Electric power material quality sampling inspection sample sealing and sending management system and management method

Abstract

The invention discloses an electric power material quality sampling inspection sample sealing and sending management system and management method, and relates to the technical field of electric power material management, the system comprises the following components: a data acquisition subsystem, a data processing center, and a user interaction and management subsystem; by integrating a plurality of subsystems of data acquisition, processing and user interaction, comprehensive automation and intellectualization of sample sealing and sample sending management of electric power material quality sampling inspection are realized, the system can acquire appearance images of electric power materials and environmental data in the sample sending process in real time, and an image pre-processing algorithm and an environmental data calibration technology are applied to improve the quality of the electric power materials. And through the machine learning analysis module, the system can accurately predict the quality risk of the electric power materials, timely generate early warning information and push the early warning information to related parties, so that the efficiency and accuracy of electric power material quality management are effectively improved.

Description

Technical Field

[0001] The present invention relates to the technical field of power material management, and specifically to a management system and method for sampling, sealing and sample delivery of power material quality inspection. Background Technique

[0002] In the field of power material management, ensuring the quality of power materials is an important link to guarantee the stable operation of the power system and the safety of power supply. With the development of the power industry, the types and quantities of power materials are increasing continuously, which puts forward higher requirements for the sampling, sealing and sample delivery management of power material quality inspection.

[0003] There are deficiencies in traditional technologies. On the one hand, the data collection means are limited, and only some basic information can be obtained, unable to comprehensively reflect the appearance characteristics of power materials and the environmental changes during the sample delivery process. On the other hand, the data processing and analysis capabilities are weak, making it difficult to efficiently process and deeply mine massive data, resulting in inaccurate quality risk prediction and untimely generation of warning information. Therefore, it is particularly important to develop a management system and method for sampling, sealing and sample delivery of power material quality inspection. Summary of the Invention

[0004] The purpose of the present invention is to make up for the deficiencies of the existing technology, and provide a management system and method for sampling, sealing and sample delivery of power material quality inspection. It can realize the full automation and intelligence of power material quality inspection sampling, sealing and sample delivery management by integrating multiple subsystems of data collection, processing and user interaction. The system can collect the appearance image data of power materials and the environmental data during the sample delivery process in real time, and use image preprocessing algorithms and environmental data calibration technologies to ensure the accuracy and integrity of the data. At the same time, by introducing a machine learning analysis module, the system can deeply mine and analyze multi-source data sets, accurately predict the quality risks of power materials, and timely generate warning information and push it to relevant parties.

[0005] To solve the above technical problems, the present invention provides the following technical solutions: A management system and method for sampling, sealing and sample delivery of power material quality inspection, the system includes the following components: a data collection subsystem, a data processing center, and a user interaction and management subsystem;

[0006] The data collection subsystem: is used to collect the appearance image data of power materials during sealing and the environmental data during the sample delivery process. The data collection subsystem includes an appearance image collection unit, an environmental data collection unit and a data transmission module. The appearance image collection unit uses a high-resolution industrial camera and intelligent image collection software to obtain the appearance image data of power materials, and preprocesses the collected images. The algorithm formula is:

[0007] I enhanced =I original×(1 + k a ×∑ i＝-n ∑ i＝-m G(i, j)×ΔI(i, j)), where I enhanced is the pixel value of the enhanced image, I original is the pixel value of the original image, G(i, j) is the Gaussian convolution kernel, ΔI(i, j) is the gradient value of the image at the position (i, j), k a is the weight coefficient. The environmental data acquisition unit consists of a high-precision temperature and humidity sensor, a vibration sensor, and a light sensor to collect transportation environmental data, and the data acquisition frequency of each sensor is determined according to the environmental sensitivity analysis of power materials. The data transmission module transmits the collected data to the data processing center, using a combination of wireless transmission technology and a wired transmission network, and encrypts the transmitted data using an encryption algorithm based on the principle of quantum key distribution;

[0008] The data processing center: includes a data repository, a machine learning analysis module, and an early warning generation and management module. The data repository stores the power material quality standard knowledge base, historical test result data, and the sealed sample and sample data of this sampling inspection, using distributed storage technology and having a data redundancy backup and efficient retrieval mechanism. Among them, the data redundancy backup strategy is set based on the classification of data importance. The machine learning analysis module uses a hybrid neural network algorithm to analyze multi-source data sets to predict the quality risk of power materials. The algorithm formula is as follows: Output = σ(w f ×f LSTM (Input CNN (Data)) + W b ×b LSTM (Input CNN (Data)) + b o ), where Output is the final predicted output, Input CNN (Data) represents the result after first inputting the data into a convolutional neural network for feature extraction. f LSTM and b LSTM are the forward and backward long short-term memory network processing functions respectively, W f , W b are the corresponding weight matrices. During the training process, different initial weight setting ranges are used for different types of power material data. b o is the bias term, σ is the activation function, and the rectified linear unit function is used. The early warning generation and management module generates early warning information according to the prediction results and pushes it. The early warning information includes the detailed information of power materials, the type of quality risk, the risk level, and the handling suggestions, and records the sending and receiving status of the early warning information;

[0009] The user interaction and management subsystem is provided with a supplier management terminal and a power enterprise management terminal. The supplier management terminal enables suppliers to view the progress of sample sealing and delivery, receive warning information, upload materials, and provide feedback on countermeasures. The power enterprise management terminal enables power enterprise managers to monitor data, review countermeasures, issue instructions, and generate reports.

[0010] Furthermore, the size n×m of the Gaussian convolution kernel G(i, j) in the image feature enhancement algorithm of the appearance image acquisition unit is determined according to the average size of the power material image and the scale of the features to be extracted. For small power material accessory images, a 3×3 Gaussian convolution kernel is usually used to finely capture local features. For large power equipment images, a 5×5 Gaussian convolution kernel is used to highlight texture features in a larger range without losing overall contour information. When calculating the image gradient ΔI(i, j), the Sobel operator and the Prewitt operator are combined. First, the gradient values under the two operators are calculated separately, and then their weighted average is taken as the final gradient value. The weighting coefficients are determined based on experimental evaluations of the edge detection accuracy of a large number of power material images. The weighting coefficient k of the Sobel operator sobel = 0.6, and the weighting coefficient k of the Prewitt operator Prewitt = 0.4. This combination method can more accurately detect edge and texture changes in power material images, providing a more reliable basis for image features for subsequent quality risk analysis.

[0011] Even further, when the environmental data acquisition unit collects data, in addition to recording the measurement values of the sensors, it also collects the working state data of the sensors themselves, including the power, calibration time, and temperature drift compensation value information of the sensors. These working state data and the environmental measurement data are transmitted to the data processing center together. When analyzing the environmental data in the data processing center, the environmental measurement data is first corrected and calibrated according to the sensor working state data. When the temperature drift compensation value of the temperature and humidity sensor exceeds the preset range, a calibration method based on polynomial fitting is used to correct the temperature and humidity measurement values. The correction formula is: H corrected = H measured + k b ×(T drift - T ref )×H measured , where H corrected is the corrected humidity value, H measured is the measured humidity value, T drift is the temperature drift value, and T refis the reference temperature value, and kb is the calibration coefficient, which is determined through calibration experiments on the sensor at different ambient temperatures and has a value range of 0.001 - 0.005. This can effectively improve the accuracy of environmental data and avoid the impact of environmental data errors caused by changes in the sensor's own state on the quality risk prediction of power materials.

[0012] Furthermore, in the encryption algorithm based on the principle of quantum key distribution of the data transmission module, the generation rate of quantum keys is dynamically adjusted according to the data transmission volume and the requirements of transmission real-time performance. When the data transmission volume is large and the requirement for transmission delay is low, a quantum key distribution scheme using multi-photon entangled states is adopted to increase the key generation rate. Its key generation rate R key is related to the number N entanglement of photon entangled states photon and the photon emission frequency f key and the formula is: R key = k c × N entanglement × f photon , where k c is the efficiency coefficient. When the data transmission volume is small but the requirement for transmission delay is high, a single-photon quantum key distribution scheme is adopted, and the transmission path and detection efficiency of photons are optimized to ensure the high efficiency and security of data encrypted transmission and meet the data transmission requirements in different sample sealing and sample sending scenarios.

[0013] Furthermore, in the distributed storage architecture of the data repository, the distribution strategy of data nodes is determined based on the geographical sources and usage area distributions of power materials. For data from concentrated production areas of power materials, it is stored in the data node cluster close to that area to reduce data transmission delay. For power material data with a high usage frequency in a specific power project, its replicas are stored on the data nodes in the area where the project is located to improve data reading speed. At the same time, in the data retrieval mechanism, a retrieval method based on dual indexing of content and geographical location is adopted. First, a preliminary screening is carried out according to the type and batch content information of power materials, and then a secondary screening is carried out in combination with the geographical location information of data storage nodes to quickly locate the required data and improve data retrieval efficiency, ensuring the efficient operation of the system when processing a large amount of power material sampling inspection data.

[0014] Furthermore, during the training process of the hybrid neural network algorithm of the machine learning analysis module, a dynamic learning rate adjustment strategy is adopted. The learning rate ηη gradually decreases as the number of training rounds t increases, and the formula is where η 0 is the initial learning rate, T is the total number of training rounds, kd is the decay coefficient, and after each round of training ends, according to the loss value L validation of the model on the validation set and the loss value L previousBased on the comparison result, an additional adjustment is made to the learning rate. If L validation > L preorious ×(1 + k e ), where k e is the tolerance coefficient, then the learning rate is multiplied by a contraction factor k f less than 1. This dynamic learning rate adjustment strategy enables the model to converge to a better solution faster during training and improves the accuracy of power material quality risk prediction.

[0015] Furthermore, when generating warning information, the warning generation and management module determines the quality risk type based on the matching result between the intermediate layer feature vector output by the hybrid neural network algorithm and the power material quality risk feature library. This quality risk feature library is constructed through in-depth analysis and feature extraction of a large number of historical power material quality problem cases. The hierarchical clustering algorithm is used to classify quality risks into different categories, and a feature vector template is established for each category. During the matching process, the similarity between the intermediate layer feature vector output by the hybrid neural network algorithm and each feature vector template is calculated using the cosine similarity calculation formula:

[0016] where V output is the intermediate layer feature vector output by the hybrid neural network algorithm, V template is the feature vector template in the power material quality risk feature library, n is the dimension of the feature vector. When the similarity exceeds the preset threshold k g , it is determined as the corresponding quality risk type, thereby improving the accuracy and pertinence of the warning information, enabling power enterprises and suppliers to take more precise countermeasures.

[0017] Furthermore, the information interaction between the supplier management terminal and the power enterprise management terminal adopts a secure multi-party computation protocol to ensure the privacy and security of both parties when communicating sensitive information such as supplier countermeasures and power enterprise review opinions. In the secure multi-party computation protocol, the encryption and decryption operations of data are based on the elliptic curve cryptosystem. The encryption key and the decryption key are respectively generated by the supplier and the power enterprise through their respective private key and public key generation algorithms, and the public key information is exchanged and registered during system initialization. During the information interaction process, data is transmitted and calculated in ciphertext form, and the original information can only be obtained after the receiving party decrypts it using the corresponding private key, preventing the information from being stolen or tampered with by a third party during the transmission process and ensuring the information security and business privacy in the process of power material quality sampling, sealing, and sample delivery management.

[0018] On the other hand, a method for power material quality sampling, sealing, and sample delivery management is characterized in that the specific steps of the method are as follows:

[0019] S1. Data collection: used to collect the appearance image data during the sealing of power materials and the environmental data during the sample delivery process. The data collection subsystem includes an appearance image acquisition unit, an environmental data acquisition unit, and a data transmission module. The appearance image acquisition unit uses a high-resolution industrial camera and intelligent image acquisition software to obtain the appearance image data of power materials, and preprocesses the collected images. The algorithm formula is:

[0020] I enhanced =I original ×(1 - k a ×∑ i＝-n ∑ i＝-m G(i, j)×ΔI(i, j)), where I enhanced is the pixel value of the enhanced image, I original is the pixel value of the original image, G(i, j) is the Gaussian convolution kernel, ΔI(i, j) is the gradient value of the image at the (i, j) position, k a is the weight coefficient. The environmental data acquisition unit consists of a high-precision temperature and humidity sensor, a vibration sensor, and a light sensor to collect the transportation environment data. The data acquisition frequency of each sensor is determined based on the environmental sensitivity analysis of power materials. The data transmission module transmits the collected data to the data processing center, using a combination of wireless transmission technology and wired transmission network, and encrypts the transmitted data using an encryption algorithm based on the principle of quantum key distribution;

[0021] S2. Data processing: includes a data repository, a machine learning analysis module, and an early warning generation and management module. The data repository stores the power material quality standard knowledge base, historical detection result data, and the sealed sample delivery data of this sampling inspection. It uses distributed storage technology and has a data redundancy backup and efficient retrieval mechanism. Among them, the data redundancy backup strategy is set based on data importance classification. The machine learning analysis module uses a hybrid neural network algorithm to analyze multi-source data sets to predict the quality risk of power materials. The algorithm formula is as follows: Output = σ(W f ×f LSTM (Input CNN (Data)) + W b ×b LSTM (Input CNN (Data)) + b o ), where Output is the final predicted output, Input CNN (Data) represents the result after first inputting the data into a convolutional neural network for feature extraction. f LSTM and b LSTM are the forward and backward long short-term memory network processing functions respectively, and W f 、W bis the corresponding weight matrix. During the training process, different initial weight setting ranges are adopted for different types of power material data, and b o is the bias term, σ is the activation function, and the rectified linear unit function is adopted. The early warning generation and management module generates early warning information according to the prediction results and pushes it. The early warning information includes detailed power material information, quality risk types, risk levels, and handling suggestions, and records the sending and receiving status of the early warning information;

[0022] S3. User interaction and management: There are a supplier management terminal and a power enterprise management terminal. The supplier management terminal allows suppliers to view the progress of sample sealing and delivery, receive early warning information, upload materials, and feedback response measures. The power enterprise management terminal allows power enterprise managers to monitor data, review response measures, issue instructions, and generate reports.

[0023] Compared with the prior art, the power material quality sampling, sample sealing, and delivery management system and management method have the following beneficial effects:

[0024] First, by integrating multiple subsystems of data collection, processing, and user interaction, the present invention realizes the full automation and intelligence of power material quality sampling, sample sealing, and delivery management. The system can collect the appearance images of power materials and environmental data during the sample delivery process in real time, and use image preprocessing algorithms and environmental data calibration technologies to ensure the accuracy and integrity of the data. Through the machine learning analysis module, the system can accurately predict the quality risks of power materials, generate early warning information in a timely manner, and push it to relevant parties, effectively improving the efficiency and accuracy of power material quality management.

[0025] Second, the present invention encrypts the transmitted data by using an encryption algorithm based on the principle of quantum key distribution to ensure the security and integrity of data transmission. At the same time, in the information interaction between the supplier management terminal and the power enterprise management terminal, a secure multi-party computing protocol is adopted to further ensure the privacy and security of sensitive information. The implementation of these security measures effectively prevents the risk of information being stolen or tampered with by a third party during the transmission process, providing strong protection for information security and business privacy in the process of power material quality sampling, sample sealing, and delivery management.

[0026] Other advantages, objectives, and features of the present invention will be described to some extent in the subsequent specification, and to some extent, will be obvious to those skilled in the art based on the study of the following text, or can be taught from the practice of the present invention. Brief Description of the Drawings

[0027] Figure 1 is the flow operation diagram of the management system of the present invention;

[0028] Figure 2This is the flowchart of the management method of the present invention. Detailed implementation manners

[0029] To further elaborate on the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the following will, in conjunction with the accompanying drawings and preferred embodiments, describe in detail the specific implementation manners, structures, features and their effects of the present invention as follows.

[0030] Embodiment 1

[0031] This embodiment describes a certain power equipment manufacturing enterprise. To ensure the quality of the small power material accessories produced, sampling inspections are carried out before leaving the factory. These accessories will be supplied to power engineering projects in multiple regions, with relatively high requirements for appearance quality, and may be affected by certain vibrations, temperature and humidity changes during transportation.

[0032] Use a high-resolution industrial camera to take pictures of the power connection terminals. Since the accessories are small in size, a 3×3 Gaussian convolution kernel is used to preprocess the collected images. I enhanced = I original ×(1 + k a ×∑ i＝-n ∑ i＝-m G(i, j)×ΔI(i, j)), where I enhanced is the pixel value of the enhanced image, I original is the pixel value of the original image, G(i, j) is the Gaussian convolution kernel, ΔI(i, j) is the gradient value of the image at the (i, j) position, and k a is the weight coefficient to enhance the image features. When calculating the image gradient, by combining the Sobel operator (weighting coefficient 0.6) and the Prewitt operator (weighting coefficient 0.4), the subtle defect features on the surface of the accessories can be accurately extracted. For example, scratches and deformations on the surface of the connection terminals can be clearly presented.

[0033] High-precision temperature and humidity sensors, vibration sensors and light sensors collect transportation environment data in real time. The temperature and humidity sensors collect data every 10 minutes. The vibration sensors continuously monitor the vibration situation during transportation. The light sensors collect data in a timely manner according to the change of ambient light. The sensors also record their own power, calibration time and working status data. When the temperature drift compensation value of the temperature and humidity sensors exceeds the preset range, a calibration method based on polynomial fitting is used to correct the humidity measurement value.

[0034] The collected appearance images and environmental data are transmitted to the data processing center by combining wireless transmission technology (such as 4G network) with wired transmission network (enterprise internal local area network). During the data transmission process, a single-photon quantum key distribution scheme is used for encryption to ensure data security. Since the data volume of small accessories is relatively small but the requirement for transmission delay is high, optimizing the photon transmission path and detection efficiency can meet the real-time demand.

[0035] Store the quality standard knowledge base of power connection terminals (such as the dimensions, materials, and appearance requirements specified by industry standards), historical detection result data, and the data of the sealed samples and submitted samples for this random inspection. Adopt distributed storage technology. Based on the geographical sources and usage area distributions of power materials, store the data from the enterprise's own production in the local data node cluster, and store copies of the data of this accessory with a high usage frequency in specific power projects on the data nodes in the project area. When retrieving data, first screen according to the accessory type and batch content information, and then quickly locate the required data in combination with the geographical location information of the data storage nodes.

[0036] Use a hybrid neural network algorithm to analyze the data, Output = σ(W f ×f LSTM (Input CNN (Data)) + W b ×b LSTM (Input CNN (Data)) + b o ), where Output is the final predicted output, Input CNN (Data) represents the result after first inputting the data into a convolutional neural network for feature extraction, f LSTM and b LSTM are the forward and backward long short-term memory network processing functions respectively, W f 、W b are the corresponding weight matrices. During the training process, different initial weight setting ranges are used for different types of power material data. b o is the bias term, and σ is the activation function. During the training process, set the initial weight range for power connection terminal data and adopt a dynamic learning rate adjustment strategy. The learning rate gradually decreases as the number of training rounds increases. After each round of training, the learning rate is adjusted additionally according to the comparison result of the loss value of the model on the validation set and the loss value of the previous round. The algorithm predicts the quality risk of power connection terminals through comprehensive analysis of appearance image data and environmental data.

[0037] Determine the quality risk type according to the matching result between the intermediate layer feature vector output by the hybrid neural network algorithm and the quality risk feature library. The quality risk feature library is constructed by analyzing a large number of historical quality problem cases of similar accessories, and is classified by the hierarchical clustering algorithm and a feature vector template is established. For example, if the similarity between the feature vector output by the algorithm and the risk feature vector template of the scratch type exceeds the preset threshold, it is determined as the appearance scratch quality risk, and a warning message is generated, including the detailed information of the wiring terminal, the risk type (appearance scratch), the risk level (classified according to the severity of the scratch), and the handling suggestion (such as repair, scrapping), and record the sending and receiving status of the warning message.

[0038] Suppliers can view the sample submission progress of the power wiring terminals in real time and receive warning messages in a timely manner. If they receive a warning message about the appearance scratch risk, the suppliers upload relevant materials, such as production process records and quality inspection reports, and feedback the countermeasures, such as improving the production process and strengthening the quality inspection link.

[0039] Managers of power enterprises monitor the data, review the countermeasures feedback by the suppliers, and issue instructions. For example, they require the suppliers to suspend the shipment of this batch of products and wait for further processing. Finally, a sampling inspection report is generated, recording the entire sampling inspection process and results, providing a basis for subsequent procurement decisions.

[0040] Embodiment 2

[0041] This embodiment describes that a power company purchases a batch of large transformers from a large power equipment manufacturer in other places and needs to conduct quality sampling inspections on them. The transformers are large in volume and high in value. Environmental factors during transportation (such as strong vibrations, temperature and humidity changes, and light) may affect their performance, and more comprehensive and accurate image data is required for their appearance quality inspection.

[0042] Use a high-resolution industrial camera and intelligent image acquisition software to obtain the appearance image data of the transformer. Due to the large volume of the equipment, a 5×5 Gaussian convolution kernel is used to enhance the image features to ensure that all details of the equipment appearance are captured. When calculating the image gradient, the Sobel operator and the Prewitt operator are also combined to accurately detect whether there are depressions on the equipment shell and whether the paint surface has peeled off.

[0043] High-precision temperature and humidity sensors, vibration sensors, and light sensors work together to collect transportation environment data. Considering that the transformer is relatively sensitive to environmental changes, the temperature and humidity sensors collect data every 5 minutes, the vibration sensors monitor the vibration conditions during transportation in real time, and the light sensors collect data at key nodes during equipment loading, unloading, and transportation. The sensors record their own working status data.

[0044] The collected data is transmitted to the data processing center through the combination of wireless transmission technology (such as 5G network) and wired transmission network. Since the transformer data volume is large and the requirement for transmission delay is relatively low, a quantum key distribution scheme based on multi-photon entangled states is adopted to encrypt and transmit the data. The key generation rate is dynamically adjusted according to the data transmission volume to ensure the security and integrity of data transmission.

[0045] Store the transformer quality standard knowledge base (such as the requirements of national standards for transformer performance, structure, and appearance), historical test result data, and the current sampling inspection data. Adopt distributed storage technology. According to the geographical origin and usage area distribution of power materials, store the data from the manufacturer's area in the corresponding regional data node cluster, and at the same time store copies at the data nodes in the area where the power engineering project using the transformer is located. Data retrieval adopts a method based on dual indexing of content and geographical location to quickly locate the required transformer data.

[0046] When training the hybrid neural network algorithm, a specific initial weight range is set for transformer data. The dynamic learning rate adjustment strategy is used, combined with the rectified linear unit activation function, to deeply analyze the appearance image data and environmental data of the transformer. Output = σ(W f ×f LSTM (Input CNN (Data)) + W b ×b LSTM (Input CNN (Data)) + b o ), where Output is the final predicted output, Input CNN (Data) represents the result after first inputting the data into a convolutional neural network for feature extraction. f LSTM and b LSTM are the forward and backward long short-term memory network processing functions respectively, W f 、W b are the corresponding weight matrices. During the training process, different initial weight setting ranges are adopted for different types of power material data. b o is the bias term, σ is the activation function, and predict its quality risk. For example, by analyzing the correlation between the vibration data during transportation and the change in the appearance image, predict whether the internal components may be displaced or damaged.

[0047] The risk type is determined based on the matching between the intermediate layer feature vector output by the hybrid neural network algorithm and the quality risk feature library. The quality risk feature library is constructed from a large number of historical transformer quality problem cases, classified using the hierarchical clustering algorithm, and a feature vector template is established. If the algorithm determines that the transformer has a risk of internal component displacement, a warning message is generated, including the detailed information of the transformer, the risk type (internal component displacement), the risk level (classified according to the displacement degree and possible impacts), and the handling suggestions (such as returning to the factory for maintenance, on-site commissioning), and the status of the warning message is recorded.

[0048] The supplier checks the progress of the transformer sealing and sample submission, uploads the detailed information during the equipment production process, such as the raw material inspection report and the assembly process record, after receiving the warning message, and feedbacks the countermeasures, such as arranging technical personnel to assist in the on-site inspection.

[0049] The power enterprise management personnel monitor the whole process data of the spot check, review the supplier's countermeasures, and issue instructions. For example, according to the risk level, they decide whether to allow the equipment to continue transporting to the designated location, or require the supplier to take special protection measures during transportation, and finally generate a detailed spot check report to provide a reference for equipment acceptance and subsequent operation and maintenance.

[0050] The above are only the preferred embodiments of the present invention, and do not impose any form of limitation on the present invention. Although the present invention has been disclosed above with the preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some changes or modifications to the above-disclosed technical content to form equivalent embodiments with equivalent changes, but as long as it does not depart from the technical content of the present invention, any brief modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention still fall within the scope of the technical solution of the present invention.

Claims

1. A power material quality sampling and sealing sample delivery management system, characterized in that: The system includes data acquisition subsystem, data processing center, user interaction and management subsystem: The data acquisition subsystem is used to collect the appearance image data of power materials when they are sealed and the environmental data during the sample delivery process. The data acquisition subsystem includes an appearance image acquisition unit, an environmental data acquisition unit and a data transmission module. The appearance image acquisition unit uses a high-resolution industrial camera and intelligent image acquisition software to obtain the appearance image data of power materials and pre-process the collected images. The algorithm formula is: I enhanced =I origianl ×(1+k a ×Σ i =-nΣ i =-m G(i,j)×ΔI(i,j)), where I enhanced is the pixel value of the enhanced image, I original is the original image pixel value, G(i, j) is the Gaussian convolution kernel, ΔI(i, j) is the gradient value of the image at position (i, j), k a is a weight coefficient. The environmental data acquisition unit is composed of a high-precision temperature and humidity sensor, a vibration sensor, and a light sensor to collect transportation environmental data. The data acquisition frequency of each sensor is determined according to the environmental sensitivity analysis of the power materials. The data transmission module transmits the collected data to the data processing center, using a combination of wireless transmission technology and wired transmission network, and uses an encryption algorithm based on the principle of quantum key distribution to encrypt the transmission data. The data processing center includes a data repository, a machine learning analysis module, and an early warning generation and management module. The data repository stores the knowledge base of power material quality standards, historical test result data, and the sealed sample delivery data of this sampling inspection. It adopts distributed storage technology and has data redundancy backup and efficient retrieval mechanism. The data redundancy backup strategy is set based on the data importance classification. The machine learning analysis module uses a hybrid neural network algorithm to analyze multi-source data sets to predict the quality risks of power materials. The algorithm formula is as follows: Output=σ(W f ×f LSTM (Input CNN (Data))+W b ×b LSTM (Input CNN (Data))+b o ), where Output is the final prediction output, Input CNN (Data) represents the result after inputting data into the convolutional neural network for feature extraction. LSTM and b LSTM are the forward and reverse LSTM network processing functions, W f , W b is the corresponding weight matrix. During the training process, different initial weight setting ranges are used for different types of power material data. o is a bias term, σ is an activation function, and a modified linear unit function is used. The warning generation and management module generates and pushes warning information according to the prediction results. The warning information includes detailed information of power materials, quality risk type, risk level and processing suggestions, and records the sending and receiving status of the warning information; The user interaction and management subsystem is provided with a supplier management terminal and a power enterprise management terminal. The supplier management terminal allows suppliers to check the progress of sample sealing and delivery, receive early warning information, upload materials and feedback response measures. The power enterprise management terminal enables power enterprise managers to monitor data, review response measures, issue instructions and generate reports.

2. According to claim 1, a power material quality sampling and sampling management system is characterized in that: The size n×m of the Gaussian convolution kernel G(i, j) in the image feature enhancement algorithm of the appearance image acquisition unit is determined according to the average size of the power material image and the scale of the required feature extraction. For the image of small power material accessories, a 3×3 Gaussian convolution kernel is usually used, while for the image of large power equipment, a 5×5 Gaussian convolution kernel is used. When calculating the image gradient ΔI(i, j), a combination of the Sobel operator and the Prewitt operator is used. The gradient values ​​under the two operators are first calculated respectively, and then the weighted average value is taken as the final gradient value. The weighting coefficient is determined according to the experimental evaluation of the accuracy of edge detection of a large number of power material images. The weighting coefficient k of the Sobe operator is Sobel =0.6, the weighting coefficient k of the Prewitt operator Prewitt =0.

4.

3. The power material quality sampling and sampling management system according to claim 1 is characterized in that: During data collection, the environmental data collection unit, in addition to recording the measured values ​​of the sensor, also collects the working status data of the sensor itself, including the power of the sensor, calibration time, and temperature drift compensation value information. These working status data are transmitted to the data processing center together with the environmental measurement data. When the data processing center analyzes the environmental data, the environmental measurement data is first corrected and calibrated according to the sensor working status data. When the temperature drift compensation value of the temperature and humidity sensor exceeds the preset range, the calibration method based on polynomial fitting is used to correct the temperature and humidity measurement values. The correction formula is: H corrected =H measured +k b ×(T drift -T ref )×H measured , where H corrected is the corrected humidity value, H measured is the measured humidity value, T drift is the temperature drift value, T ref is the reference temperature value, k b is the calibration factor.

4. The power material quality sampling and sealing and delivery management system according to claim 1 is characterized in that: In the encryption algorithm based on the quantum key distribution principle, the data transmission module dynamically adjusts the quantum key generation rate according to the data transmission volume and the transmission real-time requirements. When the data transmission volume is large and the transmission delay requirement is low, a multi-photon entangled state quantum key distribution scheme is adopted, and the key generation rate R key The number of entangled states with photons N entanglement and the photon emission frequency f photon Related, the formula is: R key =k c ×N entanglement ×f photon , where k c For the efficiency coefficient, when the data transmission volume is small but the transmission delay requirement is high, a single-photon quantum key distribution scheme is adopted, and the transmission path and detection efficiency of the photon are optimized.

5. The power material quality sampling and sampling management system according to claim 1 is characterized in that: In the distributed storage architecture of the data repository, the distribution strategy of data nodes is determined based on the geographical origin and usage area distribution of electric power materials. For data from areas where electric power material production is concentrated, they are stored in a data node cluster close to the area. For electric power material data that is frequently used in a specific electric power project, copies of them are stored on data nodes in the area where the project is located. At the same time, in the data retrieval mechanism, a retrieval method based on dual indexing of content and geographic location is adopted. First, a preliminary screening is performed based on the type of electric power materials and batch content information, and then a secondary screening is performed in combination with the geographic location information of the data storage node to quickly locate the required data.

6. The power material quality sampling and sealing and delivery management system according to claim 1 is characterized in that: The hybrid neural network algorithm of the machine learning analysis module adopts a dynamic learning rate adjustment strategy during the training process. The learning rate η gradually decreases with the increase of the number of training rounds t. The formula is: Where η0 is the initial learning rate, T is the total number of training rounds, and k d is the attenuation coefficient, and after each round of training, the loss value L of the model on the validation set is validation and the loss value L of the previous round previous The comparison results of the learning rate are adjusted additionally. If L validation >L preorious ×(1+k e ), where k e is the tolerance coefficient, then multiply the learning rate by a shrinkage factor k less than 1 f .

7. The power material quality sampling and sealing and delivery management system according to claim 1 is characterized in that: When the early warning generation and management module generates early warning information, the quality risk type is determined based on the matching result of the intermediate layer feature vector output by the hybrid neural network algorithm and the power material quality risk feature library. The quality risk feature library is constructed by in-depth analysis and feature extraction of a large number of historical power material quality problem cases, and a hierarchical clustering algorithm is used to divide quality risks into different categories, and a feature vector template is established for each category. In the matching process, the similarity between the intermediate layer feature vector output by the hybrid neural network algorithm and each feature vector template is calculated, and the cosine similarity calculation formula is used: Where V output is the intermediate layer feature vector output by the hybrid neural network algorithm, V template is the feature vector template in the quality risk feature library, n is the dimension of the feature vector, and when the similarity exceeds the preset threshold k g , it is determined to be the corresponding quality risk type.

8. The power material quality sampling and sealing and delivery management system according to claim 1 is characterized in that: The information interaction between the supplier management terminal and the power enterprise management terminal adopts a secure multi-party computing protocol. In the secure multi-party computing protocol, data encryption and decryption operations are based on the elliptic curve cryptography system. The encryption key and decryption key are generated by the supplier and the power enterprise respectively through their respective private key and public key generation algorithms, and the public key information is exchanged and registered when the system is initialized. During the information interaction process, the data is transmitted and calculated in ciphertext form. The original information can only be obtained after the recipient uses the corresponding private key to decrypt it, preventing the information from being stolen or tampered with by a third party during the transmission process, thereby ensuring the information security and commercial privacy in the process of power material quality random sampling sealing and sampling management.

9. A method for managing the sealing and delivery of samples for random inspection of power material quality, characterized in that: The specific steps of this method are: S1. Data acquisition: used to collect the appearance image data of power materials when they are sealed and the environmental data during the sample delivery process. The data acquisition subsystem includes an appearance image acquisition unit, an environmental data acquisition unit and a data transmission module. The appearance image acquisition unit uses a high-resolution industrial camera and intelligent image acquisition software to obtain the appearance image data of power materials and pre-process the collected images. The algorithm formula is: I enhanced =I original ×(1+ka×∑ i＝-n ∑ i = -mG(i, j) × ΔI(i, j)), where I enhanced is the pixel value of the enhanced image, I original is the original image pixel value, G(i, j) is the Gaussian convolution kernel, ΔI(i, j) is the gradient value of the image at position (i, j), k a is a weight coefficient. The environmental data acquisition unit is composed of a high-precision temperature and humidity sensor, a vibration sensor, and a light sensor to collect transportation environmental data. The data acquisition frequency of each sensor is determined according to the environmental sensitivity analysis of the power materials. The data transmission module transmits the collected data to the data processing center, using a combination of wireless transmission technology and wired transmission network, and uses an encryption algorithm based on the principle of quantum key distribution to encrypt the transmission data. S2. Data processing: including data repository, machine learning analysis module and early warning generation and management module. The data repository stores the knowledge base of power material quality standards, historical test result data and the sealed sample delivery data of this sampling inspection. It adopts distributed storage technology and has data redundancy backup and efficient retrieval mechanism. The data redundancy backup strategy is based on the classification of data importance. The machine learning analysis module uses a hybrid neural network algorithm to analyze multi-source data sets to predict the quality risks of power materials. The algorithm formula is as follows: Output=σ(W f ×f LSTM (Input CNN (Data))+W b ×b LSTM (Input CNN (Data))+b o ), where OutPut is the final prediction output, Input CNN (Data) represents the result after inputting data into the convolutional neural network for feature extraction. LSTM and b LSTM are the forward and reverse LSTM network processing functions, W f , W b is the corresponding weight matrix. During the training process, different initial weight setting ranges are used for different types of power material data. o is a bias term, σ is an activation function, and a modified linear unit function is used. The warning generation and management module generates and pushes warning information according to the prediction results. The warning information includes detailed information of power materials, quality risk type, risk level and processing suggestions, and records the sending and receiving status of the warning information; S3. User interaction and management: There are supplier management terminals and power enterprise management terminals. The supplier management terminal allows suppliers to check the progress of sample sealing and delivery, receive early warning information, upload materials and feedback response measures. The power enterprise management terminal allows power enterprise managers to monitor data, review response measures, issue instructions and generate reports.

Images