Edge AI processing and block chain evidence storage method and device for power distribution construction, and medium

Abstract

The invention discloses an edge AI processing and block chain evidence storage method and device for power distribution construction, and a medium. The method comprises the following steps: S1, initializing to generate a key; s2, collecting construction site data; s3, carrying out edge AI localization analysis to generate a result; s4, performing hierarchical encryption; s5, block chain evidence storage; s6, authorizing access; according to the method, the lightweight AI model is deployed on the edge side, so that the potential safety hazard identification response time is shortened, and the target identification accuracy is improved; decryptable access control is realized by only authorizing a node through an encryption cooperation mechanism, and evidence storage data has non-tampering performance and traceability through a block chain evidence storage and fragment storage strategy; the cloud data transmission quantity is reduced through edge AI local processing; an encryption proxy server and a decryption proxy server in the layered encryption mechanism undertake complex operation; based on secret key distribution of an attribute set and access strategy tree control, it is ensured that sensitive data is only limited to access by a construction unit administrator and an authorization party of a supervision unit.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of power engineering safety monitoring and information technology, in particular to an edge AI processing and blockchain storage method for power distribution construction, equipment and medium. BACKGROUND

[0002] The current power distribution construction site safety monitoring mainly relies on the centralized cloud processing mode, which has certain shortcomings:

[0003] Firstly, when using a camera and a central server architecture, the raw data needs to be returned to the cloud for analysis through the network. In complex construction environments, such as remote areas or network fluctuation scenarios, the data transmission delay significantly increases, leading to the inability to timely identify potential hazards such as abnormal safety helmet wearing and equipment failure, which seriously affects the timeliness of early warning.

[0004] Secondly, the existing scheme stores the processing results in a centralized database, which has a single point tampering risk and lacks a full-process operation traceability mechanism. At the same time, the centralized storage mode in the cloud faces a storage bottleneck and is difficult to support the long-term storage needs of large-scale monitoring data.

[0005] Finally, when deploying an artificial intelligence target detection model in the cloud, massive data transmission exacerbates network bandwidth burden. Direct application of attribute-based encryption and other privacy protection technologies will excessively consume terminal computing power, while simplified encryption schemes cannot meet the access control requirements of construction supervision parties such as safety supervision departments and supervision units, leading to the difficulty in balancing privacy protection and system efficiency.

[0006] Therefore, it is necessary to design an edge AI processing and blockchain storage method for power distribution construction, equipment and medium. SUMMARY

[0007] The purpose of the present application is to provide an edge AI processing and blockchain storage method for power distribution construction, equipment and medium to solve the problems mentioned in the background art, such as the use of a camera and a central server architecture, the raw data needing to be returned to the cloud for analysis through the network, the significant increase in data transmission delay in complex construction environments such as remote areas or network fluctuation scenarios, the inability to timely identify potential hazards such as abnormal safety helmet wearing and equipment failure, which seriously affects the timeliness of early warning, and the storage of processing results in a centralized database, which has a single point tampering risk and lacks a full-process operation traceability mechanism. At the same time, the centralized storage mode in the cloud faces a storage bottleneck and is difficult to support the long-term storage needs of large-scale monitoring data. When deploying an artificial intelligence target detection model in the cloud, massive data transmission exacerbates network bandwidth burden. Direct application of attribute-based encryption and other privacy protection technologies will excessively consume terminal computing power, while simplified encryption schemes cannot meet the access control requirements of construction supervision parties such as safety supervision departments and supervision units, leading to the difficulty in balancing privacy protection and system efficiency.

[0008] To achieve the above objectives, the present invention provides the following technical solution:

[0009] The first aspect provides an edge AI processing and blockchain evidence storage method for power distribution construction, including the following steps:

[0010] S1: System initialization, the power regulatory department to which the power distribution construction project belongs or its authorized third-party technical certification body acts as the trusted authority (TA). The trusted authority (TA) generates the system public key (PK), master key (MK), and private key (SK) through the CP-ABE key generation algorithm, which are used for subsequent decryption operations.

[0011] S2: Real-time acquisition of multi-source heterogeneous data from the power distribution construction site through the data acquisition module, including image data and physical quantity data;

[0012] S3: The edge AI processing module performs local real-time analysis on the collected multi-source heterogeneous data to generate security monitoring results. Specifically, a lightweight YOLOv8-nano model is deployed on the edge server to identify image data anomalies, and an improved LSTM model is combined to analyze sensor time-series data.

[0013] S4: Via encrypted proxy server PS enc A layered encryption mechanism is used to process security monitoring results and generate encrypted results.

[0014] S5: Write the encrypted result into the blockchain evidence storage module, use a sharding storage strategy for distributed evidence storage, and generate evidence storage data;

[0015] S6: Authorized Supervisory Node (DR) decrypts the proxy server PS dec Verify and access the stored evidence data using the key generated by S1.

[0016] As a further technical solution of the present invention, S1 specifically includes the following steps:

[0017] S1.1: The trusted authority TA generates the parameters of the bilinear group system; specifically, the trusted authority TA inputs the entire attribute set U = {Attr1, Attr2, ..., Attr...} n The power distribution scenario attributes include construction unit administrators, supervision units, safety supervision departments, and system maintenance personnel, and the CP-ABE key generation algorithm is executed.

[0018] S1.2: Generate the system public key PK and master key MK based on the attribute set; specifically: select bilinear groups G1 and G2, with order p as a prime number, bilinear mapping G1×G1→G2, and generator g∈G1;

[0019] Randomly select system master key components The computational system public key PK is used, where α is the exponential parameter of the bilinear group G1, and β is the randomization factor for attribute-based encryption.

[0020] PK = (g, g α ,g β ,e(g,g) α (H1,H2)(H1:{0,1}) * →G1,H2:G2→{0,1} 256 (For hash functions);

[0021] Master key MK = (α,β,g) α ), which is secretly stored by TA.

[0022] S1.3: Assign a private key SK to the user attribute set S; specifically: the data manager or supervisory node DR registers with TA, providing the attribute set. Includes the attribute set S = {Supervision unit, Level 1 qualification} for supervision units;

[0023] TA randomly selected r is a randomly selected blinding factor for calculating the private key.

[0024] The private key SK corresponding to user attribute set S. s The system returns information to the user via a secure channel.

[0025] As a further technical solution of the present invention, the image data in S2 is acquired by a high-definition camera with a resolution of 1920×1080 and a frame rate of 25fps; the physical quantity data includes the ambient temperature acquired by the DS18B20 temperature sensor, the ambient humidity acquired by the SHT30 humidity sensor, and the device operating current acquired by the ACS712 current sensor.

[0026] As a further technical solution of the present invention, the localized real-time analysis of S3 includes the following steps:

[0027] S3.1: Perform Gaussian filtering for noise reduction and normalization preprocessing on the image data;

[0028] S3.2: Outlier removal is performed on physical quantity data using the 3σ criterion; specifically: the sensor sequence X = (x1, x2, ..., x...) of the temperature, humidity, and current data collected by the sensors. m ), use the 3σ criterion to remove outliers: if Where x i Let X be the raw sensor reading at the i-th sampling point, μ be the sample mean of the data window, σ be the sample mean of the data window, m be the sliding window size, and output the preprocessed sensor sequence X. proc ;

[0029] S3.3: Employ a lightweight target detection model to identify security risks in image data;

[0030] S3.4: Sensor anomaly detection is performed using a time-series anomaly detection model; the time-series anomaly detection model uses an LSTM model, with the preprocessed sensor sequence X as input. proc =(x'1,x'2,...,x' 30 (30-second window, output exception score:)

[0031] s sen =σ(W oh hT+b o )

[0032] Where σ is the Sigmoid function, W oh The output gate weight matrix at time step t, h is the hidden state of the LSTM at time step t, T is the time window length, and b o Output gate bias term; T = 30; s sen When the value is greater than 0.85, it is considered abnormal, including equipment current exceeding the limit and ambient temperature being too high. The threshold is determined by training with historical normal data.

[0033] S3.5: Generate structured security monitoring results that include timestamps and GPS locations.

[0034] As a further technical solution of the present invention, in S3.1, the RGB image I∈R captured by the camera H×W×3 (H=1080, W=1920) After Gaussian filtering and denoising, the result is:

[0035]

[0036] Among them, I filter (x,y) represents the pixel value at (x,y) in the filtered image, and k is the radius of the filter kernel. Let I(x+i,y+j) be the weight of the Gaussian kernel (k=1, σ=0.8) at (i,j), where I(x+i,y+j) is the pixel value of the original image at (x+i,y+j), i,j are the relative coordinate offsets within the kernel, x,y are the absolute coordinate positions in the image, and σ is the standard deviation of the Gaussian distribution; after normalization to 640×640:

[0037]

[0038] Among them, I filter I is the pixel matrix of the image after Gaussian filtering. norm This is the normalized image pixel matrix.

[0039] As a further technical solution of the present invention, identifying safety hazards in S3.3 includes:

[0040] The lightweight object detection model uses a lightweight YOLOv8-nano model, and the output detection results are as follows:

[0041] R img ={(c1,b1,s1),(c2,b2,s2),…}

[0042] Where c is the category, including not wearing a safety helmet and not wearing a safety belt while working at height; b = (x1, y1, x2, y2) is the bounding box; s is the confidence level, and s ≥ 0.7 is considered valid;

[0043] The lightweight object detection model uses CIoU loss as its loss function:

[0044]

[0045] in, ρ is the intersection-union ratio; ρ is the Euclidean distance between the centers; c is the diagonal length of the minimum bounding rectangle; b is the coordinates of the predicted bounding box; b gt α represents the coordinates of the true bounding box; α is the adaptive weighting coefficient; v is the aspect ratio consistency measure.

[0046]

[0047] Where w and h are the width and height of the predicted bounding box; w gt ,h gt represents the width and height of the actual bounding box; IoU is the intersection-union ratio [0,1].

[0048] As a further technical solution of the present invention, the generation of structured security monitoring results in S3.5 specifically includes:

[0049] By merging image detection and sensor anomaly results, structured security monitoring results are generated.

[0050] Res={t proc ,Loc,R img ,R sen Suggested solutions}

[0051] Among them, t proc For processing time; Loc is the GPS location; R img A collection of image analysis results; R sen For sensor readings.

[0052] As a further technical solution of the present invention, the layered encryption mechanism in S4 includes the following steps:

[0053] S4.1: Use AES-256 symmetric encryption to encrypt the original data;

[0054] Specifically:

[0055] The data manager specifies access control policy T, and the specific process is as follows: 1. Policy formulation and symmetric encryption: The data manager first defines access control policy T, and then encrypts the actual project data files using a randomly generated AES-256 key; 2. Asymmetric encryption protection key: The AES key itself is not directly shared, but is protected through attribute-based encryption; 3. Final ciphertext: The ciphertext finally stored in the blockchain or cloud server consists of two parts: the data file encrypted with AES-256 and the AES key encrypted with ABE, using AES-256 symmetric encryption.

[0056] Res:C = AES - Enc(ck,Res)

[0057] Where ck is a 256-bit symmetric key; Res is the plaintext data to be encrypted; AES-Enc is the encryption algorithm function; and C is the encryption output.

[0058] S4.2: Based on access control policy T, use CP-ABE attribute base encryption to generate ciphertext and output the encryption result;

[0059] Specifically: with the encrypted proxy server PS enc Collaboration, using CP-ABE to encrypt ck:

[0060] Randomly select s, t∈Z p * s,t are components of the system master key. An access tree is constructed for the access control policy T, and a is assigned to each leaf node l. l ∈Z p a l Z is the private key component of the leaf node's attributes. p It is a group of non-zero integer multiplications;

[0061] Calculate the ciphertext:

[0062] CT=(T,C,g s C0 = e(g,g) αs ·H2(ck),

[0063] Final ciphertext:

[0064] CT final =(CT,T)

[0065] Where CT is the ciphertext, T is the access structure, C is the encrypted message body, g is the generator of the cyclic group, s is the random exponent, α is the system master key component, and e(g,g) is the ciphertext. αsThe result is a bilinear mapping, where ck is the symmetric content key, H2(ck) is the hash value of the key, l is the index of the leaf node in the access tree, and Attr is the key value. l Let t and α be the attributes corresponding to the l-th leaf node. l H1(Attr) is a random exponent related to the attribute. l ) is the hash value of the attribute.

[0066] As a further technical solution of the present invention, the step of writing the blockchain evidence storage module in S5 includes the following steps:

[0067] S5.1: A fragmented storage strategy is adopted for distributed storage of encrypted data;

[0068] The specific sharding storage strategy includes:

[0069] S5.1.1: Calculate the data hash value and generate a digital signature:

[0070] Specifically: The data manager calculates the hash signature:

[0071] Hash = SHA-256(CT) final ||DM.Identity||Address DM ||t0)

[0072] Sign with private key SK:

[0073] Sig DM =ECDSA-sign(SK) DM (Hash)

[0074] Where Hash is the digest of the message to be signed, SHA-256 is the cryptographically secure hash algorithm, and CT... final For final encrypted data, DM.Identity serves as the unique identifier for the data administrator, and Address... DM For the data manager's blockchain address, t0 is the timestamp, Sig DM For digital signatures, ECDSA-sign is the elliptic curve digital signature algorithm function, and SK... DM For data administrators' private keys;

[0075] S5.1.2: Split the encrypted data into 8 fragments;

[0076] Specifically: Send {CT final ,DM.Identity,Address DM ,t0,Sig DM} to edge server ES1;

[0077] ES1 determines storage capacity:

[0078]

[0079] Where SC represents storage capacity; peer-to-peer unloading is performed, splitting the CT into 8 shards. shardi ;

[0080] S5.1.3: Generate an independent signature for each fragment;

[0081] Specifically:

[0082]

[0083] in, The digital signature generated for edge server 1, ESDSA-sign is the elliptic curve digital signature algorithm function. This is the private key for edge server 1. SHA-256 is a cryptographically secure hash algorithm. CT shardi The ciphertext of the i-th data segment, For the blockchain address of edge server 1, Address shardi Let t be the target storage address for the i-th data fragment. 1i Timestamp; distributed to collaborative edge server ES i ;

[0084] S5.1.4: Distribute sharded storage on collaborative edge servers (ES). i node;

[0085] Specifically: storing information (ID) CT ,storage,DM.Identity,Sig DM After the offload / no-offload (s) is consensused by PBFT, it is written into the blockchain. The blockchain structure is: Block = (Index, PrevHash, Date, Timestamp, Signature).

[0086] As a further technical solution of the present invention, the S6 access to the evidence storage data includes:

[0087] S6.1: The supervisory node DR initiates an access request with attribute proof to the neighboring edge server ES1;

[0088] Specifically: The supervisory node (DR) sends a request (ID). CT The neighboring edge server ES1 records {ID} on the blockchain. CT The query retrieves the shard address and then sends it to the collaborative edge server ES. i Request CT shardi ;

[0089] S6.2: Synthesize the fragmented data of the distributed storage using the XOR reverse algorithm;

[0090] Specifically: Collaborative Edge Server (ES) i return Nearby edge server ES1 signature verification: Synthesizing CT using the XOR inverse algorithm final ; Query the blockchain to obtain Sig DM , return {ID CT CT final Sig DM The response information is recorded to the blockchain by the regulatory node DR; the CT is synthesized using the XOR reverse algorithm. final Specific synthesis mechanism: Collaborative edge server ES i Data fragmentation and encrypted CT shardi and its digital signature Returning to the coordinating node ES1, ES1 first verifies the authenticity of the signature to ensure the trustworthiness of the fragment source and the integrity of the transmission. After successful verification, it reassembles multiple valid fragments into the original ciphertext CT through an XOR inverse operation. final Subsequently, the blockchain was queried to obtain data from the data manager DM regarding CT. final authoritative signature Sig DM To prove the overall validity of the data, the final result will include the identifier ID. CT ciphertext CT final and signature Sig DM The complete data packet is submitted to the supervisory node DR, and the entire chain information of this synthetic response operation is recorded to the blockchain to form an immutable audit credential;

[0091] S6.3: The decryption proxy server performs a partial decryption operation;

[0092] Specifically: Regulatory node DR verification Sig DM Verify (PK) DM ,Hash,Sig DM After passing, extract the sub-private key. Send to decryption proxy server PS dec Decrypting the proxy server PS dec Perform semi-decryption: T = e(D) Attr C l ) al Where T is the intermediate calculation result, e is the bilinear mapping, and D is the intermediate result. Attr For user attribute private key components, C lLet `al` be the attribute-related component in the ciphertext, `al` be the secret share associated with the attribute, and `l` be the index of the leaf node in the access tree. Based on the access tree satisfaction calculation, the access control policy `T` and the signature are returned.

[0093] S6.4: The monitoring node completes the final decryption;

[0094] Specifically: Regulatory node DR verification Perform the remaining decryption: ck = H2 -1 (C0 / T), where ck is the symmetric content key, H2 -1 This is the "inverse" operation of the hash function. C0 is the core component of the ciphertext, and T is the intermediate value calculated during the decryption process. Decrypting with ck yields Res = AES - Dec(ck, C).

[0095] Secondly, a device integrating edge AI processing and blockchain evidence storage for safety monitoring data at power distribution construction sites is provided, including:

[0096] Data acquisition module: high-definition camera, temperature sensor, humidity sensor, current sensor and data aggregation unit;

[0097] Edge AI processing module: preprocessing unit, AI model inference unit, and result integration unit deployed on an 8TOPS computing power edge server;

[0098] Blockchain evidence storage module: A Hyperledger Fabric consortium blockchain network consisting of 3-5 edge nodes;

[0099] Trusted Support Layer: Trusted Authority (TA), Encrypted Proxy Server (PS) enc Decrypting the proxy server PS dec ;

[0100] User interaction module: A terminal device used for querying data at monitoring nodes;

[0101] Each layer and module is interconnected via industrial Ethernet and 5G slicing private network, with a transmission latency of less than 50ms.

[0102] Thirdly, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the edge AI processing and blockchain evidence storage method for power distribution construction as described in the first aspect.

[0103] Compared with existing technologies, the beneficial effects of this edge AI processing and blockchain evidence storage method, equipment, and medium in power distribution construction are:

[0104] By deploying lightweight AI models, YOLOv8-nano and LSTM at the edge, local data processing is performed, avoiding the network latency issues of traditional cloud transmission and shortening the response time for identifying safety hazards. At the same time, the data quality is improved by combining Gaussian filtering and 3σ criterion preprocessing algorithms, which improves the accuracy of target recognition and the recall rate of anomaly detection, ensuring that risks such as personnel violations and equipment overheating at power distribution construction sites are detected and warned of in a timely manner.

[0105] It adopts a closed-loop architecture of edge AI processing, layered encryption, and blockchain notarization. Through the collaborative mechanism of CP-ABE attribute base encryption and AES-256 symmetric encryption, it achieves access control that can only be decrypted by authorized nodes with the assistance of an encryption proxy server. Combined with the PBFT consensus and sharded storage strategy of blockchain notarization, it makes the notarized data immutable and traceable, expands the storage capacity, and provides a reliable basis for security responsibility tracing.

[0106] Edge AI local processing reduces cloud data transmission volume, significantly reducing network bandwidth pressure; the encryption and decryption proxy servers in the layered encryption mechanism undertake complex calculations, reducing the computing power consumption of terminal devices; key distribution and access policy tree control based on attribute sets ensure that sensitive data is only accessible to authorized parties, including construction unit administrators and supervision units, effectively preventing the leakage of privacy information during the construction process. Attached Figure Description

[0107] Figure 1 This is a schematic diagram of the method flow of the present invention. Detailed Implementation

[0108] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0109] Please see the appendix Figure 1 The present invention provides an embodiment of an edge AI processing and blockchain evidence storage method for power distribution construction, comprising the following steps:

[0110] S1: System initialization, where the power regulatory department to which the power distribution construction project belongs or its authorized third-party technical certification body acts as the trusted authority (TA). The trusted authority (TA) generates the system public key (PK), master key (MK), and private key (SK) using the CP-ABE key generation algorithm, which are used for subsequent decryption operations; specifically including the following steps:

[0111] S1.1: The trusted authority TA generates the parameters of the bilinear group system; specifically, the trusted authority TA inputs the entire attribute set U = {Attr1, Attr2, ..., Attr...} n The power distribution scenario attributes include construction unit administrators, supervision units, safety supervision departments, and system maintenance personnel, and the CP-ABE key generation algorithm is executed.

[0112] S1.2: Generate the system public key PK and master key MK based on the attribute set; specifically: select bilinear groups G1 and G2, with order p as a prime number, bilinear mapping G1×G1→G2, and generator g∈G1;

[0113] Randomly select system master key components The computational system public key PK is used, where α is the exponential parameter of the bilinear group G1, and β is the randomization factor for attribute-based encryption.

[0114] PK = (g, g α ,g β ,e(g,g) α (H1,H2)(H1:{0,1}) * →G1,H2:G2→{0,1} 256 (For hash functions);

[0115] Master key MK = (α,β,g) α ), which is secretly stored by TA.

[0116] S1.3: Assign a private key SK to the user attribute set S; specifically: the data manager or supervisory node DR registers with TA, providing the attribute set. Includes the attribute set S = {Supervision unit, Level 1 qualification} for supervision units;

[0117] TA randomly selected r is a randomly selected blinding factor for calculating the private key.

[0118] The private key SK corresponding to user attribute set S. s The system returns the information to the user via a secure channel.

[0119] S2: The data acquisition module acquires multi-source heterogeneous data from the power distribution construction site in real time, including image data and physical quantity data. The image data is acquired by a high-definition camera with a resolution of 1920×1080 and a frame rate of 25fps. The physical quantity data includes ambient temperature acquired by the DS18B20 temperature sensor, ambient humidity acquired by the SHT30 humidity sensor, and equipment operating current acquired by the ACS712 current sensor.

[0120] S3: The edge AI processing module performs localized real-time analysis on the collected multi-source heterogeneous data to generate security monitoring results. Specifically, it deploys a lightweight YOLOv8-nano model on the edge server to identify image data anomalies and combines an improved LSTM model to analyze sensor time-series data, including the following steps:

[0121] S3.1: Perform Gaussian filtering for noise reduction and normalization preprocessing on the image data. The RGB image I∈R acquired by the camera... H×W×3 (H=1080, W=1920) After Gaussian filtering and denoising, the result is:

[0122]

[0123] Among them, I filter (x,y) represents the pixel value at (x,y) in the filtered image, and k is the radius of the filter kernel. Let I(x+i,y+j) be the weight of the Gaussian kernel (k=1, σ=0.8) at (i,j), where I(x+i,y+j) is the pixel value of the original image at (x+i,y+j), i,j are the relative coordinate offsets within the kernel, x,y are the absolute coordinate positions in the image, and σ is the standard deviation of the Gaussian distribution; after normalization to 640×640:

[0124]

[0125] Among them, I filter I is the pixel matrix of the image after Gaussian filtering. norm This is the normalized image pixel matrix;

[0126] S3.2: Outlier removal is performed on physical quantity data using the 3σ criterion; specifically: the sensor sequence X = (x1, x2, ..., x...) of the temperature, humidity, and current data collected by the sensors. m ), use the 3σ criterion to remove outliers: if |x i If -μ|>3σ, then remove Where x i Let X be the raw sensor reading at the i-th sampling point, μ be the sample mean of the data window, σ be the sample mean of the data window, m be the sliding window size, and output the preprocessed sensor sequence X. proc ;

[0127] S3.3: A lightweight object detection model is used to identify security risks in image data. The lightweight object detection model adopts the lightweight YOLOv8-nano model, and the output detection results are as follows:

[0128] R img ={(c1,b1,s1),(c2,b2,s2),…}

[0129] Where c is the category, including not wearing a safety helmet and not wearing a safety belt while working at height; b = (x1, y1, x2, y2) is the bounding box; s is the confidence level, and s ≥ 0.7 is considered valid;

[0130] The lightweight object detection model uses CIoU loss as its loss function:

[0131]

[0132] in, ρ is the intersection-union ratio; ρ is the Euclidean distance between the centers; c is the diagonal length of the minimum bounding rectangle; b is the coordinates of the predicted bounding box; b gt α represents the coordinates of the true bounding box; α is the adaptive weighting coefficient; v is the aspect ratio consistency measure.

[0133]

[0134]

[0135] Where w and h are the width and height of the predicted bounding box; w gt ,h gt The width and height of the actual bounding box; IoU is the intersection-union ratio [0,1];

[0136] S3.4: Sensor anomaly detection is performed using a time-series anomaly detection model; the time-series anomaly detection model uses an LSTM model, with the preprocessed sensor sequence X as input. proc =(x'1,x'2,...,x' 30 (30-second window, output exception score:)

[0137] s sen =σ(W oh hT+b o )

[0138] Where σ is the Sigmoid function, W oh The output gate weight matrix at time step t, h is the hidden state of the LSTM at time step t, T is the time window length, and b o Output gate bias term; T = 30; s sen When the value is greater than 0.85, it is considered abnormal, including equipment current exceeding the limit and ambient temperature being too high. The threshold is determined by training with historical normal data.

[0139] S3.5: Generate structured security monitoring results including timestamps and GPS locations, including merging image detection and sensor anomaly results to generate structured security monitoring results:

[0140] Res={t proc ,Loc,R img ,R senSuggested solutions}

[0141] Among them, t proc For processing time; Loc is the GPS location; R img A collection of image analysis results; R sen For sensor readings;

[0142] S4: Via encrypted proxy server PS enc The security monitoring results are processed using a layered encryption mechanism to generate encrypted results, including the following steps:

[0143] S4.1: Use AES-256 symmetric encryption to encrypt the original data;

[0144] Specifically:

[0145] The data manager specifies access control policy T, and the specific process is as follows: 1. Policy formulation and symmetric encryption: The data manager first defines access control policy T, and then encrypts the actual project data files using a randomly generated AES-256 key; 2. Asymmetric encryption protection key: The AES key itself is not directly shared, but is protected through attribute-based encryption; 3. Final ciphertext: The ciphertext finally stored in the blockchain or cloud server consists of two parts: the data file encrypted with AES-256 and the AES key encrypted with ABE, using AES-256 symmetric encryption.

[0146] Res:C = AES - Enc(ck,Res)

[0147] Where ck is a 256-bit symmetric key; Res is the plaintext data to be encrypted; AES-Enc is the encryption algorithm function; and C is the encryption output.

[0148] S4.2: Based on access control policy T, use CP-ABE attribute base encryption to generate ciphertext and output the encryption result;

[0149] Specifically: with the encrypted proxy server Ps enc Collaboration, using CP-ABE to encrypt ck:

[0150] Randomly select s, t∈Z p * s,t are components of the system master key. An access tree is constructed for the access control policy T, and a is assigned to each leaf node l. l ∈Z p a l Z is the private key component of the leaf node's attributes. p It is a group of non-zero integer multiplications;

[0151] Calculate the ciphertext:

[0152]

[0153] Final ciphertext:

[0154] CT final =(CT,T)

[0155] Where CT is the ciphertext, T is the access structure, C is the encrypted message body, g is the generator of the cyclic group, s is the random exponent, α is the system master key component, and e(g,g) is the ciphertext. αs The result is a bilinear mapping, where ck is the symmetric content key, H2(ck) is the hash value of the key, l is the index of the leaf node in the access tree, and Attr is the key value. l Let t and α be the attributes corresponding to the l-th leaf node. l H1(Attr) is a random exponent related to the attribute. l ) is the hash value of the attribute;

[0156] S5: Write the encrypted result into the blockchain notarization module, use a sharding storage strategy for distributed notarization, and generate notarization data, including the following steps:

[0157] S5.1: A fragmented storage strategy is adopted for distributed storage of encrypted data;

[0158] The specific sharding storage strategy includes:

[0159] S5.1.1: Calculate the data hash value and generate a digital signature:

[0160] Specifically: The data manager calculates the hash signature:

[0161] Hash = SHA-256(CT) final ||DM.Identity||Address DM ||t0)

[0162] Sign with private key SK:

[0163] Sig DM =ECDSA-sign(SK) DM (Hash)

[0164] Where Hash is the digest of the message to be signed, SHA-256 is the cryptographically secure hash algorithm, and CT... final For final encrypted data, DM.Identity serves as the unique identifier for the data administrator, and Address... DM For the data manager's blockchain address, t0 is the timestamp, Sig DM For digital signatures, ECDSA-sign is the elliptic curve digital signature algorithm function, and SK... DM For data administrators' private keys;

[0165] S5.1.2: Split the encrypted data into 8 fragments;

[0166] Specifically: Send {CT fianl ,DM.Identity,Address DM ,t0,Sig DM} to edge server ES1;

[0167] ES1 determines storage capacity:

[0168]

[0169] Where SC represents storage capacity; peer-to-peer unloading is performed, splitting the CT into 8 shards. shardi ;

[0170] S5.1.3: Generate an independent signature for each fragment;

[0171] Specifically:

[0172]

[0173] in, The digital signature generated for edge server 1, ESDSA-sign is the elliptic curve digital signature algorithm function. This is the private key for edge server 1. SHA-256 is a cryptographically secure hash algorithm. CT shardi The ciphertext of the i-th data segment, For the blockchain address of edge server 1, Address shardi Let t be the target storage address for the i-th data fragment. 1i Timestamp; distributed to collaborative edge server ES i ;

[0174] S5.1.4: Distribute sharded storage on collaborative edge servers (ES). i node;

[0175] Specifically: storing information (ID) CT ,storage,DM.Identity,Sig DM After being written to the blockchain via PBFT consensus, the offload / no-offload (s) is structured as follows: Block = (Index, PrevHash, Date, Timestamp, Signature).

[0176] S6: Authorized Supervisory Node (DR) decrypts the proxy server PS dec Verify and access the stored evidence data using the key generated by S1, including:

[0177] S6.1: The supervisory node DR initiates an access request with attribute proof to the neighboring edge server ES1;

[0178] Specifically: The supervisory node (DR) sends a request (ID). CT The neighboring edge server ES1 records {ID} on the blockchain. CT The query retrieves the shard address and then sends it to the collaborative edge server ES. i Request CT shardi ;

[0179] S6.2: Synthesize the fragmented data of the distributed storage using the XOR reverse algorithm;

[0180] Specifically: Collaborative Edge Server (ES) i return Nearby edge server ES1 signature verification: Synthesizing CT using the XOR inverse algorithm final Query the blockchain to obtain Sig DM , return {ID CT CT final Sig DM The response information is recorded to the blockchain by the regulatory node DR; the CT is synthesized using the XOR reverse algorithm. final Specific synthesis mechanism: Collaborative edge server ES i Data fragmentation and encrypted CT shardi and its digital signature Returning to the coordinating node ES1, ES1 first verifies the authenticity of the signature to ensure the trustworthiness of the fragment source and the integrity of the transmission. After successful verification, it reassembles multiple valid fragments into the original ciphertext CT through an XOR inverse operation. final Subsequently, the blockchain was queried to obtain data from the data manager DM regarding CT. final authoritative signature Sig DM To prove the overall validity of the data, the final result will include the identifier ID. CT ciphertext CT final and signature Sig DM The complete data packet is submitted to the supervisory node DR, and the entire chain information of this synthetic response operation is recorded to the blockchain to form an immutable audit credential;

[0181] S6.3: The decryption proxy server performs a partial decryption operation;

[0182] Specifically: Regulatory node DR verification Sig DM Verify (PK) DM ,Hash,Sig DMAfter passing, extract the sub-private key. Send to decryption proxy server PS dec Decrypting the proxy server PS dec Perform semi-decryption: T = e(D) Attr C l ) al Where T is the intermediate calculation result, e is the bilinear mapping, and D is the intermediate result. Attr For user attribute private key components, C l Let `al` be the attribute-related component in the ciphertext, `al` be the secret share associated with the attribute, and `l` be the index of the leaf node in the access tree. Based on the access tree satisfaction calculation, the access control policy `T` and the signature are returned.

[0183] S6.4: The monitoring node completes the final decryption;

[0184] Specifically: Regulatory node DR verification Perform the remaining decryption: ck = H2 -1 (C0 / T), where ck is the symmetric content key, H2 -1 This is the "inverse" operation of the hash function. C0 is the core component of the ciphertext, and T is the intermediate value calculated during the decryption process. Decrypting with ck yields Res = AES - Dec(ck, C).

[0185] One embodiment of the present invention provides: an integrated device for edge AI processing and blockchain evidence storage of safety monitoring data at power distribution construction sites, comprising;

[0186] Data acquisition module: high-definition camera, temperature sensor, humidity sensor, current sensor and data aggregation unit;

[0187] Edge AI processing module: preprocessing unit, AI model inference unit, and result integration unit deployed on an 8TOPS computing power edge server;

[0188] Blockchain evidence storage module: A Hyperledger Fabric consortium blockchain network consisting of 3-5 edge nodes;

[0189] Trusted Support Layer: Trusted Authority (TA), Encrypted Proxy Server (PS) enc Decrypting the proxy server PS dec ;

[0190] User interaction module: A terminal device used for querying data at monitoring nodes;

[0191] Each layer and module is interconnected via industrial Ethernet and 5G slicing private network, with a transmission latency of less than 50ms.

[0192] One embodiment of the present invention provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements an edge AI processing and blockchain evidence storage method for power distribution construction.

[0193] In summary, this invention deploys a lightweight AI model, YOLOv8-nano, and LSTM at the edge for localized data processing, avoiding the network latency issues of traditional cloud transmission and shortening the response time for identifying safety hazards. At the same time, it combines Gaussian filtering and 3σ criterion preprocessing algorithms to improve data quality, thereby increasing the accuracy of target identification and the recall rate of anomaly detection, ensuring that risks such as personnel violations and equipment overheating at power distribution construction sites are detected and warned of in a timely manner.

[0194] It adopts a closed-loop architecture of edge AI processing, layered encryption, and blockchain notarization. Through the collaborative mechanism of CP-ABE attribute base encryption and AES-256 symmetric encryption, it achieves access control that can only be decrypted by authorized nodes with the assistance of an encryption proxy server. Combined with the PBFT consensus and sharded storage strategy of blockchain notarization, it makes the notarized data immutable and traceable, expands the storage capacity, and provides a reliable basis for security responsibility tracing.

[0195] Edge AI local processing reduces cloud data transmission volume, significantly reducing network bandwidth pressure; the encryption and decryption proxy servers in the layered encryption mechanism undertake complex calculations, reducing the computing power consumption of terminal devices; key distribution and access policy tree control based on attribute sets ensure that sensitive data is only accessible to authorized parties, including construction unit administrators and supervision units, effectively preventing the leakage of privacy information during the construction process.

[0196] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

Claims

1. An edge AI processing and blockchain evidence storage method for power distribution construction, characterized by: Includes the following steps: S1: System initialization, the power regulatory department to which the power distribution construction project belongs or its authorized third-party technical certification body acts as the trusted authority (TA). The trusted authority (TA) generates the system public key (PK), master key (MK), and private key (SK) through the CP-ABE key generation algorithm, which are used for subsequent decryption operations. S2: Real-time acquisition of multi-source heterogeneous data from the power distribution construction site through the data acquisition module, including image data and physical quantity data; S3: The edge AI processing module performs local real-time analysis on the collected multi-source heterogeneous data to generate security monitoring results. Specifically, a lightweight YOLOv8-nano model is deployed on the edge server to identify image data anomalies, and an improved LSTM model is combined to analyze sensor time-series data. S4: Via encrypted proxy server PS enc A layered encryption mechanism is used to process security monitoring results and generate encrypted results. S5: Write the encrypted result into the blockchain evidence storage module, use a sharding storage strategy for distributed evidence storage, and generate evidence storage data; S6: Authorized Supervisory Node (DR) decrypts the proxy server PS dec Verify and access the stored evidence data using the key generated by S1.

2. The edge AI processing and blockchain evidence storage method for power distribution construction according to claim 1, characterized in that: S1 specifically includes the following steps: S1.1: The trusted authority TA generates the parameters of the bilinear group system; specifically, the trusted authority TA inputs the entire attribute set U = {Attr1, Attr2, ..., Attr...} n The power distribution scenario attributes include construction unit administrators, supervision units, safety supervision departments, and system maintenance personnel, and the CP-ABE key generation algorithm is executed. S1.2: Generate the system public key PK and master key MK based on the attribute set; specifically: select bilinear groups G1 and G2, with order p as a prime number, bilinear mapping G1×G1→G2, and generator g∈G1; Randomly select system master key components The computational system public key PK is used, where α is the exponential parameter of the bilinear group G1, and β is the randomization factor for attribute-based encryption. PK = (g, g α ,g β ,e(g,g) α (h1:{0,1}) * →G1,H2:G2→{0,1} 256 (For hash functions); Master key MK = (α,β,g) α ), which is secretly stored by TA. S1.3: Assign a private key SK to the user attribute set S; specifically: the data manager or supervisory node DR registers with TA, providing the attribute set. Includes the attribute set S = {Supervision unit, Level 1 qualification} for supervision units; TA randomly selected r is a randomly selected blinding factor for calculating the private key. The private key SK corresponding to user attribute set S. s The system returns information to the user via a secure channel.

3. The edge AI processing and blockchain evidence storage method for power distribution construction according to claim 1, characterized in that: The image data in S2 is acquired by a high-definition camera with a resolution of 1920×1080 and a frame rate of 25fps; the physical quantity data includes ambient temperature acquired by a DS18B20 temperature sensor, ambient humidity acquired by an SHT30 humidity sensor, and device operating current acquired by an ACS712 current sensor.

4. The edge AI processing and blockchain evidence storage method for power distribution construction according to claim 1, characterized in that: The localized real-time analysis of S3 Includes the following steps: S3.1: Perform Gaussian filtering for noise reduction and normalization preprocessing on the image data; S3.2: Outlier removal is performed on physical quantity data using the 3σ criterion; specifically: the sensor sequence X = (x1, x2, ..., x...) of the temperature, humidity, and current data collected by the sensors. m ), use the 3σ criterion to remove outliers: if |x i If -μ|>3σ, then remove Where x i Let X be the raw sensor reading at the i-th sampling point, μ be the sample mean of the data window, σ be the sample mean of the data window, m be the sliding window size, and output the preprocessed sensor sequence X. proc ; S3.3: Employ a lightweight target detection model to identify security risks in image data; S3.4: Sensor anomaly detection is performed using a time-series anomaly detection model; The time-series anomaly detection model uses an LSTM model, with the preprocessed sensor sequence X as input. proc =(x'1,x'2,...,x' 30 (30-second window, output exception score:) s sen =σ(W oh hT+b o ) Where σ is the Sigmoid function, W oh The output gate weight matrix at time step t, h is the hidden state of the LSTM at time step t, T is the time window length, and b o Output gate bias term; T = 30; s sen When the value is greater than 0.85, it is considered abnormal, including equipment current exceeding the limit and ambient temperature being too high. The threshold is determined by training with historical normal data. S3.5: Generate structured security monitoring results that include timestamps and GPS locations.

5. The edge AI processing and blockchain evidence storage method for power distribution construction according to claim 4, characterized in that: In S3.1, the RGB image I∈R captured by the camera H×W×3 (H=1080, W=1920) After Gaussian filtering and denoising, the result is: Among them, I filter (x,y) represents the pixel value at (x,y) in the filtered image, and k is the radius of the filter kernel. Let I(x+i,y+j) be the weight of the Gaussian kernel (k=1, σ=0.8) at (i,j), where I(x+i,y+j) is the pixel value of the original image at (x+i,y+j), i,j are the relative coordinate offsets within the kernel, x,y are the absolute coordinate positions in the image, and σ is the standard deviation of the Gaussian distribution; after normalization to 640×640: Among them, I filter I is the pixel matrix of the image after Gaussian filtering. norm This is the normalized image pixel matrix.

6. The edge AI processing and blockchain evidence storage method for power distribution construction according to claim 4, characterized in that: The identification of security risks in S3.3 includes: The lightweight object detection model uses a lightweight YOLOv8-nano model, and the output detection results are as follows: R img ={(c1,b1,s1),(c2,b2,s2),…} Where c is the category, including not wearing a safety helmet and not wearing a safety belt while working at height; b = (x1, y1, x2, y2) is the bounding box; s is the confidence level, and s ≥ 0.7 is considered valid; The lightweight object detection model uses CIoU loss as its loss function: in, ρ is the intersection-union ratio; ρ is the Euclidean distance between the centers; c is the diagonal length of the minimum bounding rectangle; b is the coordinates of the predicted bounding box; b gt α represents the coordinates of the true bounding box; α is the adaptive weighting coefficient; v is the aspect ratio consistency measure. Where w and h are the width and height of the predicted bounding box; w gt ,h gt represents the width and height of the actual bounding box; IoU is the intersection-union ratio [0,1].

7. The edge AI processing and blockchain evidence storage method for power distribution construction according to claim 4, characterized in that: The specific structured security monitoring results generated by S3.5 are as follows: By merging image detection and sensor anomaly results, structured security monitoring results are generated. Res={t proc ,Loc,R img ,R sen Suggested solutions} Among them, t proc For processing time; Loc is the GPS location; R img A collection of image analysis results; R sen This refers to the sensor reading.

8. The edge AI processing and blockchain evidence storage method for power distribution construction according to claim 1, characterized in that: The layered encryption mechanism in S4 includes the following steps: S4.1: Use AES-256 symmetric encryption to encrypt the original data; Specifically: The data administrator specifies access control policy T, using AES-256 symmetric encryption: Res:C = AES - Enc(ck,Res) Where ck is a 256-bit symmetric key; Res is the plaintext data to be encrypted; AES-Enc is the encryption algorithm function; and C is the encryption output. S4.2: Based on access control policy T, use CP-ABE attribute base encryption to generate ciphertext and output the encryption result; Specifically: with the encrypted proxy server PS enc Collaboration, using CP-ABE to encrypt ck: Randomly select s, t∈Z p * s,t are components of the system master key. An access tree is constructed for the access control policy T, and a is assigned to each leaf node l. l ∈Z p a l Z is the private key component of the leaf node's attributes. p It is a group of non-zero integer multiplications; Calculate the ciphertext: Final ciphertext: CT final =(CT,T) Where CT is the ciphertext, T is the access structure, C is the encrypted message body, g is the generator of the cyclic group, s is the random exponent, α is the system master key component, and e(g,g) is the ciphertext. αs The result is a bilinear mapping, where ck is the symmetric content key, H2(ck) is the hash value of the key, l is the index of the leaf node in the access tree, and Attr is the key value. l Let t and α be the attributes corresponding to the l-th leaf node. l H1(Attr) is a random exponent related to the attribute. l ) is the hash value of the attribute.

9. The edge AI processing and blockchain evidence storage method for power distribution construction according to claim 1, characterized in that: The steps involved in writing the blockchain evidence storage module in S5 are as follows: S5.1: A fragmented storage strategy is adopted for distributed storage of encrypted data; The specific sharding storage strategy includes: S5.1.1: Calculate the data hash value and generate a digital signature: Specifically: The data manager calculates the hash signature: Hash=SHA-256(CT final ||DM.Identity||Address DM ||t0) Sign with private key SK: Say DM =ECDSA sign(SK DM (Hash) Where Hash is the digest of the message to be signed, SHA-256 is the cryptographically secure hash algorithm, and CT... final For final encrypted data, DM.Identity serves as the unique identifier for the data administrator, and Address... DM For the data manager's blockchain address, t0 is the timestamp, Sig DM For digital signatures, ECDSA-sign is the elliptic curve digital signature algorithm function, and SK... DM For data administrators' private keys; S5.1.2: Split the encrypted data into 8 fragments; Specifically: Send {CT fianl ,DM.Identity,Address DM ,t0,Sig DM } to edge server ES1; ES1 determines storage capacity: Where SC represents storage capacity; peer-to-peer unloading is performed, splitting the CT into 8 shards. shardi ; S5.1.3: Generate an independent signature for each fragment; Specifically: in, The digital signature generated for edge server 1, ESDSA-sign is the elliptic curve digital signature algorithm function. This is the private key for edge server 1. SHA-256 is a cryptographically secure hash algorithm. CT shardi The ciphertext of the i-th data segment, For the blockchain address of edge server 1, Address shardi Let t be the target storage address for the i-th data fragment. 1i For timestamps; Distributed to collaborative edge server ES i ; S5.1.4: Distribute sharded storage on collaborative edge servers (ES). i node; Specifically: storing information (ID) CT ,storage,DM.Identity,Sig DM After the offload / no-offload (s) is consensused by PBFT, it is written into the blockchain. The blockchain structure is: Block = (Index, PrevHash, Date, Timestamp, Signature).

10. The edge AI processing and blockchain evidence storage method for power distribution construction according to claim 1, characterized in that: The S6 access evidence storage data includes: S6.1: The supervisory node DR initiates an access request with attribute proof to the neighboring edge server ES1; Specifically: The supervisory node (DR) sends a request (ID). CT The neighboring edge server ES1 records {ID} on the blockchain. CT The query retrieves the shard address and then sends it to the collaborative edge server ES. i Request CT shardi ; S6.2: Synthesize the fragmented data of the distributed storage using the XOR reverse algorithm; Specifically: Collaborative Edge Server (ES) i return Nearby edge server ES1 signature verification: Synthesizing CT using the XOR inverse algorithm final Query the blockchain to obtain Sig DM , return {ID CT CT final Sig DM The response information is recorded to the blockchain by the regulatory node DR. S6.3: The decryption proxy server performs a partial decryption operation; Specifically: Regulatory node DR verification Sig DM Verify (PK) DM ,Hash,Sig DM After passing, extract the sub-private key. Send to decryption proxy server PS dec Decrypting the proxy server PS dec Perform semi-decryption: T = e(D) Attr C l ) al Where T is the intermediate calculation result, e is the bilinear mapping, and D is the intermediate result. Attr For user attribute private key components, C l Let `al` be the attribute-related component in the ciphertext, `al` be the secret share associated with the attribute, and `l` be the index of the leaf node in the access tree. Based on the access tree satisfaction calculation, the access control policy `T` and the signature are returned. S6.4: The monitoring node completes the final decryption; Specifically: Regulatory node DR verification Perform the remaining decryption: ck = H2 -1 (C0 / T), where ck is the symmetric content key, H2 -1 This is the "inverse" operation of the hash function. C0 is the core component of the ciphertext, and T is the intermediate value calculated during the decryption process. Decrypting with ck yields Res = AES - Dec(ck, C).

11. A device integrating edge AI processing and blockchain evidence storage for safety monitoring data at power distribution construction sites, characterized in that, include; Data acquisition module: high-definition camera, temperature sensor, humidity sensor, current sensor and data aggregation unit; Edge AI processing module: preprocessing unit, AI model inference unit, and result integration unit deployed on an 8TOPS computing power edge server; Blockchain evidence storage module: A Hyperledger Fabric consortium blockchain network consisting of 3-5 edge nodes; Trusted Support Layer: Trusted Authority (TA), Encrypted Proxy Server (PS) enc Decrypting the proxy server PS dec ; User interaction module: A terminal device used for querying data at monitoring nodes; Each layer and module is interconnected via industrial Ethernet and 5G slicing private network, with a transmission latency of less than 50ms.

12. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by the processor, the program implements the edge AI processing and blockchain evidence storage method for power distribution construction as described in any one of claims 1-10.

Images