Distribution network construction cost terrain classification calculation method and application

Abstract

The invention relates to the technical field of unmanned aerial vehicle application, in particular to a distribution network construction cost terrain classification calculation method and application, and is applied to cost auxiliary calculation of various terrains such as mountain land, plateau, plain, river valley, desert and gobi. The method is especially suitable for Tibetan area construction cost. Through the unmanned aerial vehicle and AI identification technology, intelligent auxiliary calculation of the distribution network project cost by the unmanned aerial vehicle is realized, and the problem of field inspection is solved. A big data standard image library is established, intelligent image identification field application is deepened, comparison with background images is carried out, a field equipment material list is automatically identified and counted, and the cost accuracy and efficiency are improved. And analyzing engineering use equipment, and performing comparative analysis in combination with different terrains to realize accurate cost of intelligent power distribution network equipment.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of unmanned aerial vehicle application, in particular to a construction cost terrain classification calculation method and application, which is applied to the auxiliary calculation of construction cost of various terrains such as mountains, plateaus, plains, valleys, deserts and gobi, and is particularly suitable for use in construction cost in the Tibetan area. BACKGROUND

[0002] Gansu Province has various terrains such as mountains, plateaus, plains, valleys, deserts and gobi. Due to the terrain factors, the company's distribution network engineering in Gansu faces many difficulties and challenges in construction, material transportation and terrain coefficient adjustment.

[0003] In addition, the terrain in the Tibetan area is more diverse and complex, and there are usually various terrains mixed in the Tibetan area. The existing calculation method is to integrate the blocks after the statistics of single terrains, and there is no way to achieve unified deployment. Therefore, the calculation of the construction cost of various terrains, especially the construction cost in the Tibetan area, is not smooth, and the standards are not unified. SUMMARY

[0005] In the current process of Gansu distribution network engineering cost, the short board of traditional cost mode is increasingly prominent. The cost personnel often need to carry out high-intensity work in complex outdoor environment, which not only consumes a lot of human and material resources, but also has low overall work efficiency. For example, in the complex terrain areas such as mountains, the cost personnel may need to walk several kilometers to reach the engineering site, which consumes a lot of time and energy, and manual recording and accounting data are prone to errors. With the continuous expansion of engineering construction scale and the continuous acceleration of construction speed, the traditional mode has been difficult to meet the actual demand. The mode has remarkable effect, on the one hand, it greatly improves the on-site verification efficiency of engineering line tower equipment, and reduces the outdoor work intensity of cost personnel; on the other hand, it realizes the automatic statistics of distribution network engineering materials, and the automatic comparison and analysis of data and doubts.

[0006] From the technical application level, the unmanned aerial vehicle can quickly reach the areas that are difficult to reach by traditional means due to its mobility advantage, and obtain high-quality image data; the monocular vision ranging technology can accurately measure the key data such as line span, which provides accurate data support for cost; the AI image intelligent recognition technology can quickly analyze the collected images and accurately identify various distribution network materials.

[0007] In summary, the comprehensive application of this series of frontier technologies has injected strong impetus into the digital transformation and upgrading of engineering field cost, greatly improving the accuracy and efficiency of cost work, and opening up a new path for distribution network engineering cost work, and improving the cost efficiency of the company.

[0008] In the face of the above difficulties, how to comprehensively strengthen the investment accuracy of distribution network, through innovation to build unmanned aerial vehicle + AI image recognition digital nuclear quantity tool, which breaks the current distribution network engineering quantity is huge, the bottleneck of on-site verification statistics.

[0009] To solve the problems of the prior art, the technical scheme is provided as follows: The method for calculating the construction cost of distribution network topography classification comprises the following steps: S1, basic model building Open source tool LabelImg, feature annotation of collected distribution network material pictures; S2, model building and training YOLOv3 deep learning model is used to build, and TensorFlow learning model is used to train and identify distribution network materials, and the distribution network materials are classified according to insulators, clamps, transformers, cables, lightning arresters, disconnectors, fuses, distribution boxes, distribution cabinets, towers, cross arms, fittings, insulator strings, conductors, grounding devices, insulator ball heads, insulator steel feet, clamp bodies, clamp pressure plates and clamp bolts; S3, data preprocessing Eliminate blurred, damaged or incomplete images; perform normalization and data enhancement processing, normalize the pixel value of the image to the [0, 1] interval, and use rotation, scaling, translation and flipping to expand the training set to improve the generalization ability of the model; S4, model training The preprocessed data set is divided into training set and validation set according to the ratio of 8:2; S5, collect the topographic information of the place where the distribution network materials are located, and count the distribution and proportion of hills, mountains, high mountains, steep ridges, swamps and river networks.

[0010] The distribution network material pictures in S1 are collected by unmanned aerial vehicle, and the specific steps are as follows: The unmanned aerial vehicle shoots the complete tower to obtain the tower information; The unmanned aerial vehicle is located at the 45° direction of the cross arm of the pole head and takes a full picture of the cross arm, insulator and clamp; Record the number of electrical equipment and stay wire on the tower; Tower coordinate acquisition, hover above the tower, take a vertical picture, and collect and obtain the tower positioning information; The unmanned aerial vehicle flies to the front of the target tower, and the unmanned aerial vehicle is located at 30-60 degrees above the tower to take a full body picture of the tower, and the tower occupies more than 70% of the height of the picture; The number of distribution network material pictures in S1 is more than 20000.

[0011] The S1 network distribution material picture is collected by a network and a drone, and the network collection network distribution material picture is used as basic data, and the S2 is used for network distribution material classification database, and the drone on-site collection real object picture is compared with the classification database.

[0012] The S1 classification database is updated or supplemented by the drone on-site collection real object picture.

[0013] The network distribution construction cost terrain classification calculation system comprises a drone module, a comment preprocessing module, a classification identification module and a training module; the network distribution material picture is collected by the drone module, the YOLOv3 deep learning model is used to build comments, and the TensorFlow learning model is used to train the classification identification network distribution material; the training module is used for multi-round training of the data set to improve the accuracy.

[0014] The present application has the following advantages: By using the drone and AI recognition technology, the network distribution engineering cost unmanned aerial vehicle intelligent auxiliary calculation is realized, and the on-site inspection problem is solved. The big data standard library is established, the image intelligent recognition on-site application is deepened, the background picture is compared, the on-site equipment material list is automatically recognized and counted, and the cost accuracy and efficiency are improved. The engineering equipment is analyzed, different terrains are compared and analyzed, and the intelligent power distribution network equipment precise cost is realized. BRIEF DESCRIPTION OF DRAWINGS

[0016] Figure 1 Pole tower photographing schematic diagram; Figure 2 Pole tower photographing schematic diagram; Figure 3 Device quantity collection schematic diagram; Figure 4 Coordinate collection schematic diagram; Figure 5 Straight line pole full body photograph schematic diagram; Figure 6 Straight line corner pole full body photograph schematic diagram; Figure 7 Straight line strain pole full body photograph schematic diagram; Figure 8 Large corner lower cross arm photographing schematic diagram; Figure 9 Terminal pole photographing schematic diagram; Figure 10 Branch pole lower cross arm photographing schematic diagram; Figure 11 Material mark schematic diagram. DETAILED DESCRIPTION

[0017] The network distribution construction cost terrain classification calculation method comprises the following steps: S1, basic model building The open source tool LabelImg is used for feature labeling of the collected network distribution material picture; S2, model building and training The YOLOv3 deep learning model was used to build the identification of distribution network materials, and the TensorFlow learning model was used to train the identification. The distribution network materials were classified into insulators, clamps, transformers, cables, surge arresters, disconnect switches, fuses, distribution boxes, distribution cabinets, poles, crossarms, hardware, insulator strings, conductors, grounding devices, insulator balls, insulator steel feet, clamp bodies, clamp plates, and clamp bolts. S3, Data Preprocessing Remove blurry, damaged, or incomplete images; perform normalization and data augmentation. Normalization normalizes the pixel values ​​of the image to the [0, 1] range, while data augmentation uses rotation, scaling, translation, and flipping to expand the training set of the original data and improve the generalization ability of the model. S4, Model Training The preprocessed dataset was divided into a training set and a validation set in an 8:2 ratio; S5. Collect terrain information on the location of distribution network materials, and statistically analyze the distribution and proportion of hills, mountains, high mountains, steep ridges, swamps, and river networks.

[0018] The images of the power distribution network materials in S1 were collected on-site using drones. The specific steps are as follows: The drone photographed the entire tower to obtain tower information; The drone was positioned at a 45° angle to the top of the pole crossarm to capture images of the crossarm, insulator, and clamps. Record the number of electrical devices and guy wires on the poles; Tower coordinate acquisition: Hover over the tower, take vertical photos, and acquire tower positioning information; The drone flies to the front of the target tower and takes a full-body photo of the tower from a 30-60 degree angle above it, with the tower occupying more than 70% of the height of the photo. The number of images of the S1 distribution network materials is more than 20,000.

[0019] The S1 distribution network material images are collected through network collection and drone acquisition. The network-collected distribution network material images serve as the basic data, and the distribution network material classification database is created through S2. The drone-collected on-site physical images are compared with the classification database.

[0020] The S1 classification database is updated or supplemented by collecting real-object images on-site using drones.

[0021] It also includes S6 cost calculation, which includes foundation construction costs, earthwork excavation and backfilling construction costs, labor transportation costs, truck transportation costs, and truck loading and unloading costs; F = Ffoundation * A + Fearthwork excavation and filling + Fhuman labor transport + Ftruck transport + Ftruck loading and unloading + Fequipment transportation and miscellaneous costs; Substitute the numerical values ​​into the formula, but do not substitute the units. The basic construction includes 10kv pole tower, 0.4kv pole tower, same pole erection, power equipment and other installation, wherein the other installation includes ordinary stay, horizontal stay, electric pole reflective film, electricity checking grounding ring and phase sequence plate; The basic construction cost Fbasic is divided into: Fbasic=F10 pole+F0.4 pole+FT+Fset+Fother+Fdemolition 10kv pole tower cost calculation F10=DX-95*4697+DX-150*6851+DX-240*9749+G*16919+T-12*1348+Z-12*1244+NZ-12*1260+D-12*1283+T-15*1581+Z-15*1477+NZ-15*1493+D-15*1516+T-18*2049+Z-18*1944+NZ-18*1961+D-18*1983 0.4kv pole tower cost calculation F0.4=DX-70 / 95*6363+DX-120 / 150*9134+T-12*1343+Z-12*1249+ZJ-12*1309+NZ-12*1317+D-12*1272+T-15*1576+Z-15*1482+ZJ-15*1542+NZ-15*1550+D-15*1505+T-18*2043+Z-18*1949+ZJ-18*2009+NZ-18*2017+D-18*1972 Same pole erection cost calculation FT=Z*1597+NZ*1672+NZJ*1680+D*1658 Equipment installation cost calculation Fset=DLQ*3512+B-100*6641+B-160*6979+B-315*7386+B-400*7659+LB-50*5639+LB-100*5855+LB-160*6193+LB-315*6601+LB-400*6874+BX-D*657+BX-S*1017 Equipment miscellaneous expenses F Device miscellaneous expenses = 1 + 1.06% + {roundup (QY-50) / 50} * 0.35% * (DLQ * 36273 + B-100 * 70512 + B-160 * 98360 + B-315 * 129066 + B-400 * 129066 + LB-50 * 17676 + LB-100 * 17676 + LB-160 * 26889 + LB-315 * 42241 + LB-400 * 42241 + BX-D * 512 + BX-S * 794) Other installation engineering cost calculation F Other = LX-P * 287 + LX-S * 1279 + FG * 4 + JD * 32 + XX * 60 Dismantling engineering cost calculation F Dismantling = BYQ * 1028 + DXG * 190 + HD * 27 + JYZ * 3 * DX * 1373 + LX * 9 + BX-10 * 123 + BX-1 * 35 Terrain comprehensive coefficient adjustment A-terrain increase coefficient Hilly-Q: 100% terrain coefficient is 1.15, every 1% terrain coefficient is 0.0015; Mountain-S: 100% terrain coefficient is 1.60, every 1% terrain coefficient is 0.006; Alpine-G: 100% terrain coefficient is 2.20, every 1% terrain coefficient is 0.012; Steep ridge-J: 100% terrain coefficient is 2.50, every 1% terrain coefficient is 0.015; Mud swamp-N: 100% terrain coefficient is 1.70, every 1% terrain coefficient is 0.007; River network-H: 100% terrain coefficient is 1.20, every 1% terrain coefficient is 0.002; A = 1 + Q% * 100 * 0.0015 + S% * 100 * 0.006 + G% * 100 * 0.012 + J% * 100 * 0.015 + N% * 100 * 0.007 + H% * 100 * 0.002 Earthwork excavation and backfill construction cost F Earthwork excavation and backfill = F10 pole earthwork + F0.4 pole earthwork + FT earthwork Earthwork components are divided into ordinary soil-PT; hard soil-JT; loose sandstone-SS; rock-YS; mud pit-NSK; quicksand pit-LSK; dry sand pit-GSK; water pit-SK; frozen soil-DT; earthwork component ratio is calculated, 10kv pole tower earthwork excavation and backfill construction cost F10 pole earthwork = G * 15 * (PT% * 67.86 + JT% * 75.88 + SS% * 120.41 + YS% * 244.18) + {4.48 * (T-12 + Z-12 + NZ-12 + D-12) + 8.58 * (T-15 + Z-15 + NZ-15 + D-15) + 12.32 * (T-18 + Z-18 + NZ-18 + D-18)} * (PT% * 25.05 + JT% * 30.72 + SS% * 120.41 + YS% * 262.33 + NSK% * 84.17 + LSK% * 138.90 + GSK% * 98.81 + SK% * 55.54 + DT% * 85.69) 0.4kv pole earthwork excavation backfill construction cost F0.4 pole earthwork = {4.48 * (T-12 + Z-12 + ZJ-12 + NZ-12 + D-12) + 8.58 * (T-15 + Z-15 + ZJ-15 + NZ-15 + D-15) + 12.32 * (T-18 + Z-18 + ZJ-18 + NZ-18 + D-18)} * (PT% * 25.05 + JT% * 30.72 + SS% * 120.41 + YS% * 262.33 + NSK% * 84.17 + LSK% * 138.90 + GSK% * 98.81 + SK% * 55.54 + DT% * 85.69) Same pole erection earthwork excavation backfill construction cost FT earthwork = 8.58 * (T-15 + Z-15 + ZJ-15 + NZ-15 + D-15) * (PT% * 25.05 + JT% * 30.72 + SS% * 120.41 + YS% * 262.33 + NSK% * 84.17 + LSK% * 138.90 + GSK% * 98.81 + SK% * 55.54 + DT% * 85.69) Manpower transportation cost, Fmanpower transportation = F10 pole manpower transportation + F0.4 pole manpower transportation + FT pole manpower transportation + Fset manpower transportation + Fother manpower transportation; XY-manpower distance, unit km; 10kv pole tower manpower transportation cost F10 pole manpower transportation = XY (DX-95 * 408.64 + DX-150 * 438.50 + DX-240 * 692.78 + G * 827.21 + T-12 * 569.29 + Z-12 * 557.01 + NZ-12 * 562.90 + D-12 * 563.57 + T-15 * 1123.38 + Z-15 * 1111.09 + NZ-15 * 1117.15 + D-15 * 1117.66 + T-18 * 1411.59 + Z-18 * 1399.30 + NZ-18 * 1405.36 + D-18 * 1405.87) 0.4kv tower manpower transportation cost F0.4 tower manpower transportation = XY (DX-70 / 95*544.93 + DX-120 / 150*584.66 + T-12*577.88 + Z-12*558.86 + ZJ-12*562.90 + NZ-12*593.20 + D-12*569.46 + T-15*1131.80 + Z-15*1109.91 + ZJ-15*1117.32 + NZ-15*1134.66 + D-15*1123.55 + T-18*1420.01 + Z-18*1397.96 + ZJ-18*1405.36 + NZ-18*1422.70 + D-18*1411.09) same tower erection manpower transportation cost FT manpower transportation = XY (Z*1116.65 + NZ*1131.46 + NZJ*1179.10 + D*1143.92) equipment installation manpower transportation cost F equipment manpower transportation = XY (DLQ*23.74 + B-100*51.85 + B-160*51.85 + B-315*51.85 + B-400*51.85 + LB-50*51.85 + LB-100*51.85 + LB-160*51.85 + LB-315*51.85 + LB-400*51.85 + BX-D*15.55 + BX-S*25.56) other installation engineering manpower transportation cost F other manpower transportation = XY (LX-P*59.49 + LX-S*494.2 + FG*0.08 + JD*0.84 + XX*1.18) automobile transportation cost, F automobile transportation = F10 tower automobile transportation + F0.4 tower automobile transportation + FT automobile transportation + F equipment automobile transportation + F other automobile transportation; QY-automobile transportation distance, unit km; 10kv tower automobile transportation cost F10 tower automobile transportation = QY (DX-95*3.24 + DX-150*3.48 + DX-240*5.50 + G*47.67 + T-12*4.12 + Z-12*3.98 + NZ-12*4.04 + D-12*4.05 + T-15*6.61 + Z-15*6.47 + NZ-15*6.53 + D-15*6.54 + T-18*8.19 + Z-18*8.05 + NZ-18*8.12 + D-18*8.12) 0.4kv tower automobile transportation cost F0.4 pole tower truck transportation = QY (DX-70 / 95*4.32 + DX-120 / 150*4.64 + T-12*4.21 + Z-12*4 + ZJ-12*4.04 + NZ-12*4.39 + D-12*4.12 + T-15*6.70 + Z-15*6.45 + ZJ-15*6.54 + NZ-15*6.73 + D-15*6.61 + T-18*8.28 + Z-18*8.03 + ZJ-18*8.12 + NZ-18*8.32 + D-18*8.18) Same pole erection truck transportation cost FT truck transportation = QY (Z*6.53 + NZ*6.7 + NZJ*7.24 + D*6.84) Equipment installation truck transportation cost F set truck transportation = QY (DLQ*0.27 + B-100*0.59 + B-160*0.59 + B-315*0.59 + B-400*0.59 + LB-50*0.59 + LB-100*0.59 + LB-160*0.59 + LB-315*0.59 + LB-400*0.59 + BX-D*0.15 + BX-S*0.25) Other installation engineering truck transportation cost F other truck transportation = QY (LX-P*0.5 + LX-S*3.45 + FG*0.01 + JD*0.01 + XX*0.01) Truck loading and unloading cost, F truck loading and unloading = F10 pole truck loading and unloading + F0.4 pole truck loading and unloading + FT truck loading and unloading + F set truck loading and unloading + F other truck loading and unloading 10kv pole tower truck loading and unloading cost F10 pole truck loading and unloading = DX-95*97.51 + DX-150*104.63 + DX-240*165.31 + G*568.12 + T-12*163.47 + Z-12*159.38 + NZ-12*161.34 + D-12*161.57 + T-15*285.78 + Z-15*281.69 + NZ-15*283.71 + D-15*283.88 + T-18*359.17 + Z-18*355.09 + NZ-18*357.10 + D-18*357.27 F0.4 pole tower truck loading and unloading cost F0.4 pole car loading and unloading = DX-70 / 95*130.03 + DX-120 / 150*149.51 + T-12*166.33 + Z-12*160 + ZJ-12*161.34 + NZ-12*171.43 + D-12*163.53 + T-15*288.58 + Z-15*281.30 + ZJ-15*283.77 + NZ-15*289.53 + D-15*285.84 + T-18*361.97 + Z-18*354.64 + ZJ-18*357.10 + NZ-18*362.87 + D-18*359.01 Same pole erection car loading and unloading cost FT car loading and unloading = Z*283.54 + NZ*288.47 + NZJ*304.32 + D*292.62 Equipment installation car loading and unloading cost Fset car loading and unloading = DLQ*7.9 + B-100*17.25 + B-160*17.25 + B-315*17.25 + B-400*17.25 + LB-50*17.25 + LB-100*17.25 + LB-160*17.25 + LB-315*17.25 + LB-400*17.25 + BX-D*4.47 + BX-S*7.3 Other installation engineering car loading and unloading cost Fother car loading and unloading = LX-P*14.88 + LX-S*144.47 + FG*0.03 + JD*0.28 + XX*0.39.

[0022] The distribution network construction cost terrain classification calculation system comprises a unmanned aerial vehicle module, a comment preprocessing module, a classification identification module and a training module; through the unmanned aerial vehicle module, distribution network material pictures are collected, YOLOv3 deep learning model is adopted to build comments, and TensorFlow learning model is adopted to train classification identification of distribution network materials; the training module trains the data set for multiple rounds to improve the accuracy. The distribution network construction cost terrain classification calculation application is characterized in that it is applied to terrain classification cost.

[0023] The specific implementation case is explained as follows: (I) Construction of distribution network material AI identification model The construction of the distribution network material AI identification model is the core link of realizing the digital cost operation of the distribution network engineering. From the collection and processing of data to the construction, training and optimization of the model, each step is closely linked and lays a foundation for accurate material identification.

[0024] To build a high-performance AI intelligent recognition model, we chose the mature open-source tool LabelImg, available from the official GitHub repository. This tool can be used to perform feature labeling for different regions and types of over 20,000 collected distribution network image materials. With the help of computer image deep learning algorithms, we can deeply mine key features from massive data, laying a solid foundation for subsequent accurate recognition.

[0025] During the model building process, we referred to the patent "A material matching method for distribution network engineering" (Publication No. CN119152266A, Application Date September 2024) applied by the Economic and Technical Research Institute of State Grid Anhui Electric Power Co., Ltd., and the patent "A contact network engineering material processing system based on intelligent recognition" (Publication No. CN119107024A, Application Date August 2024) jointly applied by the First Engineering Company of China Railway Electrification Bureau Group and China Railway Electrification Bureau Group Co., Ltd. These patents have accumulated rich practical experience and advanced technical ideas in the field of distribution network material recognition.

[0026] Based on the YOLOv3 deep learning model architecture, which has excellent performance in object detection and is fast and accurate, we used the TensorFlow machine learning framework for comprehensive and systematic training. The specific process is as follows: Data Preparation Data Collection: We collected image data containing various distribution network materials from multiple distribution network engineering sites, a total of 35,000 images, covering common materials such as insulators, clamps, transformers, and cables.

[0027] Screening and Labeling: We strictly screened the collected images, eliminating blurred, damaged, or incomplete images, and finally retained 31,740 images. Using the professional labeling tool LabelImg, experienced power engineers labeled the materials in the images, with a labeling category of 20.

[0028] Data Preprocessing: Normalization: Normalize the pixel values of the image to the interval [0, 1] to accelerate the convergence speed of model training. When performing the normalization operation, it is to process the labeled image. Because the labeled image contains the category information of the material, it is the data used for model training. The specific normalization formula is: \(x_{norm}=\frac{x- x_{min}}{x_{max}-x_{min}}\), where x is the original image pixel value, \(x_{min}\) and \(x_{max}\) are the minimum and maximum values of all pixel values of the image respectively. Through this formula, each pixel value is mapped to the interval [0, 1], so that different images remain consistent in pixel value scale, which helps the model converge faster.

[0029] Data augmentation: Use data augmentation techniques such as rotation, scaling, translation, and flipping to process 31740 high-quality image data and improve the generalization ability of the model. For example, rotate an insulator image by 45 degrees and then horizontally flip it to get new training samples.

[0030] Model construction ResNet50 based on convolutional neural network (CNN) with good feature extraction capability in image recognition tasks is selected as the basic architecture, and the model is adjusted as follows according to the characteristics of the distribution network material identification: Adjust the classification layer: Adjust the output node number of the original model's classification layer to 20 to match the number of categories of distribution network materials. The category list of distribution network materials includes insulator, clamp, transformer, cable, surge arrester, disconnector, fuse, distribution box, distribution cabinet, tower, cross arm, hardware, insulator string, conductor, grounding device, insulator ball, insulator steel foot, clamp body, clamp pressing plate, clamp bolt, etc. 20 kinds. In the model, the output node number of the classification layer determines the number of categories that the model can predict. Setting the output node number to 20 allows the model to classify and predict the 20 types of distribution network materials. For example, after training, the model will perform feature extraction and analysis on the input image data, and finally output the prediction probability of each category on the 20 output nodes. The category with the highest probability is the model's prediction result.

[0031] Add attention mechanism: Introduce SE (Squeeze-and-Excitation) attention module in the middle layer of the model to make the model pay more attention to the key features of the material and improve the recognition accuracy.

[0032] Training parameter settings Loss Function: The cross-entropy loss function is used to measure the difference between the model's prediction and the true label. The formula of the cross-entropy loss function is: \(L = -\frac{1}{N}\sum_{i = 1}^{N}\sum_{j = 1}^{C}y_{ij}\log(p_{ij})\), where N is the number of samples, C is the number of classes, \(y_{ij}\) represents the true label of the i-th sample belonging to the j-th class (usually represented by one-hot encoding, belonging to this class is 1, otherwise 0), \(p_{ij}\) is the probability of the model predicting that the i-th sample belongs to the j-th class. The loss value calculated by this formula reflects the difference between the model's prediction and the true label. The smaller the loss value, the closer the model's prediction is to the true label.

[0033] Optimizer: The Adam optimizer is selected, with the learning rate initialized to 0.001, β1 = 0.9, β2 = 0.999, and ε = 1e - 8. The learning rate determines the step size of parameter updates during training. If the learning rate is too large, the model may skip the optimal solution during training, resulting in failure to converge; if the learning rate is too small, the training speed of the model will be very slow. β1 and β2 are momentum parameters in the Adam optimizer, β1 is used to calculate the first-order moment estimate of the gradient, and β2 is used to calculate the second-order moment estimate of the gradient. β1 is set to 0.9, meaning that 90% of the previous step gradient information and 10% of the current gradient information are retained when calculating the first-order moment estimate; β2 is set to 0.999, meaning that 99.9% of the previous second-order moment information and 0.1% of the current second-order moment information are retained when calculating the second-order moment estimate. ε is a very small constant used to prevent the denominator from being zero, set to 1e - 8 to ensure numerical stability during calculation. These parameters work together to affect the way and speed of parameter updates during training, and thus affect the difference between the model's prediction and the true label.

[0034] Training Epochs: Set to 50 rounds, each round traverses the entire training dataset once. As the number of training rounds increases, the model gradually learns the characteristics of the network materials, and in the early stage, the accuracy of the model improves significantly, and the loss value decreases significantly. As the number of training rounds increases, the accuracy improvement slows down, and the loss value tends to be stable. This is because the model can quickly learn some obvious features in the data in the early stage of training, and as the training progresses, there are fewer new features to learn, so the improvement speed will slow down.

[0035] Batch Size: Set to 32, meaning 32 images are used for parameter updates during each training iteration. The choice of batch size affects the efficiency and effectiveness of model training. A larger batch size can make the model more stable when updating parameters because it incorporates information from more samples; however, it will also increase memory consumption and training time. Choosing 32 as the batch size strikes a balance between training efficiency and memory consumption.

[0036] Model training The preprocessed dataset was divided into a training set and a validation set in an 8:2 ratio. The training set contained 32,392 images, and the validation set contained 6,348 images.

[0037] During training, an evaluation is performed on the validation set after each epoch, and the model's accuracy and loss value are recorded in detail. The accuracy is calculated as: \(Accuracy = \frac{number of correctly predicted samples}{total number of samples}\). For example, if there are 6348 images in the validation set, and the model correctly predicts the material category for 5830 images, then the accuracy is \(\frac{5830}{6348}\approx 0.92\), or 92%. The loss value is calculated using the cross-entropy loss function mentioned above, reflecting the degree of difference between the model's predictions on the validation set and the true labels. The detailed records of the model's accuracy and loss value are as follows: As training progresses, the model gradually learns the characteristics of the distribution network materials. In the first few epochs, the model's accuracy improves significantly, and the loss value decreases considerably. As the number of training epochs increases, the rate of accuracy improvement gradually slows down, and the decrease in loss value also tends to stabilize, showing an overall trend of continuously improving accuracy and gradually decreasing loss value.

[0038] Model Evaluation and Optimization Evaluation indicators: Use accuracy, recall, and F1 score as evaluation indicators. The formula for calculating recall is: \(Recall=\frac{Number of correctly predicted samples}{Number of actual samples of that category}\). For example, suppose there are 1000 samples of insulating material in the validation set, and the model correctly predicts 900 samples of insulating material. The recall rate of insulating material is \(\frac{900}{1000}=0.9\), which is 90%. F1 score is an indicator that considers both accuracy and recall, and its formula is: \(F1 - Score=\frac{2\times Recall\times Accuracy}{Recall + Accuracy}\). Taking the previously calculated accuracy of 0.92 and recall of 0.9 as an example, \(F1 -Score=\frac{2\times0.9\times0.92}{0.9 + 0.92}\approx0.91\).

[0039] Evaluation results: After training, the model's accuracy on the validation set reached 92%, recall was 90%, and F1 score was 0.91.

[0040] Optimization measures: According to the evaluation results, it was found that the model's recognition accuracy for some similar-shaped materials (such as different types of wire clips) was low. To address this issue, the number of samples of such materials was further increased, and the model was fine-tuned. After retraining, the model's accuracy, recall, and F1 score on the validation set were improved, which were recalculated based on the above formulas. For example, after increasing the number of samples and fine-tuning the model, the number of correctly predicted samples in the validation set increased, and the number of actual samples of that category remained unchanged. Therefore, the accuracy and recall rates will increase accordingly, and the F1 score will also increase.

[0041] Final training results After optimization, the model was tested on an independent test set, and the final accuracy, recall, and F1 score were calculated based on the above formulas.

[0042] The recognition of each type of distribution network material by the model can be clearly seen from the confusion matrix. The diagonal elements represent the number of correctly classified samples, and the non-diagonal elements represent the number of misclassified samples. The accuracy is still calculated by dividing the number of correctly predicted samples in the test set by the total number of samples. For example, in the test set, there are 1000 samples, and the model correctly classifies 940 of them. The accuracy is \(\frac{940}{1000}=0.94\), which is 94%. By analyzing the confusion matrix, we can find that the model has high recognition accuracy for most materials, but there are still a small number of categories that are misclassified. Further optimization can be done in the future.

[0043] Gansu distribution network engineering material intelligent recognition sample library construction and graphic data labeling Using advanced unmanned aerial vehicle equipment, we take high-definition pictures of line engineering tower equipment materials from all angles and multiple angles. The flight height, angle, and shooting time of the unmanned aerial vehicle are carefully planned to ensure that the collected pictures can clearly display the details of the materials, laying a solid foundation for the construction of the sample library.

[0044] According to the characteristics of different pole types in special areas of Gansu, we have developed standard operating procedures for unmanned aerial vehicle cost collection for 6 types of poles (straight, strain, corner, terminal, branch, etc.), 31 types of cross arm heads of 4 types of State Grid standard designs, to guide grassroots personnel to quickly carry out operations and clearly define the shooting content, standardize the collection of on-site engineering cost data. Among them, the 4 types of State Grid standard designs are general tower design, substation building design, distribution equipment design, and cable line design The specific collection contents are as follows: 1. Clearly define the type of tower and obtain the pole number information. By taking pictures of the entire tower, we can understand the basic situation of the tower from a macro perspective, providing basic information for subsequent material recognition and statistics.

[0045] 2. Take pictures of the pole head size and side from 45°, and take pictures of the cross arm from 45°: Collect all the material information on the pole, such as cross arm, insulator, and clamp. This multi-angle shooting method can fully display the details of the pole head, ensuring that no important information is missed.

[0046] 3. Collect column equipment and stay wire: Record the number of electrical equipment and stay wire on the pole. These equipment and stay wire are important components of the normal operation of the distribution network, and accurate statistics of their number are of great significance to engineering construction and maintenance.

[0047] 4. Tower coordinate collection: Hover above the tower, take pictures vertically, and collect tower positioning information. Accurate coordinate positioning helps to accurately label the tower in the geographic information system, facilitating subsequent line planning and maintenance management.

[0048] 5. Tower full-body collection: The UAV takes off in front of the target tower and takes a full-body shot of the tower. The UAV is positioned directly above the wire at an angle of 30-60 degrees to ensure that the tower's full-body occupies more than 70% of the photo's height.

[0049] 6. Large-angle strain rod, terminal rod, branch rod information collection: Control the UAV to the small side of the tower, 3m away from the tower, 2m above the wire, and take a photo of the rod head at a 45-degree angle. The numerical value is for reference only, and the specific photo standard is subject to the example.

[0050] Power distribution network engineering involves a wide variety of materials, such as towers, transformers, switchgear, utility poles, insulators, wires, connectors, etc. Each type of material has different models and specifications. In the label making process, team members first use a UAV to take photos of the materials at the distribution network engineering site, then establish a material database based on the photos, and finally identify and match the models through the UAV. When implementing this solution, the UAV must ensure that the shooting angle, light, and other conditions are suitable to obtain clear and complete material images. After the photo collection is complete, image analysis software is used to process the photos, extract the key features of the materials, and establish a database containing material images, features, and corresponding model information. In subsequent identification, the UAV is used to collect images, and image recognition algorithms are used to match the new database to determine the material model. The main list of materials collected is as follows: The following table shows the classification of cross arms, insulators, and fittings. In this case, the classification is only for the convenience of data statistics, and the classification can be adjusted according to the needs of power distribution network engineering

[0051] Specific explanation and optimization comparison of the problem of numerous materials and complex statistics: 1. Numerous materials Power distribution network engineering involves a wide variety of materials, such as towers, transformers, switchgear, utility poles, insulators, wires, connectors, etc. Each type of material has different models and specifications. This diversity makes it extremely complex to identify and annotate these materials in images.

[0052] 2. Complex statistics When dealing with large amounts of image data, how to efficiently count and analyze the number, type, and location of materials to ensure the accuracy and consistency of the data is a significant challenge.

[0053] 3. Comparison before and after technical optimization

[0054] Algorithm and calculation process: A deep learning-based target detection algorithm such as the YOLO (You Only Look Once) series is introduced. Taking YOLOv5 as an example, its network structure includes a backbone network, a neck network, and a head network. The backbone network is used to extract the features of the image, the neck network further processes and fuses the features extracted by the backbone network, and the head network classifies and locates the target based on the processed features. During the training phase, a large number of image data containing various types of network engineering materials are used to train the model, and the model parameters are continuously adjusted through the backpropagation algorithm, so that the model can accurately identify different materials. In the labeling process, the model processes the input image quickly based on the learned features, and identifies the category and location information of the materials. Assuming that each picture contains an average of 5 materials, before optimization, manual labeling of each picture takes an average of 2 minutes (i.e. 120 seconds), so the time required to label 31740 pictures is: \(31740×120 = 3808800\) seconds, which is equivalent to 1058 man-hours. After introducing the YOLOv5 algorithm, the model processes each picture in an average of 0.4 seconds, so the time required to label 31740 pictures is \(31740×0.4 = 12696\) seconds, which is equivalent to 3.53 man-hours, and the labeling time is shortened to about 1 / 300 of the original, and in actual application, it will also be affected by factors such as hardware performance, etc. This simplifies the calculation to reflect the efficiency improvement.

[0055] Error rate definition: The error rate here refers to the proportion of the labeled material information that does not match the actual material information, and the calculation formula is: error rate = (number of incorrectly labeled materials ÷ total number of labeled materials) × 100%. For example, in a labeling task, a total of 1000 materials were labeled, of which 10 materials were misidentified or missed, so the error rate is \((10 ÷ 1000) × 100% = 1%\).

[0056] Review standards: The review adopts a combination of double cross-checking and random sampling review. Double cross-checking means that two different annotators check each other's labeling results for the same batch of images, and if they find inconsistencies, they discuss and determine the correct labeling together. Random sampling review is conducted by quality management personnel according to a certain proportion (such as 10%), and if the error rate exceeds the specified threshold (such as 1%), the batch of images needs to be rechecked.

[0057] Labeling specification technical requirements: The annotation specification mainly includes the following technical requirements. First, in terms of annotation format, PascalVOC format is used, that is, each annotation result is saved as an XML file, which contains basic information of the image (such as file name, image size, etc.), and object information of the annotation (such as category, bounding box coordinates, etc.). Second, for the annotation of material categories, the standard name and alias of each material are clearly specified, for example, the standard name of the pole tower is "pole tower", and the alias is "iron tower" and "cement pole tower". The standard name must be used for annotation. In terms of bounding box annotation, the bounding box must tightly enclose the material, and the coordinates of the bounding box are integer pixels, with the upper left corner as the starting point and the lower right corner as the ending point. In addition, for some special materials, such as complex equipment with multiple components, the annotation sequence from the whole to the local is specified, that is, the whole equipment is annotated first, and then each component is annotated.

[0058] Automatic statistical software and source: The selected automatic statistical software is the statistical module in OpenCV-Contrib, an extension library of OpenCV. OpenCV is an open-source computer vision library, and OpenCV-Contrib contains many extended functions. In material statistics, the related functions in the statistical module are used, such as the cv2.countNonZero() function, which can count the number of specific pixel values (corresponding to the annotated material area) in the image, thereby realizing the statistics of the number of materials. OpenCV can be downloaded and used for free from its official website (https: / / opencv.org / releases / ).

[0059] Calculation formula: Taking the number of a certain type of material in the image as an example, assuming that the annotated material area in the image is represented by a specific pixel value (such as 255), the number of pixels with a pixel value of 255 in the image is obtained through the cv2.countNonZero() function, and then the number of materials is calculated according to the pre-set pixel number threshold corresponding to each material (such as the pixel number threshold corresponding to each pole tower is 1000). The calculation formula is: number of materials = cv2.countNonZero (annotated image) ÷ pixel number threshold corresponding to each material. For example, the return value of cv2.countNonZero(annotated image) is 5000, and the pixel number threshold corresponding to each pole tower is 1000, then the number of pole towers is \(5000 ÷ 1000 = 5\). 4. Comparison and summary before and after optimization Through the above optimization measures, the labeling efficiency has been greatly improved. The labeling work that originally required a lot of manpower and time can now be completed quickly. The error rate is greatly reduced, effectively avoiding the adverse effects of labeling errors on subsequent model training and analysis. Data consistency problems have been properly solved, providing a more reliable data foundation for model training. Statistical efficiency has also made a qualitative leap, enabling real-time grasp of material data changes and providing timely and accurate data support for decision-making. These optimization measures have comprehensively improved the quality and efficiency of distribution network material labeling work, laying a solid foundation for the construction and application of distribution network material AI recognition models. Detailed description of image labeling process To ensure the efficiency and accuracy of data labeling, here are the detailed steps for labeling images using the LabelImg tool: Preparation stage Labeling process Quality check Result display Here is an example table of the labeling results: Application cases of image labeling in the project In the construction process of the distribution network material recognition model in Longnan, data labeling plays a crucial role. The project team carefully prepares and strictly controls every link, striving to provide high-quality data support for model training.

[0060] I. Data collection and preparation The project personnel carefully selected a total of 31740 material images. These images cover different geographical environments, construction conditions and equipment states, with rich diversity, laying a solid foundation for the generalization ability of the subsequent model.

[0061] II. Labeling tools and personnel Select image labeling tools and organize professional labeling personnel to participate in the work. These personnel have solid electrical professional knowledge and rich image labeling experience, ensuring the professionalism and accuracy of the labeling work.

[0062] III. Labeling specifications and implementation The marking work involves 16 kinds of materials such as cross arm, hoop, inclined support, insulator and hardware. The marking process strictly follows the established marking specification, and the marking personnel performs detailed marking process for each kind of material. Taking the cross arm as an example, the marking personnel not only accurately outlines the contour thereof by using a tool, but also records key attribute information such as model and specification thereof in depth; for the insulator, in addition to marking the position, the type thereof is also specified, and possible defect features such as crack and damage are carefully distinguished and marked in detail.

[0063] The terrain classification in the present document is obtained according to the 'Geographic Information Data Product Specification' GB / T 25528-2010 and related geographic public information.

[0064] It needs to be explained in the case that electric power materials are various, and type categories are various in different regions. In the case, only the main applied equipment in Gansu and Gansu Tibetan area is classified and summarized. The coefficient in the calculation formula is the estimated value of the cost, which is used for explaining the cost method. The coefficient can be adjusted according to local labor cost and equipment cost. Since the quantity of electric power materials is huge, all the coefficients cannot be listed. The innovation of the case lies in providing a cost calculation scheme combined with equipment and terrain.

[0065] The present application has the following beneficial effects: The unmanned aerial vehicle can quickly reach complex terrain and inconvenient transportation areas, greatly reducing the time cost of data collection. Combined with the automatic calculation technology of line span, the unmanned aerial vehicle is more efficient and accurate in data collection. From the beginning of the cost project creation, the unmanned aerial vehicle quickly enters the "patrol" work to quickly obtain positioning information and basic account data. Then, through the GIS system and unstructured data interface, automatic data collection and analysis are realized. This process reduces a lot of manual processing links and greatly saves time. According to the research results of "Unstructured Data Processing and Analysis Method for Electric Power Engineering Audit", unstructured data plays an important role in electric power engineering cost, and efficient processing is the key to improving cost efficiency. The "Electric Power Engineering Document Structured Data Extraction Method and System" patent (publication number CN114282333A) of China Southern Power Grid Company provides an effective technical solution to solve this problem. The patent describes in detail the construction and implementation of the structured data extraction model. Through text analysis, semantic understanding, and template matching techniques, it can accurately extract key information such as project name, construction unit, and project progress from unstructured engineering submission documents. This case is based on cost auxiliary calculation applied to various terrains such as mountains, plateaus, plains, valleys, deserts, and Gobi. The addition of unmanned aerial vehicle assisted collection further improves the applicability of electric power materials in Gansu terrain.

[0066] Greatly improve accuracy: In the material cost work of distribution network, the traditional method is easily disturbed by human factors, which limits the accuracy of the cost. The system introduces monocular vision ranging and AI image intelligent recognition technology, and cooperates with unmanned aerial vehicle for on-site "patrol". After the unmanned aerial vehicle collects data, the built-in material recognition model of the system starts to work, using convolutional neural network to accurately identify distribution network materials. Compared with the traditional method, the intelligent analysis model improves the accuracy and recall rate by more than 20%, greatly improving the accuracy of the cost. It is worth mentioning that the patent "Electric Power Equipment Intelligent Recognition Method and System Based on Deep Learning" (publication number: CN113012362A) disclosed by China Southern Power Grid Company also provides valuable practical guidance for us. The patent focuses on the method of using convolutional neural network to extract and fuse various features of electric power equipment to achieve accurate identification. Its technical solution in feature extraction and classification recognition is highly consistent with our material recognition model construction needs, providing an important reference for optimizing system performance.

[0067] Effectively reduce cost: The entire cost process reduces the dependence on a large number of human and material resources. With the help of digital technology, from project creation to final output of doubtful points report, it realizes efficient operation of each link, reduces the overall cost, and reduces the cost by more than 30% compared with the traditional cost.

[0068] The human cost is effectively saved. By fully exerting the advantages of artificial intelligence technology in engineering cost work, the type and quantity of materials on the pole of distribution network engineering are quickly identified and counted, and the list of suspected points is quickly reported, thereby reducing the workload of on-site cost personnel verification. At the same time, the problem of large material counting error caused by uneven business quality of cost personnel is also avoided. At present, the tool has been used in more than 10 batches of engineering cost in Pingyi and Longnan in 2023. Compared with the traditional manual on-site verification method, the on-site verification time is compressed by more than 75% on average, the accuracy rate of tool material identification reaches 77.1%, which greatly improves the work efficiency of on-site cost personnel, greatly reduces the burden of on-site "quantity verification" of cost personnel, effectively reduces the cost of human resources, and improves the quality and efficiency of cost work.

Claims

1. A method for calculating the construction cost of power distribution networks based on terrain classification, characterized in that... Includes the following steps: S1, Basic Model Building The open-source tool LabelImg is used for feature annotation of collected images of power distribution network materials. S2, Model Building and Training The YOLOv3 deep learning model was used to build the identification of distribution network materials, and the TensorFlow learning model was used to train the identification. The distribution network materials were classified into insulators, clamps, transformers, cables, surge arresters, disconnect switches, fuses, distribution boxes, distribution cabinets, poles, crossarms, hardware, insulator strings, conductors, grounding devices, insulator balls, insulator steel feet, clamp bodies, clamp plates, and clamp bolts. S3, Data Preprocessing Remove blurry, damaged, or incomplete images; Normalization and data augmentation are performed. Normalization normalizes the pixel values ​​of the image to the range of [0, 1]. Data augmentation uses rotation, scaling, translation, and flipping to expand the training set of the original data and improve the generalization ability of the model. S4, Model Training The preprocessed dataset was divided into a training set and a validation set in an 8:2 ratio; S5. Collect terrain information on the location of distribution network materials, and statistically analyze the distribution and proportion of hills, mountains, high mountains, steep ridges, swamps, and river networks.

2. The method for calculating the construction cost of power distribution networks based on terrain classification according to claim 1, characterized in that... The images of the power distribution network materials in S1 were collected on-site using drones. The specific steps are as follows: The drone photographed the entire tower to obtain tower information; The drone was positioned at a 45° angle to the top of the pole crossarm to capture images of the crossarm, insulator, and clamps. Record the number of electrical devices and guy wires on the poles; Tower coordinate acquisition: Hover over the tower, take vertical photos, and acquire tower positioning information; The drone flies to the front of the target tower and takes a full-body photo of the tower from a 30-60 degree angle above it, with the tower accounting for more than 70% of the height of the photo.

3. The method for calculating the construction cost of power distribution networks based on terrain classification according to claim 1, characterized in that... The number of images of the S1 distribution network materials is more than 20,000.

4. The method for calculating the construction cost of power distribution networks based on terrain classification according to claim 2, characterized in that... The S1 distribution network material images are collected through network collection and drone acquisition. The network-collected distribution network material images serve as the basic data, and the distribution network material classification database is created through S2. The drone-collected on-site physical images are compared with the classification database.

5. The method for calculating the construction cost of power distribution networks based on terrain classification according to claim 4, characterized in that... The S1 classification database is updated or supplemented by collecting real-object images on-site using drones.

6. The method for calculating the construction cost of power distribution networks based on terrain classification according to claim 4, characterized in that... It also includes S6 cost calculation, which includes foundation construction costs, earthwork excavation and backfilling construction costs, labor transportation costs, truck transportation costs, and truck loading and unloading costs; F = Ffoundation * A + Fearthwork excavation and filling + Fhuman labor transport + Ftruck transport + Ftruck loading and unloading + Fequipment transportation and miscellaneous costs; Substitute the numerical values ​​into the formula, but do not substitute the units. The basic construction includes 10kV poles and towers, 0.4kV poles and towers, pole erection, power equipment and other installations, including ordinary guy wires, horizontal guy wires, pole reflective film, voltage testing grounding rings and phase sequence plates; The foundation construction cost, Ffoundation, is divided into: Ffoundation = F10 pole + F0.4 pole + FT + Fdesign + Fother + Fdemolition Calculation of 10kV tower costs F10=DX-95*4697+DX-150*6851+DX-240*9749+G*16919+T-12*1348+Z-12*1244+NZ-12*1260+D-12*1 283+T-15*1581+Z-15*1477+NZ-15*1493+D-15*1516+T-18*2049+Z-18*1944+NZ-18*1961+D-18*1983 Calculation of 0.4kV tower costs F0.4=DX-70 / 95*6363+DX-120 / 150*9134+T-12*1343+Z-12*1249+ZJ-12*1309+NZ-12*1317+D-12*1272+T-15*157 6+Z-15*1482+ZJ-15*1542+NZ-15*1550+D-15*1505+T-18*2043+Z-18*1949+ZJ-18*2009+NZ-18*2017+D-18*1972 Calculation of Cost of Erecting the Same Pole FT=Z*1597+NZ*1672+NZJ*1680+D*1658 Equipment installation cost calculation F = DLQ*3512+B-100*6641+B-160*6979+B-315*7386+B-400*7659+LB-50*5639 +LB-100*5855+LB-160*6193+LB-315*6601+LB-400*6874+BX-D*657+BX-S*1017 Equipment transportation and handling fees F Equipment Freight and Miscellaneous Charges = 1 + 1.06% + {roundup(QY-50) / 50} * 0.35% * (DLQ*36273 + B-100*70512 + B-160*98360 + B-315*129066 + B-400*129066 + LB-50*17676 + LB-100*17676 + LB-160*26889 + LB-315*42241 + LB-400*42241 + BX-D*512 + BX-S*794) Calculation of other installation engineering costs F_other = LX - P * 287 + LX - S * 1279 + FG * 4 + JD * 32 + XX * 60 Demolition project cost calculation F_demolition = BYQ*1028 + DXG*190 + HD*27 + JYZ*3*DX*1373 + LX*9 + BX-10*123 + BX-1*35 Terrain Comprehensive Coefficient Adjustment A - Terrain Increase Coefficient Hilly terrain - Q: The terrain percentage coefficient is 1.15 for 100% terrain and 0.0015 for every 1% terrain percentage; Mountainous - S: The terrain percentage coefficient is 1.60 for 100% terrain and 0.006 for every 1% terrain percentage; High Mountain - G: The terrain coverage coefficient is 2.20 for 100% terrain coverage, and 0.012 for every 1% terrain coverage. Junling-J: The terrain coverage coefficient is 2.50 for 100% terrain coverage, and 0.015 for every 1% terrain coverage. Mudslide-N: The terrain coverage coefficient is 1.70 for 100% terrain coverage and 0.007 for every 1% terrain coverage. River network - H: The topography percentage coefficient is 1.20 for 100% topography and 0.002 for every 1% topography percentage; A=1+Q%*100*0.0015+S%*100*0.006+G%*100*0.012+J%*100*0.015+N%*100*0.007+H%*100*0.002 Earthwork excavation and backfilling construction costs F earthwork excavation and filling = F10 pole earthwork + F0.4 pole earthwork + FT earthwork Earthwork components are categorized as follows: ordinary soil (PT); firm soil (JT); loose sand and gravel (SS); rock (YS); muddy water pit (NSK); quicksand pit (LSK); dry sand pit (GSK); water pit (SK); and frozen soil (DT). The percentage of each earthwork component is used for calculation. 10kV tower excavation and backfilling construction cost F10 pole earthwork=G*15*(PT%*67.86+JT%*75.88+SS%*120.41+YS%*244.18)+{4.48*(T-12+Z-12+NZ-12+D-12)+8.58*(T-15+Z-15+NZ-15+D-15)+12.32*( T-18+Z-18+NZ-18+D-18)}*(PT%*25.05+JT%*30.72+SS%*120.41+YS%*262.33+NSK%*84.17+LSK%*138.90+GSK%*98.81+SK%*55.54+DT%*85.69) Construction cost of excavation and backfilling for 0.4kV power pole F0.4 rod earthwork={4.48*(T-12+Z-12+ZJ-12+NZ-12+D-12)+8.58*(T-15+Z-15+ZJ-15+NZ-15+D-15)+12.32*(T-18+Z-18+ZJ-18+NZ-18 +D-18)}*(PT%*25.05+JT%*30.72+SS%*120.41+YS%*262.33+NSK%*84.17+LSK%*138.90+GSK%*98.81+SK%*55.54+DT%*85.69) Construction cost of earthwork excavation and backfilling for the same pole erection FT earthwork=8.58*(T-15+Z-15+ZJ-15+NZ-15+D-15)*(PT%*25.05+JT%*30.72+SS%*120. 41+YS%*262.33+NSK%*84.17+LSK%*138.90+GSK%*98.81+SK%*55.54+DT%*85.69) Human transport cost, F_human transport = F10 pole human transport + F0.4 pole human transport + FT human transport + F_set human transport + F_other human transport; XY - human transport distance, unit km; 10kV pole tower manual transportation cost F10 pole manual transport = XY (DX-95*408.64+DX-150*438.50+DX-240*692.78+G*827.21+T-12*569.29+Z-12*557.01+NZ-12*562.90+D-12*563.57+T-15*1123.38+Z-15*1111.09+NZ-15*1117.15+D-15*1117.66+T-18*1411.59+Z-18*1399.30+NZ-18*1405.36+D-18*1405.87) 0.4kV pole tower manual transportation cost F0.4 pole manual transport = XY (DX-70 / 95*544.93+DX-120 / 150*584.66+T-12*577.88+Z-12*558.86+ZJ-12*562.90+NZ-12*593.20+D-12*569.46+T-15*1131.80+Z-15*1109.91+ZJ-15*1117.32+NZ-15*1134.66+D-15*1123.55+T-18*1420.01+Z-18*1397.96+ZJ-18*1405.36+NZ-18*1422.70+D-18*1411.09) Labor and transportation costs for pole erection FT manpower transportation = XY (Z*1116.65+NZ*1131.46+NZJ*1179.10+D*1143.92) Equipment installation labor and transportation costs Let F be the number of people transported by manpower = XY(DLQ*23.74+B-100*51.85+B-160*51.85+B-315*51.85+B-400*51.85+LB-50*51.85+LB-100*51.85+LB-160*51.85+LB-315*51.85+LB-400*51.85+BX-D*15.55+BX-S*25.56) Other installation work labor and transportation costs F, the distance of other human transport, = XY (LX-P*59.49+LX-S*494.2+FG*0.08+JD*0.84+XX*1.18) The cost of vehicle transportation is calculated as follows: F_vehicle transportation = F10 pole vehicle transportation + F0.4 pole vehicle transportation + FT vehicle transportation + F_set vehicle transportation + F_other vehicle transportation; QY is the transportation distance in km. 10kV power pole transportation cost F10 pole car transport = QY (DX-95*3.24+DX-150*3.48+DX-240*5.50+G*47.67+T-12*4.12+Z-12*3.98+NZ-12*4.04+D-12*4.05+T-15*6.61+Z-15*6.47+NZ-15*6.53+D-15*6.54+T-18*8.19+Z-18*8.05+NZ-18*8.12+D-18*8.12) 0.4kV power pole transportation cost by truck F0.4 pole truck transportation = QY (DX-70 / 95*4.32+DX-120 / 150*4.64+T-12*4.21+Z-12*4+ZJ-12*4.04+NZ-12*4.39+D-12*4.12+T-15*6.70+Z-15*6.45+ZJ-15*6.54+NZ-15*6.73+D-15*6.61+T-18*8.28+Z-18*8.03+ZJ-18*8.12+NZ-18*8.32+D-18*8.18) Transportation costs for vehicles erected on the same pole FT automobile transportation = QY(Z*6.53+NZ*6.7+NZJ*7.24+D*6.84) Equipment installation and transportation costs Let F be the amount of transportation by car = QY (DLQ*0.27+B-100*0.59+B-160*0.59+B-315*0.59+B-400*0.59+LB-50*0.59+LB-100*0.59+LB-160*0.59+LB-315*0.59+LB-400*0.59+BX-D*0.15+BX-S*0.25) Other installation engineering vehicle transportation costs F (Other vehicle transport distance) = QY (LX-P*0.5+LX-S*3.45+FG*0.01+JD*0.01+XX*0.01) Truck loading and unloading costs, F_truck_loading_unloading = F10-pole truck_loading_unloading + F0.4-pole truck_loading_unloading + FT_truck_loading_unloading + F_set_truck_loading_unloading + F_other_truck_loading_unloading 10kV pole loading and unloading costs F10 pole car loading / unloading = DX-95*97.51+DX-150*104.63+DX-240*165.31+G*568.12+T-12*163.47+Z-12*159.38+NZ-12*161.34+D-12*161.57+T-15*285.78+Z-15*281.69+NZ-15*283.71+D-15*283.88+T-18*359.17+Z-18*355.09+NZ-18*357.10+D-18*357.27 F0.4 pole tower vehicle loading and unloading costs F0.4 pole car loading / unloading = DX-70 / 95*130.03+DX-120 / 150*149.51+T-12*166.33+Z-12*160+ZJ-12*161.34+NZ-12*171.43+D-12*163.53+T-15*288.58+Z-15*281.30+ZJ-15*283.77+NZ-15*289.53+D-15*285.84+T-18*361.97+Z-18*354.64+ZJ-18*357.10+NZ-18*362.87+D-18*359.01 Cost of loading and unloading vehicles on the same pole FT truck loading / unloading = Z*283.54 + NZ*288.47 + NZJ*304.32 + D*292.62 Equipment installation and truck loading / unloading costs Let F be the formula for loading and unloading trucks: DLQ*7.9 + B-100*17.25 + B-160*17.25 + B-315*17.25 + B-400*17.25 + LB-50*17.25 + LB-100*17.25 + LB-160*17.25 + LB-315*17.25 + LB-400*17.25 + BX-D*4.47 + BX-S*7.3 Other installation engineering vehicle loading and unloading costs F Other vehicle loading and unloading = LX-P*14.88+LX-S*144.47+FG*0.03+JD*0.28+XX*0.

39.

7. The application of topographic classification calculation for power distribution network construction cost according to claim 1, characterized in that... It is used for terrain classification and cost estimation.

Images