Overhead line load flow probability prediction method and system considering space-time correlation

Abstract

The application discloses an overhead line load flow probability prediction method and system considering space-time correlation, and relates to the technical field of overhead line load flow prediction. Historical measurement meteorological data are acquired, load flow and intermediate variables are calculated according to the historical measurement meteorological data, and the load flow and the intermediate variables are constructed as multiple groups of input data matrices; when performing preliminary load flow prediction, a quantile regression model can consider mining and extraction of space-time characteristic information in load flow data, a connection model structure is applied to space-time correlation modeling of preliminary load flow prediction results, and a multi-line-grade load flow joint prediction probability density function considering load flow space correlation and a multi-time-period load flow joint prediction probability density function considering load flow time correlation are constructed in two steps, so that space-time correlation information of the load flow is integrated into the prediction results, prediction optimization is realized, and the load carrying capacity of the overhead line in a future period can be mastered, which can guide the formulation of a dispatching scheme.

Description

Technical Field

[0001] This invention relates to the field of overhead line current carrying capacity prediction technology, and in particular to a method and system for predicting the probability of overhead line current carrying capacity based on spatiotemporal correlation. Background Technology

[0002] In real-world environments, meteorological information along overhead power lines varies, leading to differences in the current-carrying capacity (DCC) of different spans. The DCC is the minimum DCC among all spans. With the advent of critical span selection technology, determining the DCC no longer requires calculating the DCC of all spans along the line, but only the DCC at a few pre-selected critical spans. However, the location of the critical span corresponding to the minimum DCC varies randomly and is difficult to accurately predict. Therefore, to predict the DCC of the entire line, it is necessary to predict the DCC of multiple critical spans and then determine the overall DCC by comparison.

[0003] For predicting the current carrying capacity of multiple spans along overhead power lines, current research typically involves predicting the current carrying capacity of each span separately. However, in real-world environments, the current carrying capacity of the same span exhibits temporal correlation, and the current carrying capacity of different spans along the same overhead power line within the same time period also shows spatial correlation. This is easily understood from a natural perspective: atmospheric motion evolves gradually over time and propagates continuously within a spatial region, possessing both inertia and continuity in both time and space. Consequently, meteorological conditions at different spans along the overhead power line tend to be similar within the same time period, and the changes in meteorological conditions for the same span between adjacent time periods also exhibit a certain degree of continuity. The current carrying capacity of overhead conductors is primarily determined by their surrounding meteorological conditions, therefore, the current carrying capacity of different spans at different time periods also exhibits a certain degree of spatiotemporal correlation. This spatiotemporal correlation information is highly valuable, containing the inherent patterns of load fluctuations and reflecting the comprehensive load characteristics along the overhead line and the mutual influence between loads in different spans. Integrating this spatiotemporal correlation information into multi-span load prediction would help obtain richer prediction information, thereby improving prediction accuracy. However, current technologies do not consider the spatiotemporal correlation characteristics in the data during multi-span load prediction, thus reducing the accuracy of load probability prediction. Summary of the Invention

[0004] To address the shortcomings of existing technologies, the purpose of this invention is to provide a method and system for predicting the probabilistic carrying capacity of overhead lines by considering spatiotemporal correlation. This method accurately predicts the probabilistic carrying capacity of multiple spans along an overhead line by taking into account the spatiotemporal correlation between the carrying capacity of different spans.

[0005] To achieve the above objectives, the present invention is implemented through the following technical solution: The first aspect of this invention provides a method for predicting the probability of overhead line current carrying capacity based on spatiotemporal correlation, comprising the following steps: Acquire historical meteorological data and calculate the carrying capacity and intermediate variables based on the historical meteorological data; Preprocess current carrying capacity and intermediate variable data; Based on the different time periods and spatial correlation differences, the preprocessed carrying capacity and intermediate variable data are jointly constructed into multiple sets of input data matrices; A quantile regression model is constructed, and the quantile regression model is used to predict multiple sets of input data matrices to obtain the multi-line load capacity quantiles in the corresponding prediction period. The kernel density estimation method is used to estimate the quantiles of the carrying capacity of multiple key line sections, and the probability density function of the carrying capacity prediction for each prediction period in the prediction time domain is obtained. Construct a first connection model that can characterize the spatial correlation characteristics of the carrying capacity and a second connection model that can characterize the temporal correlation characteristics of the carrying capacity; The first connection model and the second connection model are used to perform temporal and spatial feature optimization on the traffic volume prediction probability density function in turn to obtain the final traffic volume prediction probability density function for each prediction period. The probability density function of the final prediction of the current carrying capacity in each prediction period is used to predict the probability of the current carrying capacity of the overhead line under test.

[0006] A second aspect of the present invention provides an overhead line current carrying capacity probability prediction system that considers spatiotemporal correlation, comprising: The data acquisition module is configured to acquire historical meteorological data and calculate the carrying capacity and intermediate variables based on the historical meteorological data. The preprocessing module is configured to preprocess carrier flow and intermediate variable data; The matrix construction module is configured to jointly construct multiple sets of input data matrices from the preprocessed carrying capacity and intermediate variable data according to different time periods and spatial correlation differences. The preliminary prediction module is configured to construct a quantile regression model, use the quantile regression model to predict multiple sets of input data matrices, and obtain the quantiles of the carrying capacity of multiple line sections in the corresponding prediction period; use the kernel density estimation method to estimate the quantiles of the carrying capacity of multiple line sections, and obtain the probability density function of the carrying capacity prediction of each key line section in the prediction time domain for each prediction period. The prediction optimization module is configured to construct a first connection model that can characterize the spatial correlation features of the carrying capacity and a second connection model that can characterize the temporal correlation features of the carrying capacity; and to perform temporal and spatial feature optimization on the carrying capacity prediction probability density function in turn using the first connection model and the second connection model to obtain the final prediction probability density function of the carrying capacity for each prediction period. The current carrying capacity probability prediction module is configured to predict the current carrying capacity probability of the overhead line under test by using the final prediction probability density function of the current carrying capacity in each prediction period.

[0007] The above one or more technical solutions have the following beneficial effects: This invention discloses a method and system for predicting the probability of overhead line current carrying capacity based on spatiotemporal correlation. Combining the characteristics of current carrying capacity data, the method utilizes a quantile regression model to extract spatiotemporal feature information from the current carrying capacity data during the initial prediction. Furthermore, a copula model structure is applied to model the spatiotemporal correlation of the initial prediction results. The method constructs a multi-span joint prediction probability density function considering the spatial correlation of current carrying capacity and a multi-time period joint prediction probability density function considering the temporal correlation of current carrying capacity in two steps. This integrates the spatiotemporal correlation information of current carrying capacity into the prediction results, enabling post-prediction optimization and thus understanding the current carrying capacity of overhead lines in future time periods. This guides the formulation of dispatching schemes and improves the safety and economic level of power grid operation.

[0008] Advantages of additional aspects of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0009] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.

[0010] Figure 1 This is a flowchart of the overhead line current carrying capacity probability prediction method that considers spatiotemporal correlation in Embodiment 1 of the present invention. Figure 2 This is a structural diagram of the improved CNN-LSTM network prediction model in Embodiment 1 of the present invention; Figure 3 This is a schematic diagram showing the prediction results of line 1 under the two models when comparing the prediction results of the improved CNN-LSTM model and the QRNN model in Embodiment 1 of the present invention. Figure 4 This is a schematic diagram showing the prediction results of line 2 under the two models when comparing the prediction results of the improved CNN-LSTM model and the QRNN model in Embodiment 1 of the present invention. Figure 5 This is a schematic diagram showing the prediction results of line 3 under the two models when comparing the prediction results of the improved CNN-LSTM model and the QRNN model in Embodiment 1 of the present invention. Figure 6 This is a schematic diagram showing the prediction results of line 1 under the three models when comparing the prediction results of the improved CNN-LSTM model, the LSTM model, and the traditional CNN-LSTM model in Embodiment 1 of the present invention. Figure 7 This is a schematic diagram showing the prediction results of line 2 under the three models when comparing the prediction results of the improved CNN-LSTM model, the LSTM model, and the traditional CNN-LSTM model in Embodiment 1 of the present invention. Figure 8 This is a schematic diagram showing the prediction results of line 3 under the three models when comparing the prediction results of the improved CNN-LSTM model, the LSTM model, and the traditional CNN-LSTM model in Embodiment 1 of the present invention. Figure 9 This is a schematic diagram of the prediction results of line 1 when comparing result 1 and result 2 in Embodiment 1 of the present invention; Figure 10 This is a schematic diagram of the prediction results of line 2 when comparing result 1 and result 2 in Embodiment 1 of the present invention; Figure 11 This is a schematic diagram of the prediction results of line segment 3 when comparing result 1 and result 2 in Embodiment 1 of the present invention; Figure 12 This is a schematic diagram of the prediction results of line 1 when comparing results 2 and results 3 in Embodiment 1 of the present invention; Figure 13 This is a schematic diagram of the prediction results of line 2 when comparing result 2 and result 3 in Embodiment 1 of the present invention; Figure 14 This is a schematic diagram of the prediction results of line segment 3 when comparing result 2 and result 3 in Embodiment 1 of the present invention. Detailed Implementation

[0011] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0012] It should be noted that the following detailed descriptions are exemplary and intended to provide further illustration of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0013] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of exemplary embodiments according to the invention. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof. Example 1: Embodiment 1 of this invention provides a method for predicting the probability of overhead line load capacity based on spatiotemporal correlation. Targeting the spatiotemporal characteristics inherent in the load capacity data, the method first focuses on the prediction model. It combines Convolutional Neural Networks (CNN), Long Short-Term Memory (LSTM), and quantile regression theory, and considers the characteristics of the load capacity data to design an improved CNN-LSTM quantile regression prediction model. This model performs multi-line load capacity prediction by first mining the spatial characteristics of multi-line input data and then capturing the temporal dependencies of multi-time period input data. Furthermore, to improve the prediction performance, this embodiment also improves the input of the load capacity prediction model based on cross-correlation analysis results.

[0014] Next, the Copula model structure was applied to model the spatiotemporal correlation of the preliminary traffic volume prediction results. The joint prediction probability density function of multi-line traffic volume considering the spatial correlation of traffic volume and the joint prediction probability density function of multi-time period traffic volume considering the temporal correlation of traffic volume were constructed to integrate the spatiotemporal correlation information of traffic volume into the prediction results to further improve the prediction performance.

[0015] Specifically, such as Figure 1 As shown, it includes the following steps: S101: Obtain historical meteorological data and calculate the carrying capacity and intermediate variables accordingly.

[0016] S102, preprocessing current carrying capacity and intermediate variable data.

[0017] S103, based on different time periods and spatial correlation differences, the preprocessed carrying capacity and intermediate variable data are jointly constructed into multiple sets of input data matrices.

[0018] S104. Adaptively improve and construct a CNN-LSTM quantile regression model. Use the quantile regression model to perform quantile prediction on the load capacity of each line segment in the future time period for multiple sets of input data matrices, and obtain the load capacity quantile of multiple line segments in the corresponding prediction time period.

[0019] S105, the kernel density estimation method is used to estimate the quantile of the current carrying capacity of the corresponding line section in the corresponding prediction period, and the preliminary prediction probability density function (A) of the current carrying capacity of each key line section in the prediction time domain for each prediction period is obtained.

[0020] S106, construct a first connection model that can characterize the spatial correlation characteristics of the current carrying capacity, denoted as Copula model (a), and construct a second connection model that can characterize the temporal correlation characteristics of the current carrying capacity, denoted as Copula model (b); use the first connection model and the second connection model to perform temporal and spatial feature optimization on the current carrying capacity prediction probability density function in sequence to obtain the final prediction probability density function of the current carrying capacity for each prediction period, and use the final prediction probability density function of the current carrying capacity for each prediction period to predict the probability of the current carrying capacity of the overhead line under test.

[0021] Specifically, based on the Copula model (a), spatial correlation modeling of the multi-carrier flow prediction probability density function (A) is performed to construct a joint probability density function (B) for multi-line-slot flow. Sampling is then performed based on the joint probability density function (B), and an improved flow prediction probability function (C) incorporating spatial correlation is obtained from the sampling results. Based on the Copula model (b), temporal correlation modeling of the improved flow prediction probability function (C) is performed to construct a joint probability density function (D) for multi-time period flow. Sampling is then performed based on the joint probability density function (D), and a final flow prediction probability density function (E) further incorporating temporal correlation is obtained from the sampling results. Thus, by applying the Copula model to model the spatiotemporal correlation of flow in the prediction results, both spatial and temporal correlation information of flow are incorporated into the prediction results.

[0022] In S101, the historical meteorological data is the DTR (Dynamic Thermal Rating) historical data, which includes historical meteorological data around multiple key spans along the overhead line, as well as intermediate variable data and current carrying capacity data calculated based on this data.

[0023] The intermediate variable is the heat absorbed by light, q. s Convection heat dissipation q c and heat dissipation q r Their respective calculation formulas are as follows: (1) (2) (3) (4) (5) (6) in, s Q represents the light radiation absorptivity of the line. se The solar radiation flux received by the line (W / m) 2 ) A' is the effective angle of incidence of sunlight (°); A' is the projected area per unit length of the line (m²). 2 / m); D0 is the outer diameter of the overhead line (m); f Represents the air density around the line (kg / m³) 3 ); f V represents the dynamic viscosity of the surrounding air (kg / ms); w Represents the wind speed (m / s) around the line; k f k is the thermal conductivity of the surrounding air (W / (m-℃)); angle Wind direction factor; The angle (°) between the wind direction and the line axis is called the wind direction angle; T s T represents the surface temperature of the circuit (°C). a q represents ambient temperature (°C). c1 q c2 q cn These refer to the forced convection heat dissipation generated by the line at low wind speed, high wind speed, and zero wind speed, respectively. In practical applications, these values ​​should be calculated separately and the maximum value among the three should be taken as q. c ; R(T) is the thermal radiation coefficient of the line; avg The average temperature of the line reaches T. avg AC resistance per unit length (Ω / m) at that time; [T low T high The [] represents the effective temperature range in which the line temperature and its AC resistance approximately have a linear relationship, which is optional in this invention. T low and T high Given temperatures set to 25°C and 75°C respectively, calculate the maximum allowable temperature. T max When the conductor resistance is below the operating value, take T avg = T max .

[0024] The formula for calculating current carrying capacity is as follows: (7) The carrying capacity is further calculated by using intermediate variables obtained from historical meteorological data.

[0025] In this embodiment, the historical meteorological data around the conductor includes historical solar radiation intensity, temperature, wind speed, and wind direction around the conductor.

[0026] It should be noted that in other embodiments, historical meteorological data around the conductor may also include other meteorological data.

[0027] In S102, considering the differences in the value range and units of intermediate variable data and carrying capacity data, it is necessary to normalize the above data before forecasting in order to perform forecasting reasonably. The data normalization preprocessing expression is as follows: (8) Where x represents the intermediate variable data or carrying capacity data to be processed, and x' represents the standardized data obtained after preprocessing. min and x max These represent the minimum and maximum values ​​of the corresponding physical quantities in the sample data, respectively.

[0028] In S103, the intermediate variable data and carrying capacity data of multiple historical time periods involved in the prediction input data are jointly constructed into multiple sets of input data matrices according to different time periods and spatial correlation differences. Specifically, the prediction input data of multiple key line sections are first jointly constructed into multiple sets of data matrices according to different historical time periods. Based on this, the prediction input data of multiple line sections in the same period are further divided into two sets of data matrices according to different degrees of spatial correlation. One set contains carrying capacity and q... c The other group is q s and q r This allows for the construction of multiple sets of input data matrices.

[0029] In S104, the constructed multiple sets of input data matrices are fed into the adaptively improved CNN-LSTM quantile regression model to predict the multi-line load capacity quantiles for the corresponding prediction period. Specifically, a CNN is first used to mine the spatial correlation features in the multi-line load capacity data of the same period, and then an LSTM network is used to learn the temporal features contained in the load capacity data of multiple consecutive time periods. Based on this, multi-line load capacity prediction is performed. At the same time, the adaptive improvements made to the CNN-LSTM network help to coordinate the spatial correlation differences between load capacity and intermediate variables to ensure that the network extracts accurate and effective spatiotemporal feature information, thereby improving the model's prediction results in terms of accuracy, reliability, and sharpness.

[0030] In one specific implementation, considering the potential differences in spatial correlation among the various variables involved in the prediction input, directly inputting intermediate variables and traffic flow data from multiple lines simultaneously into the same CNN might result in convolutional kernels with identical parameters failing to comprehensively and effectively mine the spatial feature information of each factor. To address this issue, this embodiment adaptively improves the CNN-LSTM network based on the characteristics of traffic flow data, ensuring that the network extracts accurate and effective spatiotemporal feature information, thereby improving the model's prediction accuracy.

[0031] Table 1 shows the spatial cross-correlation coefficients of the multi-line load capacity and intermediate variables involved in the prediction input data. From this, we can see that the load capacity (DTR) and q... c (Convection heat dissipation) The spatial correlation between the two is roughly the same, q s (Solar heat absorption) and q r The spatial correlation between the two (heat radiation heat dissipation) is roughly the same, while the spatial correlation between the former two and the latter two differs significantly.

[0032] Table 1. Spatial cross-correlation coefficients between current carrying capacity and intermediate variables

[0033] Based on the spatial correlation coefficient calculation results shown in Table 1, this embodiment makes adaptive improvements to the structure of the traditional CNN-LSTM network to coordinate the spatial correlation differences between the carrying capacity and intermediate variables, thereby efficiently and accurately extracting spatial feature information from multi-line input data. The implementation process is as follows. Figure 2 As shown.

[0034] Specifically, let the prediction start time period be t. First, the prediction input data of multiple key line sections are jointly constructed into multiple data matrices according to their different historical time periods, and these matrices are respectively named C1, C2, ..., C6. n Based on this, the predicted input data for multiple lines during the same period are further divided into two data matrices according to their different degrees of spatial correlation. One set contains the carrying capacity and q. c The other group is q s and q rThe data are then fed into different CNNs for spatial feature extraction of corresponding factors. After convolution and pooling operations, the high-dimensional feature information output by the two CNNs at the same time is combined and flattened to obtain a one-dimensional spatial feature vector for the corresponding time period. With the adaptive improvements in this embodiment, carrier flows and intermediate variables with different degrees of spatial correlation are extracted using different CNNs, thus coordinating the differences in spatial correlation between different factors and ensuring that the spatial feature information of carrier flows and intermediate variables can be effectively and accurately extracted. Subsequently, the n sets of spatial feature vectors from these n time periods are input into an LSTM network in chronological order to further learn the temporal feature information contained in the input data of multiple consecutive time periods. Finally, the carrier flow prediction result is output through a fully connected layer.

[0035] Furthermore, this invention combines the CNN-LSTM network model with quantile regression theory. Referring to the construction methods of existing neural network quantile regression models (QRNN), this embodiment establishes a loss function for the CNN-LSTM quantile regression model, and obtains the corresponding model parameters by minimizing the quantile regression loss function. The specific expression of the constructed loss function is as follows: (9) Where z represents the number of current carrying capacity samples. Yh r ( h =1,…, H , r =1… z () represents the r-th current carrying capacity sample value of line slot h, where H is the number of line slots. yoh r ( τ The ) represents the predicted quantile result of the current carrying capacity τ of line segment h. η τ (·) is a modified test function constructed to ensure that the loss function is differentiable at the origin, and its expression is shown below: (10) in, This represents the given threshold parameter. This indicates the parameter to be tested.

[0036] To verify the predictive performance of the proposed model, simulations were conducted using an improved CNN-LSTM quantile regression model, a QRNN model, an LSTM quantile regression model, and a traditional CNN-LSTM quantile regression model. The dataset used for the simulations consisted of meteorological data collected by DTR measuring devices at three key spans of a 220kV overhead line from November 1, 2021 to December 31, 2021, along with calculated current carrying capacity and intermediate variable data. The data from the first 58 days was used as the training set to train the model parameters, while the remaining 3 days were used as the test set to evaluate and analyze the model's prediction results. Furthermore, during the simulations, the time period was set to 15 minutes, the prediction time domain to 2 hours, and 36 consecutive predictions were made within this time window to predict the current carrying capacity for the three days in the test set. Finally, relevant evaluation indicators were calculated to assess the merits of the different prediction results.

[0037] Table 2. Calculation results of evaluation indicators for four prediction outcomes

[0038] Figure 3 , Figure 4 , Figure 5 The comparison shows the prediction results of the improved CNN-LSTM model and the QRNN model for line segments 1, 2, and 3, respectively. Observation Figure 3 , Figure 4 , Figure 5 As can be seen from Table 2, compared with the QRNN model, the prediction results obtained based on the improved CNN-LSTM quantile prediction model have improved in terms of accuracy, reliability and sharpness. This shows that taking into account the mining and utilization of spatiotemporal related feature information in the traffic flow data when performing traffic flow prediction is of great help to improve the traffic flow prediction results.

[0039] at the same time, Figure 6 , Figure 7 , Figure 8 This section compares the prediction results of the improved CNN-LSTM model and the LSTM model for line segments 1, 2, and 3, respectively, with a comparison of the results from the traditional CNN-LSTM model. (Combined with...) Figure 6 , Figure 7 , Figure 8The evaluation metrics calculated in Table 2 also reveal that the prediction results based on the traditional CNN-LSTM network are not significantly different from those based on the LSTM network. This indicates that the prediction performance of the traditional CNN-LSTM network has not been significantly improved compared to the LSTM network. This is mainly because the traditional CNN-LSTM network does not consider the spatial correlation differences between carrying capacity and intermediate variables, resulting in the ineffective extraction of spatial feature information. However, the evaluation metrics of the prediction results obtained based on the adaptively improved CNN-LSTM network are significantly improved. This demonstrates that the adaptive improvements made to the CNN-LSTM network in this chapter help to coordinate the spatial correlation differences between carrying capacity and intermediate variables, ensuring that the model extracts accurate and effective spatiotemporal feature information, thereby improving the model's prediction results in terms of accuracy, reliability, and sharpness.

[0040] In addition, from Figure 3 , Figure 4 , Figure 5 , Figure 6 , Figure 7 and Figure 8 As can be seen, the predicted low quantile of the current carrying capacity of each line segment obtained by the improved CNN-LSTM quantile prediction model is higher, which in turn makes the predicted low quantile of the current carrying capacity of the entire line higher. This means that if the predicted low quantile of the current carrying capacity of the line is extracted by mining the spatiotemporal feature information of the current carrying capacity data to guide the operators to make power system dispatching decisions to avoid risks, it will also help to achieve a more complete exploitation of the current carrying capacity of the line.

[0041] In S105, the kernel density estimation method is used to estimate the quantiles of the carrying capacity of each key span in the corresponding forecast period, thus obtaining the probability density function of the carrying capacity for each forecast period in the forecast time domain. The kernel density estimation method is a non-parametric statistical method that estimates the probability density function of unknown data by applying a weighted average of known data using a kernel function. In multi-span carrying capacity forecasting, we can use historical data as known data and estimate the probability density function of the carrying capacity for each future period using the kernel density estimation method.

[0042] First, we collect and organize historical data, including the carrying capacity data of each key section at different time periods. Then, we select a suitable kernel function, such as the Gaussian kernel function, as the basis for estimation. Next, based on the distribution of the known data, we use the kernel function to perform a weighted average of the carrying capacity data for each time period, obtaining the carrying capacity probability density function for that period.

[0043] In S106, this embodiment performs the second-stage prediction optimization based on the Copula model structure. After the first-stage prediction, which uses an improved CNN-LSTM quantile regression model to deeply mine the spatiotemporal correlation features of the load capacity data, a suitable third-order Copula model is selected as the first connection model, and a suitable eighth-order Copula model is selected as the second connection model. The first and second connection models are used to model the spatial and temporal correlation of the load capacity in the prediction results, thereby integrating the spatiotemporal correlation information of the load capacity into the prediction results and further improving the prediction performance. Thus, this invention not only mines and extracts the spatiotemporal correlation features contained in the overhead line load capacity data during the first-stage preliminary load capacity prediction, but also further integrates the spatiotemporal correlation information of the load capacity into the prediction results during the second-stage prediction optimization by modeling the spatiotemporal correlation of the preliminary prediction results, thereby achieving a more comprehensive consideration and utilization of the spatiotemporal related information of the load capacity.

[0044] Considering the existence of various Copula models, in order to select the optimal model that can characterize the spatial and temporal correlation of carrying capacity, this embodiment first calculates the BIC values ​​of five commonly used Copula models based on historical carrying capacity data. The calculation results are shown in Tables 3 and 4.

[0045] Table 3. BIC values ​​of the Copula model for spatial correlation modeling

[0046] Table 4. BIC values ​​of the Copula model for time correlation modeling

[0047] Based on the BIC calculation results in Tables 3 and 4 above, it can be seen that the t-Copula model has the smallest BIC value, meaning that this model is the optimal model for modeling the spatial and temporal correlation of traffic volume. Therefore, this embodiment uses the t-Copula model to construct the joint prediction probability density function for multi-line-slot traffic volume and the joint prediction probability density function for multi-time period traffic volume.

[0048] In one specific implementation, the probability density function of the carrying capacity predicted based on the improved CNN-LSTM quantile regression model is used as the marginal probability density function, and a joint prediction probability density function of the carrying capacity of multiple line spans is constructed for each prediction period. The number of critical line spans for the overhead line studied in this invention is 3, and the number of time periods in the prediction time domain is 8. The carrying capacity of line span h in prediction time period k is represented by a random variable I. hk This indicates that there are a total of 24 random variables. The ternary t-Copula density function is used for modeling the spatial correlation of carrying capacity. The expression is as follows: (11) in, λ 1 T =[ t -1 ( F 1 ( I 1k )), t -1 ( F 1 ( I 2k )), t -1 ( F 1 ( I 3k ))],in t -1 (·) is the inverse function of the univariate t-distribution function. F 1 ( I hk ()( h =1,2,3) represents the probability distribution function of the carrying capacity of line h in the prediction period k, obtained by the improved CNN-LSTM quantile regression model in the first stage, and Γ(·) represents the gamma function. ρ 1 and b 1 is the third-order correlation matrix and degrees of freedom obtained through maximum likelihood estimation, in this embodiment. b The estimated value of 1 is 6. ρ The estimated value of 1 is as follows: (12) The probability density function for joint prediction of multi-line span carrying capacity considering spatial correlation is constructed for the prediction period k. f ( I 1k , I 2k , I 3k ()( k The expression for (=1,…,8) is as follows: (13) Next, 2000 random samples are performed on equation (13), and kernel density estimation is performed based on the sampling results to obtain the optimized probability density function for predicting the carrying capacity after incorporating spatial correlation information. f 2 ( I hk ). Figure 9 , Figure 10 and Figure 11 The optimized traffic flow probability prediction results for three line segments are shown below, after further incorporating spatial correlation information of traffic flow based on the improved CNN-LSTM network prediction results (Result 2). For comparison, the first-stage traffic flow prediction results based on the improved CNN-LSTM network (Result 1) are also shown below. Figure 9 , Figure 10 and Figure 11 Table 5 shows the calculation of the evaluation indicators for the two results.

[0049] observe Figure 9 , Figure 10 and Figure 11 It can be observed that incorporating spatial correlation information of multi-line load capacity into the preliminary load capacity prediction results using the Copula model can optimize and improve the prediction results. With the prediction interval including the true values, the predicted distribution after incorporating load capacity spatial correlation information is more concentrated. Furthermore, the evaluation index calculation results shown in Table 5 also indicate that the load capacity prediction results obtained after incorporating load capacity spatial correlation information for prediction optimization have higher prediction accuracy. Moreover, compared to result 1 without the second-stage prediction optimization, incorporating load capacity spatial correlation information also improves the reliability and sharpness of the load capacity probability prediction results.

[0050] Table 5. Calculation results of evaluation indicators for Results 1 and Results 2

[0051] The optimized probability density function for predicting carrying capacity incorporates spatial correlation information of carrying capacity. f 2 ( I hk ()( k =1…8; h After setting the value as 1…3), this is used as the marginal probability density function. The t-Copula model is then used to construct a joint prediction probability density function for multi-period carrying capacity in the prediction time domain for each line section, further incorporating the temporal correlation information of carrying capacity into the prediction results. The constructed multi-period joint prediction probability density function for carrying capacity… as follows: (14) in, λ 2 T =[ t -1 ( F 2 ( I h1 )), … ,t -1 ( F 2 ( I h8 ))], F 2 ( I hk ()( k =1,…,8) represents the probability distribution function of the carrying capacity of line segment h during the prediction period k, obtained after incorporating spatial correlation information of carrying capacity into the prediction results for optimization. ρ 2 and b 2. The 8th-order correlation matrix and degrees of freedom are obtained through maximum likelihood estimation.

[0052] By performing 2000 joint random samplings on the joint probability density function and estimating the kernel density based on the sampling results, the final traffic volume prediction probability density function, which further incorporates multi-period traffic volume temporal correlation information, can be obtained. f 3 ( I hk ()( k =1…8; h =1…3). Figure 12 , Figure 13 and Figure 14 The final probability prediction results for the carrying capacity are shown in Result 3, after incorporating spatial correlation information and temporal correlation information of the carrying capacity for prediction optimization in three different line profiles. For comparison, Result 2, which only incorporates spatial correlation information of the carrying capacity, is also shown. Figure 12 , Figure 13 and Figure 14 Table 6 also shows the calculation of relevant evaluation indicators for the two prediction results.

[0053] Table 6. Calculation results of evaluation indicators for Results 2 and 3

[0054] observe Figure 12 , Figure 13 and Figure 14 It can be observed that further modeling the temporal correlation of the prediction results to incorporate the temporal correlation information of the carrying capacity still helps improve prediction performance. Compared to result 2, which only incorporates the spatial correlation information of the carrying capacity, result 3, which incorporates the temporal and spatial correlation information of the carrying capacity, shows a more concentrated prediction distribution. Furthermore, the evaluation index calculation results shown in Table 6 also reveal that further incorporating the temporal correlation information of the carrying capacity for prediction optimization will further improve the prediction accuracy, while the reliability and sharpness of the prediction results are also largely improved accordingly.

[0055] Example 2: Embodiment 2 of the present invention provides an overhead line current carrying capacity probability prediction system that considers spatiotemporal correlation, comprising: The data acquisition module is configured to acquire historical meteorological data and calculate the carrying capacity and intermediate variables based on the historical meteorological data. The preprocessing module is configured to preprocess carrier flow and intermediate variable data; The matrix construction module is configured to jointly construct multiple sets of input data matrices from the preprocessed carrying capacity and intermediate variable data according to different time periods and spatial correlation differences. The preliminary prediction module is configured to construct a quantile regression model, use the quantile regression model to predict multiple sets of input data matrices, and obtain the quantiles of the carrying capacity of multiple line sections in the corresponding prediction period; use the kernel density estimation method to estimate the quantiles of the carrying capacity of multiple line sections, and obtain the probability density function of the carrying capacity prediction of each key line section in the prediction time domain for each prediction period. The prediction optimization module is configured to construct a first connection model that can characterize the spatial correlation features of the carrying capacity and a second connection model that can characterize the temporal correlation features of the carrying capacity; and to perform temporal and spatial feature optimization on the carrying capacity prediction probability density function in turn using the first connection model and the second connection model to obtain the final prediction probability density function of the carrying capacity for each prediction period. The current carrying capacity probability prediction module is configured to predict the current carrying capacity probability of the overhead line under test by using the final prediction probability density function of the current carrying capacity in each prediction period.

[0056] The steps and methods involved in the above embodiment two correspond to those in embodiment one. For specific implementation details, please refer to the relevant description section of embodiment one.

[0057] Those skilled in the art will understand that the modules or steps of the present invention described above can be implemented using general-purpose computer devices. Optionally, they can be implemented using computer-executable program code, thereby allowing them to be stored in a storage device for execution by a computer device, or they can be fabricated as separate integrated circuit modules, or multiple modules or steps can be fabricated as a single integrated circuit module. The present invention is not limited to any particular combination of hardware and software.

[0058] While the specific embodiments of the present invention have been described above in conjunction with the accompanying drawings, this is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art without creative effort based on the technical solutions of the present invention are still within the scope of protection of the present invention.

Claims

1. A method for predicting the probability of overhead line current carrying capacity based on spatiotemporal correlation, characterized in that, Includes the following steps: Acquire historical meteorological data and calculate the carrying capacity and intermediate variables based on the historical meteorological data; Preprocess current carrying capacity and intermediate variable data; Based on the different time periods and spatial correlation differences, the preprocessed carrying capacity and intermediate variable data are jointly constructed into multiple sets of input data matrices; A quantile regression model is constructed and used to predict multiple sets of input data matrices to obtain the quantiles of multi-line load capacity in the corresponding prediction period. The quantile regression model is composed of an improved CNN-LSTM network. First, the CNN network is used to mine the spatial correlation features in the multi-line load capacity data of the same period through multiple sets of input data matrices. Then, the LSTM network is used to learn the temporal features contained in the load capacity data of multiple consecutive time periods, thereby completing the prediction of multi-line load capacity. The kernel density estimation method is used to estimate the quantiles of the carrying capacity of multiple key line sections, and the probability density function of the carrying capacity prediction for each prediction period in the prediction time domain is obtained. Construct a first connection model that can characterize the spatial correlation characteristics of the carrying capacity and a second connection model that can characterize the temporal correlation characteristics of the carrying capacity; The first and second connection models are used to perform temporal and spatial feature optimization on the traffic volume prediction probability density function in turn to obtain the final traffic volume prediction probability density function for each prediction period. The third-order Copula model is selected as the first connection model and the eighth-order Copula model is selected as the second connection model. The probability density function of the final prediction of the current carrying capacity in each prediction period is used to predict the probability of the current carrying capacity of the overhead line under test.

2. The overhead line current carrying capacity probability prediction method based on spatiotemporal correlation as described in claim 1, characterized in that, Historical meteorological data includes historical meteorological data around several key spans along the overhead line.

3. The overhead line current carrying capacity probability prediction method based on spatiotemporal correlation as described in claim 2, characterized in that, Historical meteorological data in the surrounding area includes historical solar radiation intensity, temperature, wind speed, and wind direction around the conductor.

4. The overhead line current carrying capacity probability prediction method based on spatiotemporal correlation as described in claim 1, characterized in that, Preprocessing the carrying capacity and intermediate variable data involves normalizing the carrying capacity and intermediate variable data.

5. The overhead line current carrying capacity probability prediction method based on spatiotemporal correlation as described in claim 1, characterized in that, The loss function of the quantile regression model is established based on the quantile regression theory, and the model parameters of the quantile regression model are obtained by minimizing the quantile regression loss function.

6. The overhead line current carrying capacity probability prediction method based on spatiotemporal correlation as described in claim 1, characterized in that, The specific steps for performing temporal and spatial feature optimization on the traffic volume prediction probability density function using the first and second connection models to obtain the final traffic volume prediction probability density function for each prediction period are as follows: Based on the first connection model, the spatial correlation model of the multi-carrier flow prediction probability density function is used to construct the joint probability density function of multi-line-slot flow. Sampling is performed based on the joint probability density function of multi-line load capacity, and the improved prediction probability function of load capacity after incorporating spatial correlation of load capacity is obtained based on the sampling results. Based on the second connection model, the time correlation model of the improved prediction probability function of carrying capacity is used to construct the joint probability density function of carrying capacity in multiple time periods; Sampling is performed based on the joint probability density function of the carrying capacity over multiple time periods, and the final predicted probability density function of the carrying capacity after further incorporating the temporal correlation of the carrying capacity is obtained based on the sampling results.

7. The overhead line current carrying capacity probability prediction method based on spatiotemporal correlation as described in claim 1, characterized in that, The Copula model is a t-Copula model.

8. A probabilistic prediction system for the overhead line current carrying capacity probability prediction method based on spatiotemporal correlation as described in any one of claims 1-7, characterized in that, include: The data acquisition module is configured to acquire historical meteorological data and calculate the carrying capacity and intermediate variables based on the historical meteorological data. The preprocessing module is configured to preprocess carrier flow and intermediate variable data; The matrix construction module is configured to jointly construct multiple sets of input data matrices from the preprocessed carrying capacity and intermediate variable data according to different time periods and spatial correlation differences. The preliminary prediction module is configured to construct a quantile regression model, use the quantile regression model to predict multiple sets of input data matrices, and obtain the quantiles of the carrying capacity of multiple line sections in the corresponding prediction period; use the kernel density estimation method to estimate the quantiles of the carrying capacity of multiple line sections, and obtain the probability density function of the carrying capacity prediction of each key line section in the prediction time domain for each prediction period. The prediction optimization module is configured to construct a first connection model that can characterize the spatial correlation features of the carrying capacity and a second connection model that can characterize the temporal correlation features of the carrying capacity; and to perform temporal and spatial feature optimization on the carrying capacity prediction probability density function in turn using the first connection model and the second connection model to obtain the final prediction probability density function of the carrying capacity for each prediction period. The current carrying capacity probability prediction module is configured to predict the current carrying capacity probability of the overhead line under test by using the final prediction probability density function of the current carrying capacity in each prediction period.

Images