monitoring data and other monitoring data. Based on the slope failure characteristics, select key monitoring data that impacts photovoltaic slope safety for use in the subsequent MIC calculation in step 2.2.

[0016] Preferably, the step 2 further comprises the following steps:

[0017] Step 2.1: The multi-source parameter is a spatial variable G = {g1, g2, ..., g n} (random distribution determined by mean and standard deviation), while multi-source monitoring data is time series D t ={d1,d2,…,d n}(changes with time t). The matching strategy enables the two to be correlated, that is, at each time point t, a multi-source parameter value G is associated t , to match multi-source parameter variables with key monitoring data. To avoid contingency, the Monte Carlo method is used to randomly sample multi-source parameters N times.

[0018] Step 2.2: Calculate the MIC of the multi-source parameters sampled N times and the key monitoring data respectively. The average value of the MIC calculation is taken. The higher the average value of the MIC calculation, the stronger the correlation. Set a threshold to filter out the multi-source parameters with high correlation with the key monitoring data. The MIC calculation can be expressed as follows:

[0019]

[0020] In the formula, integers a and b represent the number of segments in the x and y directions, I(D|G) represents the mutual information (MI) based on the grids G and D, Ω(a,b) represents a set of two-dimensional grids of size a×b, L(n) is used to limit the size of the grid division area, and n is the length of the input data sequence. In this invention, L(n) takes n 0.6 .

[0021] Step 2.3: The multi-source monitoring data includes key monitoring data and other monitoring data. Based on the multi-source monitoring data of the photovoltaic slope collected in step 1.2, calculate the MIC value of the remaining monitoring data and the key monitoring data, and set a threshold to filter out the remaining key monitoring data required for the subsequent step 4.2.

[0022] Preferably, the step 3 further comprises the following steps:

[0023] Step 3.1: Based on the on-site geological survey data of the photovoltaic slope, a three-dimensional photovoltaic slope simulation model is established. M random sample combinations are uniformly drawn using optimal Latin hypercube sampling. The spatiotemporal variability of the multi-source parameters is set based on random field theory. The multi-source parameters are used as the model input of the three-dimensional simulation model and simulation calculations are performed to obtain the corresponding correlation response value g(θ). This can be expressed as follows:

[0024] g(θ)=g(C,α,E,v,…,RF,AV,P)

[0025] Step 3.2: Obtain the required correlation response values. Except for the safety factor FS, the obtained correlation response values must correspond to the existing monitoring data. Construct a dataset by combining the multi-source parameter inputs and the correlation response outputs.

[0026] Preferably, the step 4 further comprises the following steps:

[0027] Step 4.1: Build a multi-input and multi-output 1D-CNN deep learning framework model to predict the associated response value. Divide the dataset obtained in step 3 into a training set, a validation set, and a test set. Adjust the hyperparameters to obtain a proxy model with high prediction accuracy and strong generalization ability, which can be expressed as follows;

[0028] 1D-CNN={θ,g(θ)}

[0029] Step 4.2: Combine the remaining key monitoring data D screened by MIC in step 2.3 Mon Based on Bayesian theorem and Markov chain sampling algorithm, the Markov chain Monte Carlo method MCMC is used to obtain the posterior distribution of multi-source parameters. This process can dynamically update the posterior distribution with the real-time input of monitoring data, so as to obtain the posterior distribution θ of multi-source parameters considering the time effect. post , as shown below:

[0030]

[0031] Where l is the normalization factor, which ensures the consistency of the cumulative probability in the entire range of θ.

[0032] Step 4.3: The posterior distribution of the multi-source parameters obtained by the dynamic update in step 4.2 is predicted by the 1D-CNN proxy model to obtain the posterior probability distribution of the associated response value (safety factor FS) in real time and quickly, as shown in the following formula:

[0033] 1D-CNN={θ post ,g(θ post )}

[0034] Preferably, the step 5 further comprises the following steps:

[0035] Step 5.1: Based on the reliability theory, establish the performance function Z of the photovoltaic slope, as shown below:

[0036] Z=FS-1

[0037] Step 5.2: Based on the posterior distribution of the safety factor FS obtained in step 4.3 and the performance function Z of the photovoltaic slope established in step 5.1, the Monte Carlo method is used to simulate and calculate the real-time failure probability P of the slope. f , as shown below:

[0038]

[0039] Where, P f is the probability of PV-slope instability, P(FS<1) is the probability of instability determined by the safety factor; N (k) is the kth sample generated by sampling, I{·} is the indicative function, if FS<1, I{·}=1, otherwise I{·}=0; n FS is the total number of FS<1.

[0040] The real-time failure probability P of the slope calculated by the safety factor FS f Carry out safety evaluation and early warning classification, as shown in Table 1:

[0041]

[0042] The safety levels in Table 1 are divided into five levels: Level I, Level II, Level III, Level IV and Level V. The warning levels are also divided into five levels: green, blue, yellow, orange and red. Green represents safety, blue represents the warning state, and yellow, orange and red represent the alarm state.

[0043] Compared with the prior art, the present invention has the following beneficial effects:

[0044] 1. Comprehensive consideration of multi-source parameter selection and photovoltaic structure

[0045] Breaking through the limitation of traditional slope evaluation that ignores the load effect of photovoltaic structures, an evaluation method integrating geotechnical parameters, environmental factors and photovoltaic structure parameters is proposed.

[0046] A matching strategy between multi-source parameters and key monitoring data based on the MIC algorithm is proposed. The Monte Carlo random sampling method is used to eliminate the influence of chance and screen multi-source parameters and monitoring data.

[0047] 2. Innovative Application of Spatiotemporal Variability Simulation Models

[0048] Based on random field theory, the spatiotemporal variability of multi-source parameters is set, and the multi-source parameters are used as model inputs of a three-dimensional simulation model to obtain corresponding associated response values.

[0049] A multi-input and multi-output 1D-CNN-MCMC deep learning framework is constructed to predict the associated response values and training data sets, breaking through the static analysis limitations of traditional numerical simulations and realizing real-time updating of the posterior probability distribution of the safety factor FS.

[0050] 3. Data-driven dynamic evaluation of time effects

[0051] Based on the posterior distribution of multiple source parameters updated using Bayesian theorem and Markov Chain Monte Carlo (MCMC) methods, the posterior probability distribution of the associated response value (safety factor FS) can be quickly and in real time through surrogate model prediction. A dynamic update mechanism for the posterior distribution of the safety factor FS is proposed, allowing the evaluation results to evolve in real time with the monitoring data.

[0052] The real-time failure probability P of the slope calculated by the safety factor FS f Conduct safety evaluation and early warning classification, establish a five-level early warning system, and improve the engineering practicality of long-term predictions. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 This is a flow chart of a photovoltaic slope safety assessment method considering the spatiotemporal variability of multi-source parameters according to the present invention;

[0054] Figure 2A deep learning model framework for dynamically updating the posterior distribution of photovoltaic slope safety factor FS provided by an embodiment of the present invention DETAILED DESCRIPTION

[0055] In order to more clearly understand the above-mentioned objects, features and advantages of the present invention, a photovoltaic slope safety assessment method considering the spatiotemporal variability of multi-source parameters of the present invention is further described below with reference to the accompanying drawings and embodiments.

[0056] The photovoltaic slope safety assessment method considering the spatiotemporal variability of multi-source parameters comprises the following steps:

[0057] Step 1: Collect geological survey data and multi-source monitoring data of the photovoltaic slope, determine the random distribution type of the initial multi-source parameters, and select key monitoring data;

[0058] Specifically, step 1 includes the following steps:

[0059] The specific implementation of step 1.1 is to collect geological survey data of the photovoltaic slope site, including but not limited to slope topographic maps, slope geological conditions, and photovoltaic structure design drawings. The multi-source parameter information collected includes: the mean and standard values of the physical and mechanical parameters of the rock and soil (cohesion C, internal friction angle α, elastic modulus E, Poisson's ratio v); the mean and standard deviation of rainfall RF; the mean and standard deviation of wind speed AV; the mean and standard deviation of the position P of the photovoltaic structure on the slope. After determining the initial multi-source parameters, the random distribution type of these corresponding initial multi-source parameters is queried according to the corresponding specifications or references. Cohesion C and internal friction angle α are log-normal distributions, elastic modulus E, Poisson's ratio v, the position P of the photovoltaic structure on the slope and rainfall RF are normally distributed, and wind speed AV is extreme value type I distribution;

[0060] The specific implementation of step 1.2 involves collecting multi-source monitoring data for photovoltaic slopes, including but not limited to surface displacement SD, deep displacement DD, soil moisture content SM, relative displacement PD of photovoltaic structure columns, inclination PA of photovoltaic structure columns, and strain PS of photovoltaic structure braces and columns. Considering that the type of this embodiment is a photovoltaic slope on a highway, the project site has little rainfall all year round and there are few high and steep slopes along the route. Therefore, the landslide usually takes the form of a deep sliding mode. Based on this, deep displacement DD is selected as the key monitoring data for the MIC calculation in the subsequent step 2.2.

[0061] Step 2: Perform data matching between multi-source parameters and monitoring data, and use the MIC algorithm to screen multi-source parameters and monitoring data;

[0062] The MIC algorithm is used to perform correlation calculation. The MIC calculation is shown in the following formula:

[0063]

[0064] Where the integers a and b represent the number of segments in the x and y directions, I(D|G) represents the mutual information (MI) based on the grids G and D, Ω(a,b) represents a set of two-dimensional grids of size a×b, L(n) is used to limit the size of the grid division area, and n is the length of the input data sequence. Usually L(n) = n 0.6 In this embodiment, L(n) takes n 0.6 .

[0065] Specific implementation of step 2.1: The multi-source parameter is a spatial variable G = {g1, g2, ..., g n} (random distribution determined by mean and standard deviation), while the key monitoring data deep displacement DD is a time series D t ={d1,d2,…,d n}(changes with time t). The matching strategy enables the two to be correlated, that is, at each time point t, a multi-source parameter value G is associated t , to match the multi-source parameter variables with the key monitoring data deep displacement DD. To avoid contingency, the Monte Carlo method is used to randomly sample the multi-source parameters N times.

[0066] The specific implementation of step 2.2 is to calculate the MIC of the multi-source parameters sampled N times and the key monitoring data deep displacement DD, and take the average value of the MIC as the result, as shown in the following formula:

[0067]

[0068] Where, MIC j,i is the MIC value of the j-th multi-source parameter and the key data at the i-th sampling point, and N is the number of sampling times.

[0069] The higher the MIC value, the stronger the correlation. A threshold is set to screen out multi-source parameters with high correlation with the key monitoring data deep displacement DD. In this embodiment, the threshold is set to 0.3, and multi-source parameters below 0.3 will be discarded.

[0070] In the specific implementation of step 2.3, the multi-source monitoring data includes key monitoring data and remaining monitoring data. Based on the photovoltaic slope multi-source monitoring data collected in step 1.2, the MIC value of the remaining monitoring data and the key monitoring data deep displacement (DD) is calculated. A threshold is set to select the remaining key monitoring data with a high correlation with the key monitoring data deep displacement (DD) for use in the subsequent Markov Chain Monte Carlo (MCMC) update in step 4.2. In this embodiment, the MIC threshold is set to 0.3; remaining monitoring data with values below 0.3 are discarded.

[0071] Step 3: Based on the on-site geological survey data, a three-dimensional simulation model considering the spatiotemporal variability of multi-source parameters of the photovoltaic slope is constructed to obtain a data set;

[0072] Specifically, step 3 includes the following steps:

[0073] Step 3.1 is specifically implemented by establishing a three-dimensional photovoltaic slope simulation model based on on-site geological survey data. Based on random field theory, multi-source parameters are linked to space through correlation distances. A simple spatial random field distribution is used to set the spatiotemporal variability of the multi-source parameters. M random sample combinations of multi-source parameters are uniformly extracted using optimal Latin hypercube sampling. These M random sample combinations of multi-source parameters are used as model inputs for simulation calculations to obtain the corresponding implicit correlation response value g(θ), which can be expressed as follows:

[0074] g(θ)=g(C,α,E,v,…,RF,AV,P)

[0075] In step 3.2, numerical simulations are performed to obtain the required correlation response values, including slope surface displacement SD, deep displacement DD, soil moisture SM, PV structure column relative displacement PD, and PV structure column inclination PA. With the exception of the safety factor FS, the resulting correlation response values must correspond to the remaining monitoring data selected in step 2.3. The multi-source parameter inputs and correlation response outputs are constructed into a dataset.

[0076] Step 4: Based on the 1D-CNN-MCMC deep learning model framework, the dataset is trained and the posterior probability distribution of the photovoltaic slope safety factor FS is dynamically updated in combination with the field monitoring data;

[0077] Specifically, step 4 includes the following steps:

[0078] The step 4.1 is specifically implemented to build a multi-input and multi-output 1D-CNN deep learning framework model for predicting the associated response value, such as Figure 2 The constructed 1D-CNN consists of 5 convolutional layers, 1 residual layer, activation function layer, 1 expansion layer, 1 dropout layer and 2 fully connected layers. The loss function is the mean square error (MSE) and the evaluation index determination coefficient (R-Squared, R 2 ), setting an early stopping strategy of 300 epochs. Divide the dataset obtained in step 3 into a training set, a validation set, and a test set. The training set is used to train the model, the validation set is used to adjust the model hyperparameters, and the test set is used to test the model's accuracy. Finally, a proxy model with high prediction accuracy and strong generalization ability is obtained, which can be expressed as follows:

[0079] 1D-CNN={θ,g(θ)}

[0080] The step 4.2 is specifically implemented, combined with the remaining key monitoring data D screened by MIC in step 2.3 Mon Based on Bayesian theorem, the Markov algorithm is used to extract the Markov chain, and the Markov chain Monte Carlo method MCMC is used to update the posterior distribution θ of the multi-source parameters. post This process can dynamically update the posterior distribution as the remaining key monitoring data are input in real time, thereby obtaining the parameter posterior distribution that takes into account the time effect, as shown in the following formula:

[0081]

[0082] Where l is the normalization factor, which ensures the consistency of the cumulative probability in the entire range of θ.

[0083] In the specific implementation of step 4.3, the posterior distribution of the multi-source parameters obtained by the dynamic update of step 4.2 is predicted by the 1D-CNN proxy model, as shown in the following formula. The posterior distribution of the associated response value (safety factor FS) can be obtained quickly and in real time.

[0084] 1D-CNN={θ post ,g(θ post )}

[0085] Step 5: Calculate the probability of failure of the photovoltaic slope P f , conduct safety assessment and early warning.

[0086] Specifically, step 5 includes the following steps:

[0087] The specific implementation of step 5.1 is as follows: Based on the reliability theory, the function function Z of the photovoltaic slope is established, as shown in the following formula:

[0088] Z=FS-1

[0089] The specific implementation of step 5.2 is: based on the posterior distribution of the safety factor FS obtained in step 4.3 and the function function Z of the photovoltaic slope established in step 5.1, the Monte Carlo method is combined to simulate and calculate the real-time failure probability P of the slope f , as shown below:

[0090]

[0091] Where, P f is the probability of photovoltaic slope instability, P(FS<1) is the probability of instability determined by the safety factor, N (k) is the kth sample generated by sampling, I{·} is the indicative function, if FS<1, I{·}=1, otherwise I{·}=0, n FS is the total number of FS<1.

[0092] The real-time failure probability P of the slope calculated by the safety factor FS f Carry out safety evaluation and early warning classification, as shown in Table 1:

[0093]

[0094] The safety levels in Table 1 are divided into five levels: Level I, Level II, Level III, Level IV and Level V. The warning levels are also divided into five levels: green, blue, yellow, orange and red. Green represents safety, blue represents the warning state, and yellow, orange and red represent the alarm state.

[0095] Safety Assessment Level I indicates the highest level of safety, with the PV slope in a stable state and a failure probability between 0% and 5%, a low-probability event. The warning level is green, and no alarm is issued. Similarly, Safety Assessment Level V indicates the lowest level of safety, with the PV slope in an unstable state and a failure probability between 90% and 100%, a high-probability event. The warning level is red, and an alarm is issued.

[0096] By integrating multi-source monitoring data with a deep learning model, the present invention solves the problem of delayed early warning caused by the static assumption of geological parameters in traditional methods, and realizes dynamic probabilistic assessment of the safety status of the slope.

[0097] Although this specification has provided a detailed description of the present invention using general descriptions and specific embodiments, it will be apparent to those skilled in the art that modifications and improvements may be made based on the present invention. Therefore, such modifications and improvements, which do not depart from the spirit of the present invention, are intended to be within the scope of protection claimed herein.

Claims

1. A photovoltaic slope safety assessment method considering the spatiotemporal variability of multi-source parameters, characterized by: The following steps are involved: Step 1: Collect geological survey data and multi-source monitoring data of the photovoltaic slope, determine the random distribution type of the initial multi-source parameters, and select key monitoring data; Step 2: Perform data matching between multi-source parameters and monitoring data, and use the MIC algorithm to screen multi-source parameters and monitoring data; Step 3: Based on the on-site geological survey data, a three-dimensional simulation model considering the spatiotemporal variability of multi-source parameters of the photovoltaic slope is constructed to obtain a data set; Step 4: Based on the 1D-CNN-MCMC deep learning model framework, the dataset is trained and the posterior probability distribution of the photovoltaic slope safety factor FS is dynamically updated in combination with the field monitoring data; Step 5: Calculate the probability of failure of the photovoltaic slope P f , conduct safety assessment and early warning.

2. A photovoltaic slope safety assessment method considering the spatiotemporal variability of multi-source parameters according to claim 1, characterized in that: The step 1 also includes the following steps: Step 1.1: Collect geological survey data at the photovoltaic slope site, including but not limited to slope topographic maps, slope geological conditions, and photovoltaic structure design drawings. The collected initial multi-source parameters include the mean and standard values of the physical and mechanical parameters of the rock and soil (cohesion C, internal friction angle α, elastic modulus E, Poisson's ratio v); the mean and standard deviation of rainfall RF; the mean and standard deviation of wind speed AV; and the mean and standard deviation of the position P of the photovoltaic structure on the slope. Once the initial multi-source parameters are determined, the prior distribution of the multi-source parameters is obtained, which can be expressed as the following formula: θ prio =C,α,E,v,…,RF,AV,P Where C is the cohesion, α is the internal friction angle, E is the elastic modulus, v is the Poisson's ratio, RF is the rainfall, AV is the wind speed, and P is the photovoltaic structure location parameter; According to the corresponding specifications or references, the random distribution types of these corresponding parameters are found; Step 1.2: Collect multi-source monitoring data for the photovoltaic slope, including but not limited to surface displacement SD, deep displacement DD, soil moisture SM, relative displacement PD and inclination PA of photovoltaic structure columns, and strain PS of photovoltaic structure braces and columns. The multi-source monitoring data is divided into key monitoring data and other monitoring data. Based on the slope failure characteristics, select key monitoring data that affect the safety of the photovoltaic slope for use in the MIC calculation in the subsequent step 2.

3. A photovoltaic slope safety assessment method considering the spatiotemporal variability of multi-source parameters according to claim 1, characterized in that: The step 2 also includes the following steps: Step 2.1: The multi-source parameter is a spatial variable G = {g1, g2, ..., g n } (random distribution determined by mean and standard deviation), while multi-source monitoring data is time series D t ={d1,d2,…,d n }(changes with time t); the matching strategy is used to enable the two to perform correlation calculations, that is, at each time point t, a multi-source parameter value G is associated t , to match multi-source parameter variables with key monitoring data, in order to avoid contingency, the Monte Carlo method is used to randomly sample multi-source parameters N times; Step 2.2: Calculate the MIC of the multi-source parameters sampled N times and the key monitoring data respectively. Take the average value of the MIC calculation results. Set the threshold to filter out the multi-source parameters with high correlation with the key monitoring data. The MIC calculation can be expressed as follows: Where the integers a and b represent the number of segments in the x and y directions, I(D|G) represents the mutual information (MI) based on the grids G and D, Ω(a,b) represents a set of two-dimensional grids of size a×b, L(n) is used to limit the size of the grid division area, n is the length of the input data sequence, and L(n) = n 0.6 ; Step 2.3: The multi-source monitoring data includes: key monitoring data and other monitoring data; Calculate the MIC values of the remaining monitoring data and key monitoring data, set the threshold to filter out the remaining key monitoring data needed in the subsequent step 4.

4. A photovoltaic slope safety assessment method considering the spatiotemporal variability of multi-source parameters according to claim 1, characterized in that: The step 3 also includes the following steps: Step 3.1: Based on the on-site geological survey data of the photovoltaic slope, a three-dimensional photovoltaic slope simulation model is established. M random sample combinations are uniformly extracted using optimal Latin hypercube sampling. The spatiotemporal variability of multi-source parameters is set based on random field theory. The multi-source parameters are used as model inputs of the three-dimensional simulation model and simulation calculations are performed to obtain the corresponding correlation response value g(θ), which can be expressed as the following formula: g(θ)=g(C,α,E,ν,…,RF,AV,P) Step 3.2: Obtain the required correlation response values; except for the safety factor FS, the obtained correlation response values must correspond to the existing monitoring data; construct the multi-source parameter input and correlation response value output into a data set.

5. The photovoltaic slope safety assessment method considering the spatiotemporal variability of multi-source parameters according to claim 1 is characterized in that: The step 4 also includes the following steps: Step 4.1: Build a multi-input and multi-output 1D-CNN deep learning framework model to predict the associated response value. Divide the dataset obtained in step 3 into a training set, a validation set, and a test set. Adjust the hyperparameters to obtain a proxy model with high prediction accuracy and strong generalization ability, which can be expressed as follows; 1D-CNN={θ,g(θ)} Step 4.2: Combine the remaining key monitoring data D screened by MIC in step 2.3 Mon Based on Bayesian theorem and Markov chain sampling algorithm, the Markov chain Monte Carlo method MCMC is used to obtain the posterior distribution of multi-source parameters. This process can dynamically update the posterior distribution with the real-time input of monitoring data, so as to obtain the posterior distribution θ of multi-source parameters considering the time effect. post , as shown below: Where l is the normalization factor, which ensures the consistency of the cumulative probability in the entire range of θ; Step 4.3: The posterior distribution of the multi-source parameters obtained by the dynamic update in step 4.2 is predicted by the 1D-CNN proxy model, and the posterior probability distribution of the associated response value (safety factor FS) can be obtained quickly and in real time, as shown in the following formula: 1D-CNN = {θ post ,g(θ post )}.

6. A photovoltaic slope safety assessment method considering the spatiotemporal variability of multi-source parameters according to claim 1, characterized in that: The step 5 also includes the following steps: Step 5.1: Based on the reliability theory, establish the performance function Z of the photovoltaic slope, as shown below: Z=FS-1 Step 5.2: Based on the posterior distribution of the safety factor FS obtained in step 4.3 and the performance function Z of the photovoltaic slope established in step 5.1, the Monte Carlo method is used to simulate and calculate the real-time failure probability P of the slope. f , as shown below: Where, P f is the probability of PV-slope instability, P(FS<1) is the probability of instability determined by the safety factor; N (k) is the kth sample generated by sampling, I{·} is the indicative function, if FS<1, I{·}=1, otherwise I{·}=0; n FS is the total number of FS<1; The real-time failure probability P of the slope calculated by the safety factor FS f Safety evaluation and early warning classification are carried out. The safety evaluation level is divided into five levels: Level I, Level II, Level III, Level IV and Level V. The early warning level is divided into five levels: green, blue, yellow, orange and red, corresponding to different failure probability ranges.

7. The method according to claim 3, characterized in that The threshold of the MIC algorithm is set to 0.3, and multi-source parameters and monitoring data with correlation higher than 0.3 are screened.

8. The method according to claim 2, characterized in that On the photovoltaic slopes of highways, where rainfall is low throughout the year and there are few high and steep slopes along the line, deep displacement DD is selected as the key monitoring data.

9. The method according to claim 4, characterized in that In step 3.2, the required associated response values obtained by numerical simulation calculation include slope surface displacement SD, deep displacement DD, soil moisture content SM, photovoltaic structure column relative displacement PD and photovoltaic structure column inclination PA.

10. The method according to claim 1, characterized in that The constructed 1D-CNN consists of 5 convolutional layers, 1 residual layer, activation function layer, 1 expansion layer, 1 dropout layer and 2 fully connected layers. The loss function is the mean square error (MSE) and the evaluation index determination coefficient (R-Squared, R 2 ), set the early stopping strategy for 300 rounds.

Images