A real-time electricity fee settlement and intelligent deduction method for a smart meter

Abstract

The present invention discloses a method for real-time electricity bill settlement and intelligent deduction for a smart meter, belonging to the field of smart grid technology. The method comprises obtaining current waveform data through high-frequency sampling, generating a feature vector through differential exponential smoothing, and dynamically classifying equipment load types based on spectral characteristics and a lightweight convolutional neural network; generating dynamic rate parameters by combining the grid peak demand index and the user's historical credit score, and constructing a hybrid bill; forming a cross-chain transaction message by parsing the hash value and embedding the blockchain timestamp, and activating a three-stage settlement smart contract; and monitoring the load fluctuation data after deduction to generate abnormal fluctuation markers and optimize through federated learning. The present invention achieves the comprehensive optimization of high-precision real-time billing, tamper-proof trusted transactions, and adaptive anomaly monitoring, significantly improving the grid's peak-shaving efficiency and electricity fairness.

Description

Technical Field

[0001] The present invention relates to the field of smart grid technology, and in particular to a method for real-time electricity fee settlement and intelligent fee deduction using a smart meter. Background Art

[0002] In recent years, with the development of smart grid technology, smart meters, as core equipment for collecting electricity usage information, have been widely used in power systems. Traditional electricity bill settlement models, based on scheduled meter reading and fixed-rate mechanisms, are difficult to adapt to the needs of real-time electricity price response, user credit assessment, and dynamic grid peak regulation.

[0003] Currently, the sampling frequency of traditional electricity meters is unable to capture the transient characteristics of current waveforms, and feature extraction often relies on mean filtering or simple frequency domain analysis. This results in inadequate recognition of new electrical appliances by load classification algorithms, leading to frequent billing errors. Existing dynamic electricity pricing mechanisms are often based on pre-defined peak and off-peak periods, failing to incorporate the grid's real-time peak-shaving demand, user credit scores, and the risk of transient equipment overloads. This results in inefficient grid load balancing and unfair electricity costs for users, especially during periods of significant supply-demand imbalances. Electricity transaction data relies on centralized servers for storage and processing, exposing risks of bill tampering, delayed settlement, and malicious defaults. While some systems have incorporated blockchain technology, cross-chain transactions lack efficient verification mechanisms, and the execution logic of smart contracts is simplistic (e.g., relying solely on balance thresholds), making them inadequate for multi-tiered fund management. Detection of abnormal electricity usage, such as electricity theft and equipment failure, typically relies on manual verification or offline data analysis, resulting in long response times and high audit costs. Existing algorithms lack sensitivity to load fluctuation patterns, making it difficult to detect hidden anomalies in a timely manner. To address the above issues, the industry urgently needs an intelligent electricity fee management method that can achieve high-frequency real-time data collection, dynamic multi-dimensional rate integration, tamper-resistant trusted settlement, and a self-closed loop for abnormal fluctuations. Summary of the Invention

[0004] To solve the above problems, the present invention provides a real-time settlement and intelligent deduction method for electricity charges of smart meters. It adopts high-frequency sampling data coupled with waveform fingerprint compression, multi-dimensional dynamic rate generation and blockchain cross-chain verification technology, which can achieve millisecond-level analysis of load characteristics, tamper-proof real-time settlement and adaptive closed-loop monitoring of abnormal electricity consumption behavior, taking into account the peak-shaving efficiency of the power grid and the fairness of user electricity consumption.

[0005] The above objectives can be achieved through the following solutions:

[0006] A method for real-time settlement and intelligent deduction of electricity charges for smart meters, comprising: obtaining continuously sampled current waveform data on a power line, generating an original electric parameter sequence at a frequency of not less than 100 sampling points per second; receiving the original electric parameter sequence, and using differential exponential smoothing processing on a TSA chip based on the sampling timestamp of the original electric parameter sequence to generate a feature vector carrying a waveform fingerprint; outputting a device load type classification result through a lightweight convolutional neural network based on the spectral distribution characteristics of the feature vector; receiving the load type classification result, and combining it with the power grid peak demand index Generate dynamic rate parameters; receive the dynamic rate parameters, weight the user's historical credit score to generate a hybrid billing bill with a transient overload coefficient; parse the hash value of the hybrid billing bill, and embed the blockchain timestamp to form a cross-chain transaction message; after monitoring the cross-chain transaction message, parse the account pre-deposit status to activate the three-stage settlement smart contract; capture the load fluctuation data after deduction according to the three-stage settlement smart contract, and generate an abnormal fluctuation mark by comparing it with the equipment shutdown characteristic curve; receive the abnormal fluctuation mark, update the federated learning model parameters and write it into the tamper-proof ledger to complete the closed-loop processing.

[0007] Optionally, the generating of a feature vector carrying a waveform fingerprint by using differential exponential smoothing processing on a TSA chip based on the sampling timestamp of the original electrical parameter sequence includes: receiving the original electrical parameter sequence, extracting power differential values ​​of adjacent points and compressing them into a Huffman dictionary; parsing the Huffman dictionary, superimposing voltage phase mutations to generate a feature summary with a check bit; performing nonlinear filtering based on the feature summary, outputting a denoised feature vector, and irreversibly restoring the original waveform.

[0008] Optionally, the outputting of the equipment load type classification result through a lightweight convolutional neural network based on the spectral distribution characteristics of the feature vector includes: preprocessing the Fourier transform result of the feature vector to generate a device fingerprint mapping dictionary; receiving real-time current harmonic distribution data, traversing the mapping dictionary to output the equipment identification result; parsing unrecognized power curve patterns and writing them into the knowledge base to be optimized.

[0009] Optionally, the generation of dynamic rate parameters in combination with the power grid peak demand index includes: parsing the JSON data of the power grid peak demand index to generate a basic rate matrix; correcting the weight distribution of the basic rate matrix based on the credit adjustment coefficient generated according to the user's historical electricity usage behavior; correcting the weight distribution based on the loss compensation factor of the real-time temperature of the distribution transformer, and outputting the final dynamic rate parameters.

[0010] Optionally, parsing the hash value of the hybrid billing bill and incorporating the blockchain timestamp to form a cross-chain transaction message includes: parsing the JSON structure of the hybrid billing bill to separate the business logic data flow and the capital flow; pushing the separated business logic data flow unidirectionally to the alliance chain node for pre-confirmation; after receiving the pre-confirmation result, permanently writing the capital flow hash value into the genesis block of the transaction chain.

[0011] Optionally, the parsing of the account deposit status to activate the three-stage settlement smart contract includes: when the account deposit amount is greater than or equal to a preset first threshold, activating the full deduction module and generating a unique deduction voucher; when the account deposit amount is less than the first threshold and greater than or equal to a preset second threshold, generating a phased withholding agreement and locking the fund pool; when the account deposit amount is less than the second threshold, triggering the relay control sequence and freezing the electricity usage permission.

[0012] Optionally, triggering the relay control sequence and freezing electricity usage permissions includes: receiving grid load capacity data within the geographic fence, calculating a preset allowable power gradient; generating a device load reduction curve based on the preset allowable power gradient, and outputting a relay control instruction queue; burning the relay control instruction queue into the meter firmware and erasing the original control code.

[0013] Optionally, the capturing of load fluctuation data after deduction according to the three-stage settlement smart contract and the generation of abnormal fluctuation marks by comparing the equipment shutdown characteristic curve include: capturing the load fluctuation data after deduction to extract the power attenuation slope; comparing the power attenuation slope with the benchmark value of the equipment feature knowledge base to generate a deviation coefficient vector; parsing the line impedance mutation parameters in the deviation coefficient vector to construct an abnormal score matrix associated with multi-source data; when the abnormal score matrix exceeds a preset safety threshold, locking the funds transfer of the associated transaction and initiating manual review to generate an audit report containing the abnormal fluctuation mark.

[0014] Optionally, the updating of the federated learning model parameters and writing into the tamper-proof ledger to complete the closed-loop processing includes: receiving the abnormal fluctuation mark and the corresponding audit report, taking the time period of the abnormal fluctuation mark as the index, extracting the feature vector from the edge node to generate the data fingerprint, and outputting the global optimization requirements; according to the global optimization requirements, reconstructing the convolution kernel attention weight matrix, and writing the old parameter hash value into the tamper-proof ledger; compiling the updated weight matrix into a binary instruction set carrying the abnormal fluctuation mark version number, and encrypting and transmitting it to the designated meter cluster.

[0015] Based on the same inventive concept, the present invention also provides a real-time settlement and intelligent deduction system for electricity charges of smart meters, the system comprising: a data acquisition module for acquiring continuously sampled current waveform data on the power line, and generating an original electric parameter sequence at a frequency of not less than 100 sampling points per second; a feature vector generation module for receiving the original electric parameter sequence, and generating a feature vector carrying a waveform fingerprint by using differential exponential smoothing processing on a TSA chip based on the sampling timestamp of the original electric parameter sequence; an equipment load classification module for outputting the equipment load type classification result through a lightweight convolutional neural network according to the spectrum distribution characteristics of the feature vector; a rate parameter generation module for receiving the load type classification result, and combining it with the peak load regulation demand of the power grid. The exponent is calculated to generate dynamic rate parameters; a billing bill generation module is used to receive the dynamic rate parameters and weight the user's historical credit score to generate a hybrid bill containing a transient overload coefficient; a transaction message generation module is used to parse the hash value of the hybrid bill and embed the blockchain timestamp to form a cross-chain transaction message; a smart contract generation module is used to monitor the cross-chain transaction message, parse the account pre-deposit status and activate the three-stage settlement smart contract; an abnormal fluctuation mark generation module is used to capture the load fluctuation data after deduction according to the three-stage settlement smart contract, and generate an abnormal fluctuation mark by comparing the equipment shutdown characteristic curve; a feedback recording module is used to receive the abnormal fluctuation mark, update the federated learning model parameters and write it into the tamper-proof ledger to complete the closed-loop processing.

[0016] Compared with the prior art, the present invention has the following advantages:

[0017] 1. By combining high-frequency sampling technology with intelligent algorithms, refined analysis of current waveforms and real-time dynamic classification of load types are achieved. Differential exponential smoothing effectively extracts waveform features, and combined with lightweight convolutional neural networks, it significantly improves the ability to identify nonlinear devices and transient loads. This solves the problem of traditional methods having difficulty capturing complex electricity consumption characteristics, providing a reliable data foundation for accurate billing.

[0018] 2. Build a dynamic rate model based on real-time peak-shaving demand, user credit assessment, and equipment operating status, generating a hybrid billing strategy that integrates multiple regulatory factors. By flexibly responding to grid load fluctuations and personalized electricity consumption behavior, this model improves grid peak-shaving efficiency while ensuring fairness in user rights, effectively balancing supply and demand and optimizing energy distribution efficiency.

[0019] 3. The cross-chain blockchain architecture is used to decouple and verify the billing process, and the business logic and capital flow are separated to ensure the traceability of the entire transaction chain. The three-stage smart contract is combined with a hierarchical account management mechanism to achieve hierarchical prevention and control of arrears risks and automatic execution. The alliance chain pre-confirmation and genesis block anchoring technology enhance the transaction non-repudiation and build a transparent and reliable electricity bill settlement ecosystem.

[0020] 4. Capture equipment shutdown characteristics and line status changes in real time, establish an abnormal fluctuation marking mechanism to quickly identify hidden power consumption problems; the federated learning model continuously optimizes the load classification algorithm through distributed collaborative training, and combines the blockchain ledger solidification iterative process to form an autonomous evolutionary system for abnormal warning, parameter update and closed-loop disposal, comprehensively improving the system's robustness and long-term adaptability.

[0021] Other features and advantages of the present invention will be described in the following description, and in part will become apparent from the description, or will be understood by practicing the present invention. The purpose and other advantages of the present invention can be realized and obtained by the structures pointed out in the description, claims and drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0023] Figure 1 The present invention is a flowchart of a method for real-time electricity bill settlement and intelligent deduction of a smart meter according to an embodiment of the present invention.

[0024] Figure 2 The present invention is a schematic diagram of a system for real-time electricity bill settlement and intelligent deduction of smart meters. DETAILED DESCRIPTION

[0025] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.

[0026] Reference Figure 1 One embodiment of the present invention proposes a real-time electricity bill settlement and intelligent deduction method for a smart meter. By coupling high-frequency sampling data with waveform fingerprint compression, multi-dimensional dynamic rate generation, and blockchain cross-chain verification technology, it can achieve millisecond-level analysis of load characteristics, tamper-proof real-time settlement, and adaptive closed-loop monitoring of abnormal electricity consumption behavior, taking into account both the peak-shaving efficiency of the power grid and the fairness of electricity consumption for users.

[0027] The method of this embodiment specifically includes:

[0028] Obtain continuously sampled current waveform data on the power line and generate an original electrical parameter sequence at a frequency of not less than 100 sampling points per second;

[0029] Receiving the original electrical parameter sequence, and generating a feature vector carrying a waveform fingerprint by using differential exponential smoothing processing on a TSA chip based on a sampling time stamp of the original electrical parameter sequence;

[0030] Specifically, the raw data of the voltage and current waveforms are acquired through high-speed sampling at a rate of not less than 100 times per second, and the differential exponential smoothing algorithm is used to remove high-frequency noise while retaining the local mutation characteristics of the waveform, such as harmonic phase jumps or steep waveforms at the moment of equipment start and stop. The differential calculation obtains the gradient value by weighted averaging the previous and next data windows to generate a fingerprint feature vector containing waveform trends and details. Compared with traditional mean filtering, this method is more suitable for the accurate expression of short-term impact loads. Differential exponential smoothing is a dynamic filtering method that combines sliding windows and gradient calculations. It adjusts the smoothing intensity according to the time weight of the data points, and highlights the waveform change trend when reducing noise. The waveform fingerprint is a set of characteristic parameters extracted from the waveform segment formed by multiple samplings, which can uniquely characterize a specific type of electrical appliance or abnormal state.

[0031] Among them, the sliding window The value range is 50-200 sampling points, and the smoothing coefficient The dynamic adjustment formula is

[0032] ,

[0033] In the formula is the standard deviation of the power differential in the current window, is the mean standard deviation of the historical window, To adjust the parameters, you can preset ;

[0034] Backward difference gradient value The calculation formula is

[0035] ,

[0036] Where, for The current value at the moment, is the sampling interval, which can be set to 0.01 seconds;

[0037] The calculation formula for weighted smoothing output is:

[0038] .

[0039] Outputting the equipment load type classification result through a lightweight convolutional neural network according to the eigenvector spectrum distribution characteristics;

[0040] Specifically, a fast Fourier transform is performed on the feature vector to extract the power spectral density distribution in the 0.1-2kHz range, identifying the spectrum differences between constant impedance and electronic switching loads. The standardized feature vector is then fed into a pre-trained lightweight convolutional neural network model. This network uses depthwise separable convolution to reduce computational complexity, enabling efficient classification of nonlinear devices such as variable-frequency air conditioners and charging stations. Classification results are updated every 50 milliseconds. Lightweight convolutional neural networks are deep learning models optimized by reducing the number of network layer parameters and employing low-rank decomposition techniques. They are suitable for real-time inference in embedded devices with limited computing power.

[0041] Among them, the lightweight convolutional neural network architecture includes: the input layer receives the spectrum vector with a dimension of 128×1; convolution layer 1 consists of four groups of depthwise separable convolutions with a kernel size of 3×1, a stride of 2, and 32 output channels; the attention module uses a compression-excitation network (SENet) for channel attention weight calculation with a compression ratio of r=16; convolution layer 2 consists of two groups of low-rank decomposition convolutions, which decompose the original 5×1 kernel into two layers of 5×3 and 3×1, reducing the number of parameters by 40%; the fully connected layer has the number of output nodes corresponding to N known device types, with Softmax activation; the training dataset contains harmonic feature data of typical industrial equipment in the GB / T 17215 standard, such as welding machines and inverters.

[0042] Receiving the load type classification result, and generating dynamic rate parameters in combination with the power grid peak demand index;

[0043] receiving the dynamic rate parameters and weighting the user's historical credit score to generate a hybrid bill including a transient overload coefficient;

[0044] Specifically, based on the real-time peak demand index released by the power grid dispatching center, such as regional line load factors and transformer temperature overload warnings, a power supply cost adjustment factor is calculated. This factor, combined with a user credit score database, generates differentiated rates for each time period. The user credit score database can include historical payment records and the number of defaults. For transient equipment overloads detected in the load classification results, such as when the instantaneous current exceeds 80% of the nominal value when the air conditioner compressor starts, an overload factor is introduced to weightedly bill the excess power, ultimately generating a multi-layered billing structure consisting of a base fee, a peak demand surcharge, and a credit discount. The peak demand index is a quantitative indicator reflecting the current supply and demand balance of the power grid. It is dynamically updated based on real-time load fluctuations and is used to adjust the time-zone electricity price ratio. The transient overload factor is a punitive billing weight for short-term abnormal energy consumption behavior of the equipment, and its value is determined by the product of the degree of overload and the duration.

[0045] Parsing the hash value of the hybrid bill and combining it with the blockchain timestamp to form a cross-chain transaction message;

[0046] After detecting the cross-chain transaction message, the account deposit status is analyzed to activate the three-stage settlement smart contract;

[0047] Specifically, the hash value of the bill's core fields and the latest timestamp from the grid's blockchain main chain are embedded in the transaction message. Cross-chain verification with the user-side microgrid consortium chain ensures the immutability and temporal consistency of the transaction data. When the three-stage settlement smart contract is triggered, the current fee is preferentially deducted from the user's prepaid account. When the balance falls below a threshold, a portion of the prepaid amount is frozen and an alert is issued. If the critical value is reached, a remote power-off command is automatically sent to the meter circuit breaker module. Cross-chain verification involves relay nodes confirming the legitimacy of transactions across different blockchain networks to prevent double-spending attacks or counterfeit bills. The three-stage settlement system incorporates a multi-level fund management mechanism consisting of pre-withholding fees, freezing buffers, and emergency power outages, with response strategies triggered in layers based on balance status.

[0048] Capturing the load fluctuation data after deduction according to the three-stage settlement smart contract and generating abnormal fluctuation marks by comparing the data with the equipment shutdown characteristic curve;

[0049] Receive the abnormal fluctuation mark, update the federated learning model parameters and write them into the tamper-proof ledger to complete the closed-loop processing.

[0050] Specifically, after the deduction is executed, the current fluctuation curve is continuously monitored. The actual waveform is matched against the attenuation slope and oscillation frequency parameters in the standard feature library when the device is normally shut down. If the deviation exceeds 15%, it is marked as an abnormal event and triggers an on-site inspection signal. At the same time, the anonymized abnormal data is uploaded to the federated learning center node, and the load classification model is updated in conjunction with multiple meter terminals to optimize the sensitivity of the next detection. Federated learning updates improve the anomaly recognition ability of each terminal model through distributed collaborative training while protecting user privacy, avoiding the security risks of centralized data collection.

[0051] For example, this method forms a refined characterization of the equipment's electricity consumption behavior through high-precision waveform sampling and feature extraction, implements dynamic rate decisions based on the grid's real-time peak-shaving needs and the user's credit status, uses blockchain cross-chain verification to ensure the non-repudiation of billing data throughout the entire process, and achieves system self-optimization through anomaly detection and model iteration.

[0052] Optionally, the generating of a feature vector carrying a waveform fingerprint by using differential exponential smoothing processing on a TSA chip based on the sampling timestamp of the original electrical parameter sequence includes:

[0053] receiving the original electrical parameter sequence, extracting power differential values ​​of adjacent points and compressing them into a Huffman dictionary;

[0054] Specifically, after receiving the original electrical parameter sequence, the system first calculates the power differential values ​​of adjacent sampling points. These differential values ​​reflect the instantaneous rate of current change. Huffman coding is then used to compress the differential values. An optimal prefix code table is constructed based on the frequency of occurrence of different values ​​to form a compact Huffman dictionary. This process effectively reduces transmission bandwidth consumption while retaining the timing characteristics of power mutations, such as current step data at the moment of motor startup or switching. The Huffman dictionary is a variable-length coding structure based on the statistical characteristics of power differentials. High-frequency small-amplitude fluctuations are represented by short codes, while low-frequency large-amplitude mutations retain long code configurations, achieving a balance between data volume compression and feature integrity.

[0055] Parsing the Huffman dictionary, and superimposing the voltage phase mutation amount to generate a feature summary with a check bit;

[0056] Specifically, after parsing the compressed data from the Huffman dictionary, voltage phase mutations, such as zero-crossing offset and harmonic phase difference, are superimposed to enhance waveform characterization. To ensure data credibility, a cyclic redundancy check (CRC) code is added as an additional bit to generate a self-verifying feature summary. This verification mechanism can detect data tampering or loss during transmission and prevent feature distortion caused by signal interference. Voltage phase mutations are abnormal phase offsets of the voltage waveform within a specific cycle and are used to capture phase distortion caused by capacitor switching or power electronic equipment switching.

[0057] Nonlinear filtering is performed based on the feature summary, a denoised feature vector is output, and the original waveform is irreversibly restored.

[0058] Specifically, nonlinear filtering is performed based on the feature summary, using a piecewise function to differentially suppress noise of varying amplitudes. Small background noise is significantly attenuated, while large mutation signals, such as equipment startup and shutdown, are enhanced. The filtered output feature vector is encrypted using a one-way hash function, ensuring that even if the feature vector is obtained, the original current waveform cannot be deduced, thus achieving dual protections for user privacy and data security. Nonlinear filtering is a data processing method that dynamically adjusts the filter strength based on the signal amplitude. Different suppression strategies are used for small noise and significant mutation signals to avoid the signal edge blurring caused by traditional linear filtering.

[0059] For example, data storage efficiency is optimized through power differential compression, feature representation reliability is improved by combining phase mutation enhancement and verification mechanisms, and high-fidelity denoising is achieved through nonlinear filtering. The resulting irreversible feature vector not only retains the fingerprint of key electricity consumption behavior, but also effectively prevents the risk of data leakage. This method significantly improves the real-time processing capability of the meter edge computing device, enhances the anti-interference ability of load feature analysis, and meets privacy protection requirements, providing a high-precision and high-security data foundation for subsequent dynamic rate calculation and anomaly detection. Its beneficial effects include improving the efficiency of waveform feature generation, strengthening the protection capability of sensitive data, and ensuring the robustness of feature extraction in complex electricity consumption scenarios.

[0060] Optionally, outputting a device load type classification result through a lightweight convolutional neural network according to the characteristic vector spectrum distribution characteristics includes:

[0061] Preprocessing the Fourier transform result of the feature vector to generate a device fingerprint mapping dictionary;

[0062] Specifically, the Fourier transform results of the feature vectors are preprocessed. Key frequency band parameters are extracted based on the spectral energy distribution and fundamental frequency harmonic components. The typical spectral characteristics of different device types are clustered and encoded into a hash index structure to form a device fingerprint mapping dictionary. This dictionary serves as a reference library for load identification, supporting fast comparison and retrieval. The device fingerprint mapping dictionary is a device feature index table constructed through Fourier transform and data clustering, storing the power spectrum characteristics of different devices in specific frequency bands in hash form.

[0063] receiving real-time current harmonic distribution data, traversing the mapping dictionary and outputting a device identification result;

[0064] Specifically, the system receives real-time current harmonic distribution data, calculates the amplitude and phase differences of its main harmonic components, and traverses the known device feature templates stored in the device fingerprint mapping dictionary. A cosine similarity algorithm is used to measure the degree of match between the real-time data and the template. When the match exceeds a set threshold, the corresponding device type is output, enabling rapid classification of electronic devices such as variable-frequency air conditioners and LED drivers. Real-time current harmonic distribution data is a collection of multiple harmonic components generated by electrical equipment during operation. A Fourier transform is used to separate the spectrum vector consisting of the fundamental wave and high-frequency harmonic components such as the third and fifth harmonics.

[0065] Analyze unrecognized power curve patterns and write them into the knowledge base to be optimized.

[0066] Specifically, when analyzing unrecognized power curve patterns, characteristic parameters such as transient startup waveforms and steady-state operating harmonic distortion are extracted, annotated with timestamps and environmental parameters, such as grid voltage fluctuations, and then written into the knowledge base to be optimized. This knowledge base serves as a data source for offline training, enabling expansion of device type recognition capabilities during subsequent model iterations. The knowledge base to be optimized is a structured database specifically designed to store the characteristics of unrecognized power usage patterns. New device labels are generated through manual labeling or automated clustering for model retraining.

[0067] For example, by constructing a device fingerprint dictionary to achieve efficient matching of known loads, combined with dynamic identification of harmonic distribution and autonomous accumulation of unknown patterns, a scalable load classification system is formed. This method can improve the coverage of load classification, promptly capture the characteristics of new electrical devices, and optimize model performance through continuous iteration, thereby reducing manual intervention and enhancing the system's adaptability to new scenarios. Its beneficial effects include improving device classification accuracy, reducing missed detection rates, and effectively supporting the power grid's needs for refined management of diverse loads.

[0068] Optionally, generating a dynamic rate parameter in combination with a power grid peak-shaving demand index includes:

[0069] Parse the JSON data of the power grid peak demand index and generate the basic rate matrix;

[0070] Specifically, the system parses the JSON-formatted data of the power grid's peak demand index, extracting key parameters such as regional load factor and renewable energy consumption rate. Based on pre-defined rules for dividing peak and valley periods, it generates a two-dimensional matrix containing the base electricity price and peak shaving surcharge rates for different time periods. This matrix dynamically maps the current power grid supply and demand, providing an initial framework for subsequent rate adjustments. The basic rate matrix is ​​a price structure model constructed based on the dual dimensions of time and load type, with rows representing time periods and columns corresponding to benchmark price units for different load types.

[0071] Correcting the weight distribution of the basic rate matrix based on the credit adjustment coefficient generated by the user's historical electricity usage behavior;

[0072] Specifically, the system reads a user's historical electricity usage data set, calculates a credit score based on the number of delayed payments and excessive electricity usage violations, and normalizes this score into a credit adjustment coefficient. This coefficient is applied as a weighted factor to the corresponding column weights in the base rate matrix, reducing the peak-shaving surcharge rate for users with good credit and implementing a premium policy for high-risk users, thus forming a preliminary adjusted rate distribution. The credit adjustment coefficient is a dynamic weight parameter that reflects the user's credit rating. An algorithm quantifies the user's historical behavior into a rate discount or penalty ratio.

[0073] The weight distribution is modified based on the loss compensation factor of the real-time temperature of the distribution transformer, and the final dynamic rate parameter is output.

[0074] Specifically, real-time temperature data from the distribution transformer windings is collected and a loss compensation factor is calculated based on the loss model under rated operating conditions. The compensation factor corresponding to periods of temperature exceeding the standard is added to the row weights of the basic rate matrix. This apportions the additional O&M costs incurred by equipment overload, and outputs a multi-dimensional dynamic rate parameter that integrates grid peak regulation, user credit, and equipment loss. The loss compensation factor is a rate correction coefficient to compensate for additional energy loss caused by overload or aging of distribution equipment. It is dynamically calculated based on the deviation between the real-time temperature and the standard operating conditions.

[0075] For example, a multidimensional dynamic rate model is constructed by integrating data on grid peak-shaving demand, user credit assessments, and distribution equipment losses. This method enables refined rate regulation, responding to real-time grid load fluctuations while taking into account differences in user behavior and equipment health, thereby promoting the efficient allocation of power resources. Its beneficial effects include improving the fairness and transparency of rate setting, incentivizing users to optimize their electricity usage habits, reducing the risk of grid equipment overload, and supporting the coordinated optimization of power system economics and security.

[0076] Optionally, parsing the hash value of the hybrid bill and combining the blockchain timestamp to form a cross-chain transaction message includes:

[0077] Parse the JSON structure of the mixed billing invoice to separate the business logic data flow and the capital flow;

[0078] Specifically, the JSON structure is a lightweight data exchange format that uses key-value pairs to hierarchically store billing information such as rate parameters, electricity consumption, and credit discounts. The business logic data flow contains billing rules associated with electricity pricing strategies and load types, while the capital flow involves actual settlement amounts and payment path information. This step is separated by structured fields to ensure the independence and traceability of business configuration and capital transactions during subsequent blockchain processing. Hybrid billing is a multi-layered billing data set generated by integrating dynamic rates, credit scores, and transient overload factors. The JSON structure is a nested tree-like data format that supports efficient parsing and cross-platform transmission. The business logic data flow is a collection of non-financial information that represents electricity price calculation rules and electricity usage strategies. The capital flow is a record of financial operations that includes transaction amounts, account identifiers, and settlement timestamps.

[0079] Push the separated business logic data flow unidirectionally to the alliance chain node for pre-confirmation;

[0080] Specifically, consortium chain nodes are blockchain network nodes composed of authorized members such as power grid companies and regulatory agencies. The legitimacy of business logic is verified through a consensus mechanism. The pre-confirmation process checks data format compliance and permission validity, for example, verifying that the user's credit adjustment factor is within the contractually agreed range. One-way push uses asymmetric encryption to ensure irreversible data transmission from the meter to the blockchain, preventing tampering. Consortium chain nodes are server clusters of authorized blockchain members that participate in business logic verification. Pre-confirmation serves as a preliminary review of data legitimacy before formal accounting, preventing invalid transactions from being posted on the blockchain.

[0081] After receiving the pre-confirmation result, the hash value of the fund flow is permanently written into the genesis block of the transaction chain.

[0082] Specifically, the transaction chain is a blockchain branch dedicated to fund settlement. The genesis block is the initial block of the chain, recording an unalterable hash value fingerprint. The fund flow hash is a unique digital summary generated using the SHA-3 algorithm. Once written, it forms a chain link with the timestamp and hashes of adjacent blocks, ensuring the auditability of transaction records throughout their lifecycle. The transaction chain is a dedicated financial blockchain network independent of the business chain, optimizing the efficiency of high-frequency trading. The genesis block is the first block of the blockchain network, storing the initial state of the chain and core verification rules. Permanently writing the hash value is an irreversible operation that cryptographically binds the transaction summary to the blockchain.

[0083] Among them, the cross-chain transaction verification mechanism is the business logic chain, that is, the consortium chain adopts the improved PBFT consensus, and the node response threshold is set to 2 / 3 majority; the transaction chain genesis block defines the hash anchoring rule: the hash value of the generated new block is synchronized to the Hyperledger Fabric main chain through the Merkle tree every 10 minutes; the smart contract trigger condition can be that when the user balance is less than the second threshold, the Chaincode function FreezeAccount(bytes32 userID) is called and an SGX encrypted log is generated.

[0084] For example, a cross-chain transaction architecture enables the separated and trusted processing of business logic and funds settlement. When parsing bills, a JSON structure separation mechanism ensures the independent flow of business rules and financial data. Consortium chain pre-confirmation ensures compliance with business policies, and the transaction chain's genesis block writes solidify the fingerprint of fund operations. Technically, combining the distributed ledger nature of blockchain with a one-way data pipeline design effectively isolates the impact of business changes from transaction history, forming a tamper-proof, full-link evidence chain. Beneficial effects include: improving the credibility and non-repudiation of billing transactions and shortening the verification cycle for cross-institutional settlements; decoupling business and capital flows enhances system scalability and supports dynamic adjustments to rate policies without interfering with existing transaction records; and a genesis block anchoring mechanism strengthens transaction traceability, providing irrefutable audit evidence for electricity bill disputes. This approach ultimately achieves a highly secure and efficient real-time electricity bill settlement system that balances the compliance requirements of grid operations with the protection of user rights.

[0085] Optionally, the analyzing the account pre-deposit status to activate the three-stage settlement smart contract includes:

[0086] When the pre-deposit amount in the account is greater than or equal to a preset first threshold, the full amount deduction module is activated and a unique deduction voucher is generated;

[0087] Specifically, the full deduction module is a program component that directly deducts the full amount payable from the user's account based on the electricity bill amount, and is suitable for scenarios where the user has sufficient funds. The unique deduction voucher is an electronic receipt containing the transaction serial number, timestamp, and encrypted signature, which ensures the immutability and uniqueness of the transaction record. This step prevents the accumulation of arrears through immediate deductions, and at the same time generates a unique voucher that provides a traceable basis for subsequent inquiries and disputes. The full deduction module is a software functional unit that performs a real-time, one-time deduction of the electricity bill amount. The unique deduction voucher is a unique encrypted data packet that identifies a single transaction and is used to verify the authenticity of the transaction.

[0088] When the account deposit amount is less than the first threshold and greater than or equal to a preset second threshold, a phased withholding agreement is generated and the fund pool is locked;

[0089] Specifically, a phased withholding agreement is a temporary contract that deducts the amount to be deducted in batches according to a preset period, such as by the hour or by electricity consumption. A fund pool lock freezes a portion of the balance in a user's account as a performance bond to ensure the reliability of subsequent deductions. This step alleviates the electricity consumption pressure of users with short-term cash shortages through flexible deduction rules, while also mitigating cash flow risks for power grid companies. A phased withholding agreement is a temporary fund management solution that allows for installment deductions. A fund pool lock restricts the liquidity of a specific amount in an account to ensure the feasibility of phased deductions.

[0090] When the pre-deposit amount in the account is less than the second threshold, the relay control sequence is triggered and the electricity usage authority is frozen.

[0091] Specifically, the relay control sequence is a prioritized set of hardware switching instructions that freeze electricity usage rights by disconnecting or limiting power supply lines. For example, this prioritizes disconnecting non-essential load circuits or entering low-power mode. This step directly blocks electricity use for users in arrears through hardware-level control, ensuring the collection of electricity bills and the safety of grid operations. The relay control sequence is an ordered set of instructions that controls the switching state of relays within the meter. Freezing electricity usage rights terminates a user's continued use of electricity by physically disconnecting the power circuit or limiting the power supply.

[0092] For example, a hierarchical response mechanism is used to implement tiered fund management, dynamically adjusting settlement strategies based on account balance status. Its core principle is to divide credit risk levels by preset thresholds. When the balance is above the first threshold, settlement is quickly completed to ensure fund recovery. When the balance is in the middle range, phased deductions are enabled to balance user experience and grid rights. When the balance is below the second threshold, permission freezes are enforced to eliminate the risk of arrears. The technical effect is reflected in the multi-level prevention of capital chain breaks caused by arrears, enhancing the proactive control capability of electricity fee recovery; phased agreements reduce users' short-term financial pressure and improve electricity continuity; and hardware-level permission freezes ensure the mandatory and timely management of arrears. The overall approach integrates flexible billing and rigid control, avoiding the drawbacks of frequent power outages in the traditional prepaid model while strengthening the security of electricity fee transactions through hierarchical responses, ultimately achieving dual optimization of user rights protection and grid enterprise risk control.

[0093] Optionally, triggering the relay control sequence and freezing the power usage permission includes:

[0094] Receive grid load capacity data within the geo-fence and calculate a preset allowable power gradient;

[0095] Specifically, the geofence is the boundary of the virtual power supply area divided according to the distribution network topology, and the grid load capacity data includes parameters such as the remaining capacity of the transformer in the current area, the upper limit of the line current, and the load fluctuation rate. The preset allowable power gradient is calculated based on real-time data to determine the maximum power supply reduction threshold available to users in arrears during a specific period of time. For example, when the regional load rate exceeds the limit, the gradient calculation is used to limit the rate of decrease of the user's instantaneous power consumption to avoid voltage drops causing equipment damage. The geofence is the geographical boundary of power supply management delineated by grid node coordinates and load distribution data. The grid load capacity data is a set of dynamic parameters that characterize the current maximum power that the distribution equipment can carry. The preset allowable power gradient is a power adjustment rate limit value set based on grid security constraints.

[0096] Generate a device load reduction curve based on the preset allowable power gradient and output a relay control instruction queue;

[0097] Specifically, the device load reduction curve is an execution plan for reducing the power supply in stages according to a time series, generating a stepped power reduction trajectory based on the gradient threshold. For example, when the gradient limit is 5% of the rated power per second, a command queue is generated to gradually cut off non-critical loads every 10 seconds, prioritizing basic lighting power. The relay control command queue contains the switching action timing, target circuit number, and execution delay parameters to ensure that the meter relay module accurately executes power regulation actions. The device load reduction curve is a timing control plan for reducing the power supply in stages. The relay control command queue is a collection of hardware switching operation instructions arranged by priority and time order.

[0098] Burn the relay control instruction queue into the meter firmware and erase the original control code.

[0099] Specifically, the burning operation involves writing compiled machine code instructions into the read-only memory area of ​​the meter control chip through a secure channel, replacing the original control logic. Erasing the original control code clears the storage area of ​​the meter's original relay driver to prevent instruction conflicts or malicious code injection. For example, page erase technology in FLASH memory can be used to ensure that the new instruction queue has exclusive hardware execution rights. Burning is an irreversible operation that permanently stores control instructions in the meter's memory through a programming interface. Erasing the original control code completely deletes the data in the storage area of ​​the meter's original relay driver.

[0100] For example, refined power supply control is achieved for arrears scenarios, and its core principle lies in the dynamic binding of the real-time capacity of the power grid with the management of user power usage rights. The grid load bottleneck area is precisely located through geographic fencing, and a progressive load reduction strategy is generated in combination with the preset power gradient, which not only avoids the negative impact of direct power outages, but also protects the safety of distribution equipment through smooth power adjustment. The technical effect is reflected in the integration of regional power grid status perception and adaptive control capabilities, giving priority to ensuring the continuity of power supply to critical loads; the timed execution mechanism of the relay instruction queue improves the accuracy and safety of power regulation; the firmware burning and code erase operations ensure the irreversibility and anti-interference ability of the control logic. Ultimately, a flexible power outage management system is formed that takes into account grid stability, equipment protection and user experience, significantly improving the intelligence level and execution reliability of power regulation in arrears scenarios.

[0101] Optionally, capturing the load fluctuation data after deduction according to the three-stage settlement smart contract and generating an abnormal fluctuation mark by comparing the data with the device shutdown characteristic curve includes:

[0102] Capture post-charge load fluctuation data to extract power attenuation slope;

[0103] Specifically, the power decay slope is the rate of change of the current or power over time after the device is shut down. It is obtained by sampling the millisecond-level current waveform data after the deduction and calculating the first-order derivative of its decline curve or the slope value of the fitted linear phase. This slope reflects the physical response characteristics of the device when it is powered off, such as the difference between the inertial coasting of a motor and the instantaneous power outage of an electronic device, providing dynamic feature input for anomaly detection. The power decay slope is a quantitative indicator that characterizes the speed at which energy consumption decreases when the device is powered off. Load fluctuation data is a record of the instantaneous changes in current or power on the power grid circuit after the electricity bill is settled.

[0104] Comparing the power attenuation slope with a reference value in a device feature knowledge base to generate a deviation coefficient vector;

[0105] Specifically, the device feature knowledge base is a feature database that pre-stores the standard power attenuation parameters of various electrical appliances when the power is normally cut off, including typical slope ranges, time-series vibration patterns, etc. By calculating the absolute difference and fluctuation variance between the attenuation slope collected in real time and the corresponding device benchmark value in the knowledge base, a deviation coefficient vector reflecting the degree of deviation is generated. This vector contains multi-dimensional abnormal indicators such as slope deviation and phase offset, providing basic data for subsequent correlation analysis. The device feature knowledge base is a reference library of normal power-off features of equipment based on historical data or experiments. The deviation coefficient vector is a set of numerical indicators that describe the difference between real-time data and benchmark features.

[0106] Analyzing the line impedance mutation parameters in the deviation coefficient vector and constructing an anomaly score matrix associated with multi-source data;

[0107] Specifically, the line impedance mutation parameter represents the rapid change in impedance in a power grid line due to switching operations, poor contact, or illegal access. It is calculated by measuring parameters such as the voltage and current phase difference and harmonic distortion rate during a power outage. The impedance mutation data is analyzed for spatiotemporal correlation with other indicators in the deviation coefficient vector to construct an anomaly score matrix. Each element in this matrix represents a comprehensive weighted score for different anomaly types, which is used to quantify the likelihood and severity of an anomaly event. The line impedance mutation parameter is a physical quantity that reflects the abnormal instantaneous change in impedance in a power grid line. The anomaly score matrix is ​​a quantitative risk assessment model formed by weighted fusion of multiple anomaly indicators.

[0108] When the abnormal score matrix exceeds the preset security threshold, the fund transfer of the related transaction is locked and a manual review is initiated, and an audit report containing abnormal fluctuation marks is generated.

[0109] Specifically, the preset safety threshold is a dynamic threshold parameter set based on historical abnormal case statistics and power grid safety regulations. By monitoring the weighted sum of each dimension in the anomaly score matrix in real time to see if it exceeds the threshold, the automatic freezing of funds transfers for suspicious transactions is triggered. The abnormal time period, device type, and score details are recorded in the audit report, creating a traceable abnormal event archive that provides a basis for subsequent verification and model optimization. The preset safety threshold is the graded alert value that triggers the risk control mechanism. The audit report is an audit document that records the characteristics of the abnormal event and the handling measures.

[0110] For example, an anomaly detection model is established by capturing the dynamic power consumption characteristics when the equipment is shut down. Its core principle lies in the collaborative analysis mechanism that integrates the physical characteristics of the equipment and the status of the power grid line. By extracting the power attenuation slope and comparing it with the standardized knowledge base, the rationality and consistency of the equipment's power-off behavior are quantified; a multi-source correlation scoring model is constructed in combination with the line impedance parameters to effectively distinguish normal fluctuations from abnormal scenarios such as power theft and equipment failure. The technical effect is reflected in improving the recognition sensitivity of hidden abnormal power consumption behaviors and avoiding false alarms and missed alarms caused by single feature detection; the dynamic threshold mechanism realizes accurate risk classification and control, reducing the workload of manual verification; the audit report generation function improves the abnormal event tracing chain and strengthens the anti-power theft and equipment health management capabilities. Ultimately, an intelligent closed-loop abnormal monitoring system is formed, which significantly improves the security and fairness of the electricity bill settlement environment.

[0111] Optionally, updating the federated learning model parameters and writing them into the tamper-proof ledger to complete the closed-loop processing includes:

[0112] Receive abnormal fluctuation marks and corresponding audit reports, use the period of occurrence of the abnormal fluctuation marks as an index, extract feature vectors from edge nodes to generate data fingerprints, and output global optimization requirements;

[0113] Specifically, the data fingerprint is a unique identifier generated by compressing the feature vector using a hash algorithm, which is used to quickly index and match the electricity consumption feature data for a specific time period. The global optimization requirement is the model update instruction generated based on the anomaly analysis results, marking the model modules that need to be optimized and the enhanced anomaly detection dimensions. This step accurately locates the associated data through the abnormal time period index and generates the target parameters for the collaborative optimization of the federated learning system. The data fingerprint is a feature summary formed by encrypting the electricity consumption characteristics of a specific time period. The global optimization requirement is the target instruction set that triggers the update of the federated learning model parameters. The edge node is a local computing device deployed at the meter terminal or regional gateway.

[0114] According to the global optimization requirements, the convolution kernel attention weight matrix is ​​reconstructed, and the old parameter hash value is written into the tamper-proof ledger;

[0115] Specifically, the convolution kernel attention weight matrix is ​​a set of parameters used in lightweight neural networks to distribute spatial attention to input features. The reconstruction operation adjusts the weights of different feature channels based on abnormal fluctuation characteristics, for example, enhancing the detection weight of harmonic distortion parameters. The old parameter hash value is a parameter fingerprint generated using a one-way hashing algorithm. This is written to the blockchain's tamper-proof ledger to form a version traceability baseline, ensuring the integrity and auditability of the model iteration process. The convolution kernel attention weight matrix is ​​a network parameter structure used to distribute spatial importance of features. The old parameter hash value is an encrypted summary record of the parameter set before the model update. The tamper-proof ledger is an immutable data storage system implemented by blockchain technology.

[0116] The updated weight matrix is ​​compiled into a binary instruction set carrying a version number that marks abnormal fluctuations and encrypted and transmitted to the designated meter cluster.

[0117] Specifically, the binary instruction set is a compiled low-level machine code that is adapted to the instruction set architecture of the meter terminal processor. The abnormal fluctuation mark version number is a unique identifier embedded in the update file and is used to trace the abnormality type and occurrence time corresponding to the model optimization. Encrypted transmission uses an asymmetric encryption algorithm to protect the communication channel and ensure that the update instructions are not maliciously intercepted or tampered with during transmission. The binary instruction set is a set of machine language codes that can be directly embedded in the meter firmware for execution. The abnormal fluctuation mark version number is a model update identifier associated with a specific abnormal event. The designated meter cluster is a grouped set of terminal devices that need to synchronize model updates.

[0118] For example, by constructing a dynamic update mechanism for federated learning based on abnormal events, the continuous optimization of the electricity consumption monitoring model can be achieved. Its core principle is to transform abnormal fluctuation characteristics into model optimization signals, and to jointly improve the feature recognition capability of multiple terminals while protecting local privacy data. The technical effect is reflected in the enhancement of the model's detection sensitivity to new electricity theft modes and equipment failures through elastic weight adjustment; the version traceability function of the tamper-proof ledger ensures the credibility of model iteration and prevents parameter tampering attacks; the encrypted compilation and directional transmission mechanism improves the security of model updates and device compatibility. This method forms a closed-loop optimization system with simultaneously enhanced privacy security and technical attack and defense capabilities, providing smart meters with long-term self-evolving anomaly monitoring capabilities, and significantly improving the intelligence level of power grid anti-fraud and equipment health management.

[0119] Among them, the privacy protection mechanism and parameter update rules are: each terminal node uses its local anomaly dataset Calculating the loss function Relative to model parameters Gradient

[0120] ;

[0121] To ensure differential privacy, for each gradient Add Gaussian noise , where is the standard deviation of the noise, which can be set to 0.01. The mean is 0 and the variance is Gaussian noise, is the identity matrix.

[0122] Privacy Budget for Differential Privacy and The relationship is usually estimated by the following formula:

[0123] ,

[0124] Where, is the L2 sensitivity of the gradient, which can usually be set to 1. is the number of iterations, For a small probability, usually set to 0.00001 or less, you can set is 0.01, satisfying =1.5 privacy budget constraint.

[0125] The server collects the noise gradients of all terminal nodes and updates the parameters. ,have

[0126] ,

[0127] Where, is the learning rate, which controls the step size of parameter updates, is the total number of terminal nodes, is the gradient The L2 norm of , used for normalization.

[0128] Based on the same inventive concept, Figure 2 As shown, the present invention also provides a real-time electricity fee settlement and intelligent fee deduction system for a smart meter, the system comprising:

[0129] The data acquisition module is used to obtain the continuously sampled current waveform data on the power line and generate the original electrical parameter sequence at a frequency of not less than 100 sampling points per second;

[0130] A feature vector generation module is configured to receive the original electrical parameter sequence and generate a feature vector carrying a waveform fingerprint by using differential exponential smoothing processing on a TSA chip based on a sampling time stamp of the original electrical parameter sequence;

[0131] An equipment load classification module, configured to output equipment load type classification results through a lightweight convolutional neural network based on the spectral distribution characteristics of the feature vector;

[0132] A rate parameter generation module, configured to receive the load type classification result and generate dynamic rate parameters in combination with the power grid peak demand index;

[0133] A billing bill generation module, configured to receive the dynamic rate parameters and weight the user's historical credit score to generate a hybrid bill including a transient overload coefficient;

[0134] A transaction message generation module, configured to parse the hash value of the hybrid bill and combine it with the blockchain timestamp to form a cross-chain transaction message;

[0135] A smart contract generation module is used to monitor the cross-chain transaction message, analyze the account deposit status and activate the three-stage settlement smart contract;

[0136] An abnormal fluctuation mark generation module is used to capture the load fluctuation data after deduction according to the three-stage settlement smart contract, and generate an abnormal fluctuation mark by comparing it with the equipment shutdown characteristic curve;

[0137] The feedback recording module is used to receive the abnormal fluctuation mark, update the federated learning model parameters and write them into the tamper-proof ledger to complete the closed-loop processing.

[0138] It should be noted that the electrical connections between the above-mentioned units do not necessarily mean direct connections of lines. Indirect connections are applicable to the embodiments of the present invention as long as the purpose of the present invention is achieved. The above description is only an exemplary embodiment of the present invention and is not intended to limit the scope of the present invention.

[0139] That is, any equivalent changes and modifications made according to the teachings of the present invention are still within the scope of the present invention. Those skilled in the art will readily conceive of other embodiments of the present invention after considering the disclosure of the specification and practical truths. This application is intended to cover any variations, uses, or adaptations of the present invention that follow the general principles of the present invention and include common knowledge or customary technical means in the art not described herein.

Claims

1. A method for real-time electricity bill settlement and intelligent deduction of smart meters, characterized in that: The method comprises: Obtain continuously sampled current waveform data on the power line and generate an original electrical parameter sequence at a frequency of not less than 100 sampling points per second; The method comprises the following steps: receiving the original electrical parameter sequence, and using differential exponential smoothing processing on a TSA chip based on the sampling timestamp of the original electrical parameter sequence to generate a feature vector carrying a waveform fingerprint; receiving the original electrical parameter sequence, extracting power differential values ​​of adjacent points and compressing them into a Huffman dictionary; parsing the Huffman dictionary, superimposing voltage phase mutations to generate a feature summary with a check bit; performing nonlinear filtering based on the feature summary, outputting a denoised feature vector, and irreversibly restoring the original waveform; Outputting the equipment load type classification result through a lightweight convolutional neural network according to the eigenvector spectrum distribution characteristics; Receive the load type classification result and generate dynamic rate parameters in combination with the power grid peak demand index; including: parsing the JSON data of the power grid peak demand index to generate a basic rate matrix; modifying the weight distribution of the basic rate matrix based on the credit adjustment coefficient generated by the user's historical electricity consumption behavior; modifying the weight distribution based on the loss compensation factor of the real-time temperature of the distribution transformer, and outputting the final dynamic rate parameters; receiving the dynamic rate parameters and weighting the user's historical credit score to generate a hybrid bill including a transient overload coefficient; Parsing the hash value of the hybrid bill and combining it with the blockchain timestamp to form a cross-chain transaction message; After detecting the cross-chain transaction message, the account deposit status is analyzed to activate the three-stage settlement smart contract; The three-stage settlement smart contract captures post-fee load fluctuation data and compares it with the device shutdown characteristic curve to generate an abnormal fluctuation mark; this includes: capturing the post-fee load fluctuation data to extract the power attenuation slope; comparing the power attenuation slope with the benchmark value of the device feature knowledge base to generate a deviation coefficient vector; parsing the line impedance mutation parameters in the deviation coefficient vector to construct an abnormal score matrix associated with multi-source data; when the abnormal score matrix exceeds a preset safety threshold, locking the funds transfer of the associated transaction and initiating manual review, generating an audit report containing the abnormal fluctuation mark; Receive the abnormal fluctuation mark, update the federated learning model parameters and write them into the tamper-proof ledger to complete the closed-loop processing; including: receiving the abnormal fluctuation mark and the corresponding audit report, using the time period of the abnormal fluctuation mark as the index, extracting the feature vector from the edge node to generate the data fingerprint, and outputting the global optimization requirements; according to the global optimization requirements, reconstructing the convolution kernel attention weight matrix, and writing the old parameter hash value into the tamper-proof ledger; compiling the updated weight matrix into a binary instruction set carrying the abnormal fluctuation mark version number, and encrypting and transmitting it to the designated meter cluster.

2. The method for real-time electricity fee settlement and intelligent deduction of a smart meter according to claim 1, characterized in that: Outputting the equipment load type classification result through a lightweight convolutional neural network according to the characteristic vector spectrum distribution characteristics includes: Preprocessing the Fourier transform result of the feature vector to generate a device fingerprint mapping dictionary; receiving real-time current harmonic distribution data, traversing the mapping dictionary and outputting a device identification result; Analyze unrecognized power curve patterns and write them into the knowledge base to be optimized.

3. The method for real-time electricity fee settlement and intelligent deduction of a smart meter according to claim 1, characterized in that: The step of parsing the hash value of the hybrid bill and combining it with the blockchain timestamp to form a cross-chain transaction message includes: Parse the JSON structure of the mixed billing invoice to separate the business logic data flow and the capital flow; Push the separated business logic data flow unidirectionally to the alliance chain node for pre-confirmation; After receiving the pre-confirmation result, the hash value of the fund flow is permanently written into the genesis block of the transaction chain.

4. The method for real-time electricity bill settlement and intelligent deduction of a smart meter according to claim 1, characterized in that: The three-stage settlement smart contract activated by analyzing the account deposit status includes: When the pre-deposit amount in the account is greater than or equal to a preset first threshold, the full amount deduction module is activated and a unique deduction voucher is generated; When the account deposit amount is less than the first threshold and greater than or equal to a preset second threshold, a phased withholding agreement is generated and the fund pool is locked; When the pre-deposit amount in the account is less than the second threshold, the relay control sequence is triggered and the electricity usage authority is frozen.

5. The method for real-time electricity fee settlement and intelligent deduction of a smart meter according to claim 4, characterized in that: The triggering of the relay control sequence and freezing of the power usage authority includes: Receive grid load capacity data within the geo-fence and calculate a preset allowable power gradient; Generate a device load reduction curve based on the preset allowable power gradient and output a relay control instruction queue; Burn the relay control instruction queue into the meter firmware and erase the original control code.

6. A system for real-time electricity bill settlement and intelligent deduction of smart meters, applying a method for real-time electricity bill settlement and intelligent deduction of smart meters as claimed in any one of claims 1 to 5, characterized in that: The system comprises: The data acquisition module is used to obtain the continuously sampled current waveform data on the power line and generate the original electrical parameter sequence at a frequency of not less than 100 sampling points per second; A feature vector generation module is configured to receive the original electrical parameter sequence and generate a feature vector carrying a waveform fingerprint using differential exponential smoothing on a TSA chip based on the sampling timestamp of the original electrical parameter sequence. The module comprises the following steps: receiving the original electrical parameter sequence, extracting power differential values ​​of adjacent points and compressing them into a Huffman dictionary; parsing the Huffman dictionary, superimposing voltage phase mutations to generate a feature summary with a check bit; performing nonlinear filtering based on the feature summary, outputting a denoised feature vector, and irreversibly restoring the original waveform. An equipment load classification module, configured to output equipment load type classification results through a lightweight convolutional neural network based on the spectral distribution characteristics of the feature vector; A rate parameter generation module is configured to receive the load type classification results and generate dynamic rate parameters in combination with the power grid peak demand index; the module comprises: parsing the JSON data of the power grid peak demand index to generate a basic rate matrix; modifying the weight distribution of the basic rate matrix based on the credit adjustment coefficient generated by the user's historical electricity usage behavior; modifying the weight distribution based on the loss compensation factor of the real-time temperature of the distribution transformer, and outputting the final dynamic rate parameters; A billing bill generation module, configured to receive the dynamic rate parameters and weight the user's historical credit score to generate a hybrid bill including a transient overload coefficient; A transaction message generation module, configured to parse the hash value of the hybrid bill and combine it with the blockchain timestamp to form a cross-chain transaction message; A smart contract generation module is used to monitor the cross-chain transaction message, analyze the account deposit status and activate the three-stage settlement smart contract; An abnormal fluctuation mark generation module is used to capture the load fluctuation data after deduction according to the three-stage settlement smart contract, and generate abnormal fluctuation marks by comparing it with the equipment shutdown characteristic curve; the module includes: capturing the load fluctuation data after deduction to extract the power attenuation slope; comparing the power attenuation slope with the benchmark value of the equipment feature knowledge base to generate a deviation coefficient vector; parsing the line impedance mutation parameters in the deviation coefficient vector to construct an abnormal score matrix associated with multi-source data; when the abnormal score matrix exceeds a preset safety threshold, locking the funds transfer of the associated transaction and initiating manual review, and generating an audit report containing the abnormal fluctuation mark; The feedback recording module is used to receive the abnormal fluctuation mark, update the federated learning model parameters and write them into the tamper-proof ledger to complete the closed-loop processing; it includes: receiving the abnormal fluctuation mark and the corresponding audit report, using the time period of the abnormal fluctuation mark as the index, extracting the feature vector from the edge node to generate the data fingerprint, and outputting the global optimization requirements; according to the global optimization requirements, reconstructing the convolution kernel attention weight matrix, and writing the old parameter hash value into the tamper-proof ledger; compiling the updated weight matrix into a binary instruction set carrying the abnormal fluctuation mark version number, and encrypting and transmitting it to the designated meter cluster.

Images