Multi-agent collaborative auditing system for power bills

Abstract

This invention relates to the field of power data processing technology and discloses a multi-agent collaborative auditing system for power invoices, comprising an intelligent agent for extracting invoice elements and an intelligent agent for estimating power load, connected by communication. The intelligent agent for extracting invoice elements parses the power invoice data object, extracts the time-of-use electricity consumption value and the maximum demand value, and divides the metering time slots according to the settlement cycle to generate a power consumption time-series dataset containing the electricity consumption and duration of each time period. The intelligent agent for estimating power load receives this dataset, calculates the lower limit of the loss product of the square integral of power for each time period, and performs constrained discrete estimation of energy distribution based on the maximum demand value to construct an upper limit of the loss product that satisfies the maximum demand boundary. Then, based on the numerical inclusion relationship between the lower and upper limits of the loss product, a power supply logic discrimination quantity is generated, and an audit conclusion is output accordingly. This invention can effectively identify metering or billing anomalies through multi-dimensional logical verification even in the absence of load curve data.

Description

Technical Field

[0001] This invention relates to the field of power data processing technology, and more specifically, to a multi-agent collaborative review system for power invoices. Background Technology

[0002] In scenarios where electricity is not directly supplied by the power grid, such as industrial parks, commercial complexes, and office buildings, the "overall application, internal transfer" operation model is typically adopted. The power supply entity (such as the property management company or energy management provider) settles accounts with the power grid as the main meter user and supplies electricity to end users through the internal distribution network. In electricity billing in such scenarios, a two-part tariff policy is usually implemented, meaning that the electricity bill consists of a consumption-based charge and a basic charge based on maximum demand (or transformer capacity). At the same time, to promote peak shaving and valley filling, the consumption-based charge usually adopts a time-of-use (TOU) billing mechanism, dividing the day into different periods such as peak, high, flat, and low periods, and applying different electricity price standards to each period.

[0003] Furthermore, due to line losses and transformer losses in the internal power distribution network, these losses should be reasonably allocated or attributed when calculating the electricity consumption and charges of resellers, according to relevant electricity supply and consumption business rules. Studies have shown that variable losses (such as copper losses) in the distribution network are proportional to the square of the current, exhibiting a non-linear variation pattern.

[0004] However, existing electricity bill verification or electricity bill review work faces serious challenges. Traditional electricity bills (such as electricity bills and settlement statements) usually only provide summary data for the settlement period, mainly including the cumulative electricity consumption for each period, the maximum demand for the month, and statistical indicators such as power factor, but often lack continuous high-frequency load curve data.

[0005] Under such sparse data conditions, existing auditing systems can typically only perform simple arithmetic summation and comparison or unit price calculation. When metering clock drift, time period mapping errors, demand window alignment deviations, or inadequate data filling occur, the time-of-use electricity distribution on the invoice may logically be impossible to generate from any actual load sequence that meets the maximum demand limit. For example, under the premise of strictly limited maximum demand, it is objectively impossible to achieve an excessively high cumulative electricity consumption in a certain time period. Existing linear auditing methods struggle to utilize residual nonlinear loss clues in the invoices to identify such hidden logical contradictions, resulting in insufficient accuracy and reliability of the audit results. Summary of the Invention

[0006] This invention provides a multi-agent collaborative review system for electricity bills, which solves the technical problems mentioned in the background art.

[0007] This invention provides a multi-agent collaborative review system for electricity bills, including a bill element extraction agent and an electricity load projection agent connected by communication. The bill element extraction agent is configured to parse the electricity bill data object to extract time-of-use electricity values, maximum demand values, and billing cycle data. Based on the settlement cycle determined by the billing cycle data, it divides the data into multiple metering time slots to generate an electricity consumption time-series dataset containing time-period electricity consumption and time-period duration. The electricity load projection agent is configured to calculate the lower limit of the loss product of the power square integral within each time period based on the electricity consumption time-series dataset. Based on the maximum demand value, it performs constrained discrete projection of the energy distribution within each time period to construct an upper limit of the loss product that satisfies the maximum demand boundary. The system generates a power supply logic discrimination quantity based on the numerical inclusion relationship between the lower limit and the upper limit of the loss product. If the lower limit of the loss product is greater than the upper limit or the power supply logic discrimination quantity exceeds a preset range, it outputs a conclusion of audit failure.

[0008] The beneficial effects of this invention are as follows: When continuous load curves are missing from the electricity consumption settlement voucher data, this invention can construct a loss product envelope interval reflecting changes in line losses by utilizing the inherent logical constraints between the electricity consumption time-series dataset and the maximum demand value. Through collaborative interaction between two intelligent agents, it can effectively identify logical contradictions caused by clock drift, misaligned time period mapping, or data filling errors without relying on external master meter data comparison. This invention solves the technical problem of verifying the authenticity of data based solely on summary data in the context of electricity transfer, significantly improving the accuracy and automation level of electricity cost verification, and ensuring the objectivity and traceability of the evidence chain. Attached Figure Description

[0009] Figure 1 This is a flowchart of a multi-agent collaborative review system for electricity bills according to the present invention; Figure 2 This is a schematic diagram illustrating a scenario implementation of the present invention. Detailed Implementation

[0010] The subject matter described herein will now be discussed with reference to exemplary embodiments. It should be understood that these embodiments are discussed only to enable those skilled in the art to better understand and implement the subject matter described herein, and changes may be made to the function and arrangement of the elements discussed without departing from the scope of this specification. Various processes or components may be omitted, substituted, or added as needed in the examples. Furthermore, features described in some examples may be combined in other examples.

[0011] It should be noted that, unless otherwise defined, the technical or scientific terms used in one or more embodiments of the present invention should have the ordinary meaning understood by one of ordinary skill in the art to which this invention pertains. The terms "first," "second," and similar terms used in one or more embodiments of the present invention do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Terms such as "comprising" or "including" indicate that the element or object preceding the term encompasses the elements or objects listed following the term and their equivalents, without excluding other elements or objects. Terms such as "connected" or "linked" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. Terms such as "upper," "lower," "left," and "right" are used only to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0012] like Figure 1 As shown, a multi-agent collaborative review system for electricity bills includes a bill element extraction agent and an electricity load projection agent connected by communication. The bill element extraction agent is configured to parse the electricity bill data object to extract time-of-use electricity values, maximum demand values, and billing cycle data. Based on the settlement cycle determined by the billing cycle data, it divides the data into multiple metering time slots to generate an electricity consumption time-series dataset containing time-period electricity consumption and time-period duration. The electricity load projection agent is configured to calculate the lower limit of the loss product of the power square integral within each time period based on the electricity consumption time-series dataset. Based on the maximum demand value, it performs constrained discrete projection of the energy distribution within each time period to construct an upper limit of the loss product that satisfies the maximum demand boundary. The system generates a power supply logic discrimination quantity based on the numerical inclusion relationship between the lower limit and the upper limit of the loss product. If the lower limit of the loss product is greater than the upper limit or the power supply logic discrimination quantity exceeds a preset range, it outputs a conclusion of audit failure.

[0013] Preferably, the bill element extraction intelligent agent parses the electricity bill data object to extract the time-of-use electricity consumption value and the maximum demand value, and determines multiple metering time slots according to the settlement cycle to generate an electricity consumption time-series dataset containing the electricity consumption and duration of each time slot, specifically including: The intelligent agent for extracting bill elements identifies the billing start and end dates of the electricity bill data object to determine the settlement period, and uses a pre-set thesaurus to map the original labels of the time-of-use electricity values ​​to a preset set of standard time periods. The cumulative electricity consumption for each standard time period is extracted as the electricity consumption for that time period. ; The document element extraction agent identifies the maximum demand field, contract demand field, or transformer capacity field from the power document data object according to a preset priority order, and uses the identified value as the maximum demand value. ; The document element extraction agent is set to a demand cycle. The settlement period is divided into multiple consecutive metering time slots, with the smallest metering unit being the settlement period itself. The intelligent agent for extracting invoice elements determines the time period type of each metering time slot based on a pre-set time-of-use electricity price template and a calendar database. And count the number of the metering time slots belonging to the same time period type. ; The document element extraction agent calculates the duration of the time period for this time period type according to the following formula. : ; in, It belongs to the time period type The number of the aforementioned measurement time slots, The demand cycle is mentioned above; The bill element extraction agent will extract the electricity amount for the specified time period. With the corresponding time period duration Combined, the electricity consumption time-series dataset is formed. ,in , .

[0014] The time-of-use electricity consumption figure is the cumulative energy consumption recorded in the electricity ticket according to different electricity usage periods. The table or text area of ​​the electricity ticket can be deciphered using optical character recognition technology.

[0015] The maximum demand value is the maximum average power per unit time within the settlement period recorded in the electricity bill. Key field areas of the electricity bill can be analyzed using optical character recognition (OCR) technology.

[0016] Billing cycle data is the relevant data recorded in electricity bills used to determine the time range for electricity settlement. The billing instructions area of ​​the electricity bill can be deciphered using optical character recognition (OCR) technology.

[0017] The billing start and end dates are the beginning and end dates of the electricity billing cycle specified in the electricity bill. The billing cycle field of the electricity bill can be deciphered using optical character recognition (OCR) technology.

[0018] The thesaurus is a pre-compiled dictionary used to match the original time-of-use electricity labels in electricity bills with the standard time period names. Preferably, it is a label matching set that includes peak, high, peak, flat, flat period, valley, and low valley, based on the common time period division type of time-of-use electricity pricing in the power industry, and can cover the time period label descriptions of most electricity bills.

[0019] The standard time period set is a pre-defined set of electricity consumption time periods used to standardize the time-of-use electricity labels on electricity bills. Preferably, it is a four-time period set comprising peak, off-peak, and valley periods. Alternatively, a three-time period set comprising peak, off-peak, and valley periods, or a two-time period set comprising peak and valley periods, can be selected based on the actual scenario, taking into account the mainstream time period division methods of time-of-use electricity pricing policies in various regions.

[0020] The electricity consumption per time period is the cumulative energy consumption value for each time period within a standard time period set. The electricity consumption field for each time period in an electricity ticket can be analyzed using optical character recognition (OCR) technology.

[0021] The maximum demand field is the field name area in an electricity bill that identifies the maximum demand value. Electricity bills can be parsed using optical character recognition (OCR) technology combined with keyword matching.

[0022] The contract demand field is the area in an electricity invoice that identifies the contracted demand value. Electricity invoices can be parsed using optical character recognition (OCR) technology combined with keyword matching.

[0023] The transformer capacity field is the field name area in electricity invoices that identifies the transformer capacity value. Electricity invoices can be parsed using optical character recognition (OCR) technology combined with keyword matching.

[0024] The demand period is a pre-set time period that serves as the smallest metering unit. A value of 0.25 hours is preferred because the power industry commonly uses a 15-minute average power consumption to measure maximum demand, aligning with industry metering rules.

[0025] A metering time slot is a continuous time unit obtained by dividing the settlement cycle according to the demand cycle.

[0026] Time-of-use pricing templates are pre-compiled templates that include time ranges for different date types. Preferably, they include weekdays, weekends, and public holidays to accommodate the varying time ranges required by local time-of-use pricing policies, which differentiate time periods based on date type.

[0027] The calendar library is a pre-configured database of date attributes, including statutory holidays, workdays, and weekends. Ideally, it should cover statutory holidays and workday adjustments for all regions of China from 2000 to 2100, meeting long-term electricity invoice verification needs and adapting to date attribute determination in different regions.

[0028] The time slot type is a standard time slot category obtained by matching the time-of-use electricity price template and calendar database based on the metering time slot.

[0029] The number of metering slots is the total number of metering slots belonging to the same time period type within the settlement cycle.

[0030] The duration of a time period is the product of the number of metering time slots corresponding to each time period type and the demand cycle.

[0031] Electricity consumption time series datasets are structured data collections formed by combining the electricity consumption of each standard time period with the corresponding time period duration.

[0032] By using a pre-defined thesaurus to map the original labels of time-of-use electricity values ​​to a pre-defined set of standard time periods, the system ensures that electricity bills may have different labels for time-of-use electricity. For example, peak electricity on a bill is mapped to the peak period in the standard time period set, and off-peak electricity is mapped to the off-peak period in the standard time period set. This unified label mapping eliminates the differences in label descriptions between different bills, providing a consistent basis for subsequent calculations. For instance, if one bill indicates a peak electricity of 1000 kWh and another bill indicates a peak electricity of 800 kWh, both can be mapped to the peak period of the standard time period for subsequent processing.

[0033] Identifying the maximum demand, contract demand, and transformer capacity fields according to a preset priority order and determining the maximum demand value means prioritizing the extraction of the maximum demand value directly stated on the invoice. If the invoice lacks this field, the contract demand value is extracted as the maximum demand value. If the invoice also lacks a contract demand field, the transformer capacity value is extracted as the maximum demand value. This priority ensures that the power boundary values ​​used for subsequent calculations can be stably obtained even when the invoice fields are described differently. For example, if an invoice only specifies a transformer capacity field of 500 kVA, this value is taken as the maximum demand value.

[0034] Dividing the settlement period into metering slots by setting the demand cycle as the smallest metering unit means using a fixed demand cycle as the unit and breaking down the entire settlement period into continuous and non-overlapping time units. The duration of each unit is consistent with the demand cycle, allowing subsequent energy distribution projections to match the demand metering rules of the power industry. For example, if the settlement period is 720 hours and the demand cycle is 0.25 hours, it can be divided into 2880 continuous metering slots.

[0035] Determining the time slot type for each metering time slot based on a pre-set time-of-use pricing template and a calendar database involves first determining whether the date of the metering time slot is a weekday, weekend, or public holiday using the calendar database, and then matching the time slot's time range with the standard time period corresponding to that date type using the time-of-use pricing template. This ensures that the time slot type determination for each metering time slot aligns with the actual time-of-use pricing policy. For example, if a metering time slot is a weekday from 9:00 AM to 9:15 AM, and the time-of-use pricing template indicates that 9:00 AM on weekdays is a peak period, then the time slot type for that metering time slot is peak.

[0036] Counting the number of metering slots of the same time period type and multiplying them by the demand cycle to calculate the duration of the time period refers to counting the metering slots belonging to the same standard time period, and then calculating the actual total duration of that time period within the settlement cycle through multiplication. This method can accurately calculate the duration of each time period and avoids the expression error when directly extracting the time period duration from the invoice. For example, if the number of metering slots in the peak period is 800 and the demand cycle is 0.25 hours, then the duration of the peak period is 200 hours.

[0037] Combining electricity consumption data with corresponding time-period durations to form a time-series electricity consumption dataset means mapping the electricity consumption of each standard time period to the calculated time period duration, creating a structured dataset. This allows subsequent calculations of the upper and lower bounds of the loss product to directly utilize the standardized time-period data, eliminating the need for secondary parsing of the invoice data. For example, peak-period electricity consumption of 9000 kWh and duration of 50 hours, normal-period electricity consumption of 12000 kWh and duration of 100 hours, and valley-period electricity consumption of 5000 kWh and duration of 50 hours can be combined into a time-series electricity consumption dataset containing three sets of data.

[0038] The thesaurus provides a bidirectional correspondence between peak and high, valley and low, and flat and flat periods. It is constructed to cover all common time-of-use expressions in the power industry and is updated by supplementing and revising the dictionary based on new time-of-use labels issued by the power industry in various regions, ensuring that the dictionary can continuously match the label expressions of various invoices.

[0039] The specific rules for dividing the standard time period set are as follows: the four-time period set includes peak, peak, flat, and valley. Peak is the time period with the highest electricity load, peak is the time period with the second highest electricity load, flat is the time period with stable electricity load, and valley is the time period with the lowest electricity load; the three-time period set includes peak, flat, and valley; and the two-time period set includes peak and valley. The corresponding set can be selected according to the actual time period marking on the ticket.

[0040] The preset priority order for identifying the maximum demand, contract demand, and transformer capacity fields is that the maximum demand field takes precedence over the contract demand field, and the contract demand field takes precedence over the transformer capacity field. This priority is set according to the metering rules of the power industry. The maximum demand is the actual metered value, which best matches the power boundary requirements for subsequent calculations.

[0041] The specific standard for the demand period is 0.25 hours by default. If the demand period value is clearly marked in the electricity bill, then the value marked in the bill shall be used, because the general metering period for the maximum demand in the power industry is 15 minutes, which is 0.25 hours.

[0042] The specific content of the time-of-use electricity pricing template includes three date types: weekdays, weekends, and statutory holidays. Each date type corresponds to the specific time range of peak, peak, flat, and valley periods, which is consistent with the time-of-use electricity pricing policies issued by local development and reform commissions. The matching rule with invoices is to match the time-of-use electricity pricing template of the corresponding region based on the region where the invoice was issued. The version update rule is to update the time range of the corresponding template synchronously after the time-of-use electricity pricing policies of various regions are adjusted.

[0043] The calendar database contains date information for statutory holidays, workdays, and weekends in all provinces and cities across the country from 2000 to 2100. The rules for determining holidays and workdays follow the annual holiday arrangement notice issued by the State Council. The time zone adaptation rule is to use the East Eighth Time Zone by default. If the ticket indicates a different time zone, the date will be adjusted according to the time zone indicated on the ticket.

[0044] The specific rule for dividing the metering time slots is to divide them sequentially according to the demand period, starting from the start time of the billing start and end dates. If the last metering time slot is less than the demand period, it is still treated as a complete metering time slot. The alignment of the first and last time slots of the settlement period is aligned with the time nodes of the demand period. For example, if the demand period is 15 minutes, the start time of the time slots is the hour, 15 minutes, 30 minutes, and 45 minutes.

[0045] The specific data format of the electricity consumption time series dataset is a two-dimensional list format, with each row containing three data items: standard time period name, time period electricity consumption, and time period duration.

[0046] Preferably, the power load projection agent calculates the lower limit of the loss product of the power square integral in each time period based on the power consumption time series dataset, and performs constrained discrete projection of the energy distribution in each time period based on the maximum demand value, constructing an upper limit of the loss product that satisfies the maximum demand boundary, specifically including: The power load projection agent targets each time period type. The lower limit of the loss product for this time period type is calculated according to the following formula. : ; in, The electricity consumption for that time period type. The duration of the time period of this time period type; The power load projection agent targets each time period type. Calculate the number of fully loaded time slots for this time period type using the following formula. With remaining power value : ; ; in, The electricity consumption for that time period type. The demand cycle is mentioned above. This refers to the maximum required value. This indicates the floor function; The power load projection agent calculates the upper limit of the loss product for this time period type according to the following formula. : ; in, The demand cycle is mentioned above. The number of full-load time slots, This refers to the maximum required value. The remaining power value is given.

[0047] The lower limit of the loss product is calculated by the power load simulation agent for each time period type, based on the ratio of the square of the power consumption in that time period to the duration of that time period. It reflects the minimum value of the square integral of the power during that time period.

[0048] The total power of time slots is obtained by dividing the power consumption of each time slot type by the demand cycle, which is converted by the power load projection agent. It is the sum of the power values ​​of all metered time slots in that time period.

[0049] The number of full-load time slots is obtained by dividing the sum of the time slot power by the maximum demand value and taking the integer part of the result using the power load projection intelligent agent. It is the number of time slots in which the power reaches the maximum demand value during this period.

[0050] The remaining power value is the difference between the sum of the time slot power calculated by the power load simulation agent and the product of the number of full-load time slots and the maximum demand value. It is the remaining power value outside the full-load time slots during this period.

[0051] The upper limit of the loss product is calculated by the power load simulation agent for each time period type, combining the demand cycle, the number of full-load time slots, the maximum demand value, and the remaining power value. It reflects the maximum value of the square integral of power during that time period.

[0052] The lower limit of the loss product is calculated by the ratio of the square of the electricity consumption during a time period to the duration of that time period. This means that the variable losses in a distribution network are proportional to the square of the current, and the current is approximately positively correlated with active power. The power square integral is related to the variable losses. Using this ratio, the lower limit of the power square integral can be derived from the electricity consumption and duration of the ticket, establishing a correlation between summary data and nonlinear losses without the need for continuous load curves. For example, if the electricity consumption during a time period is 9000 kWh and the duration is 50 hours, the calculated lower limit of the loss product is 1,620,000, which is the minimum possible value of the power square integral for that time period.

[0053] Dividing the electricity consumption during a time period by the demand cycle to convert it into the total power in each time slot means that electricity consumption is the product of power and time. This division transforms the energy value into a power value, allowing subsequent energy distribution projections to match the metering time slots based on the demand cycle, thus aligning with the power industry's demand metering rules. For example, if the electricity consumption during a certain time period is 9000 kWh and the demand cycle is 0.25 hours, the calculated total power in each time slot is 36000 kW, corresponding to the sum of power in all metering time slots for that time period.

[0054] The method of extracting the number of full-load time slots from the total time slot power using rounding down means that, in order to conform to the power boundary constraint of the maximum demand, only the integer part of the total time slot power that is divisible by the maximum demand value is taken as the number of time slots to measure the power reaching the maximum demand, ensuring that the calculated power value does not exceed the maximum demand boundary. For example, if the total time slot power is 36,000 kW and the maximum demand is 300 kW, rounding down yields 120 full-load time slots.

[0055] The difference between the total power of the time slots and the power of the full-load time slots is used to obtain the remaining power value. This refers to calculating the unallocated remaining power within a given time period after the power value of the full-load time slots has been determined. This achieves a preliminary discrete allocation of energy under the constraint of maximum demand, providing a basis for the subsequent calculation of the upper limit of the loss product. For example, if the total power of the time slots is 36,000 kW, the number of full-load time slots is 120, and the maximum demand is 300 kW, the calculated remaining power value is 0 kW.

[0056] The upper limit of the loss product is constructed by multiplying the demand period by the product of the number of fully loaded time slots and the square of the maximum demand, and then adding the sum of the squares of the remaining power. This refers to using the principle of extreme values ​​of convex functions to push the power of as many metered time slots as possible to the maximum demand, while concentrating the remaining power in a single time slot, thus obtaining the maximum value of the power square integral. The value calculated based on this is the upper limit of the loss product. For example, with a demand period of 0.25 hours, 120 fully loaded time slots, a maximum demand of 300 kW, and a remaining power of 0 kW, the calculated upper limit of the loss product is 2,700,000.

[0057] The overall approach employs a constrained discrete extrapolation method based on the maximum demand value to construct the upper and lower bounds of the loss product envelope. This means using the maximum demand as a hard power constraint, discretizing the continuous power distribution into metering time slots with the demand period as the unit, and deriving the upper and lower limits of the power square integral for each time period to form the envelope interval of the loss product, rather than using conventional linear calculation methods. This approach ensures that the extrapolation results closely match the actual power load metering patterns.

[0058] The physical meaning of the lower limit of the loss product is the minimum possible value of the integral of the square of power over time within that time period. The detailed derivation of the calculation is based on Cauchy's inequality. The amount of electricity in a time period is the integral of power over time. From the inequality, it can be deduced that the integral of the square of power is greater than or equal to the square of the amount of electricity divided by the duration. Since the integral of the square of power is positively correlated with the variable loss, this ratio is set as the lower limit of the loss product to conform to the laws of electricity.

[0059] The specific constraints of the constrained discrete extrapolation are that the average power of the metering time slot is not greater than the maximum demand value, and the sum of the products of the power of all metering time slots and the demand period is equal to the electricity consumption of that time period, ensuring that the boundary conditions of the extrapolation are clear and in line with the actual power metering requirements.

[0060] The specific method for rounding down when calculating the number of full-load time slots is to directly take the integer part of the sum of the time slot power divided by the maximum demand value, discard the decimal part, and not perform any rounding. The power corresponding to the decimal part is included in the remaining power value to ensure that the power of all full-load time slots does not exceed the maximum demand value.

[0061] The rule for allocating the remaining power value is to concentrate all the remaining power value into a single metering time slot for that time period. The power value of that time slot is the remaining power value, and the power value of the other metering time slots that are not fully loaded is 0. It is not allowed to allocate the remaining power value to multiple metering time slots to ensure that the derivation of the maximum value of the power square integral conforms to the principle of the extreme value of convex function.

[0062] Preferably, the step of generating a power supply logic discrimination quantity based on the numerical inclusion relationship between the lower limit of the loss product and the upper limit of the loss product specifically includes: The power load projection agent calculates the lower limit of the full-cycle loss product according to the following formula. Total electricity consumption during the settlement period Total duration of settlement cycle : ; in, This represents the total number of time period types. This is the lower limit of the loss product. The amount of electricity consumed during the specified time period. The duration of the time period; The power load projection agent calculates the time-period energy slack for each time period type according to the following formula. : ; in, The upper limit of the loss product, This is the lower limit value of the loss product; The power load projection agent calculates the global relaxation compression value according to the following formula. : ; in, It is an exponential function with the natural constant as its base. This is the preset saturation sensitivity coefficient; The power load projection agent calculates the load pattern amplification index according to the following formula. : ; The power load projection agent combines hard unrealizable indicators and morphological anomaly indicators to generate the power supply logic discrimination quantity according to the following formula. : ; in, This is the upper limit of the total lifecycle loss product. This refers to the maximum required value. The preset morphological amplification threshold, This is a logical indicator function that takes the value 1 when the condition is met, and 0 otherwise.

[0063] The saturation sensitivity coefficient is a preset exponential decay transformation coefficient used for calculating the global relaxation compression value, and is a custom parameter. It is preferably 0.05 to adapt to the standard resolution of electricity metering, balancing the feature magnification effect during edge-fitting periods with the stability of numerical calculations, while also covering the needs of most scenarios involving the review of electricity transfer invoices.

[0064] The lower limit of the total cycle loss product is the overall minimum value of the power square integral within the settlement period obtained by summing the lower limits of the loss products for all time periods by the power load projection intelligent agent.

[0065] The total electricity consumption during the settlement period is the total energy consumption value within the settlement period obtained by summing up the electricity consumption of all time periods of all time periods by the power load projection intelligent agent.

[0066] The total settlement cycle duration is the overall length of the settlement cycle obtained by summing the durations of all time periods of all time periods by the power load projection intelligent agent.

[0067] The energy slack for a given time period is calculated by the power load projection agent for each time period type. The difference between the upper and lower limits of the loss product is then divided by the upper limit of the loss product. This value is used to quantify the margin of the upper and lower limits of the loss product for that time period.

[0068] The global relaxation compression value is a value obtained by using an electrical load extrapolation agent to perform an exponential decay transformation on the energy relaxation of all time periods based on the natural constant and the saturation sensitivity coefficient, and then taking the average value. It is used to comprehensively reflect the degree of edge contact in each time period of the entire cycle.

[0069] The load pattern amplification index is a value obtained by using the power load projection intelligent agent to multiply the product of the total settlement cycle duration and the lower limit of the full cycle loss product, and then dividing it by the square of the total electricity consumption in the settlement cycle. It is used to quantify the degree of square law amplification of the load structure.

[0070] The upper limit of the total cycle loss product is the overall maximum value of the power square integral within the settlement period obtained by summing the upper limit values ​​of the loss products of all time period types by the power load projection intelligent agent.

[0071] The load shape amplification threshold is a preset critical value used to determine whether the load shape is abnormal; it is a user-defined parameter. A value of 4.0 is preferred to meet the load shape statistics of power industry power transfer scenarios. This value is a reasonable dividing point for load structure uniformity; exceeding it indicates extremely uneven load shape.

[0072] Indicator functions are used to quantify hard unrealizable conditions and morphological anomalies. They take a value of 1 when the condition is met and a value of 0 when the condition is not met.

[0073] The power supply logic discrimination quantity is a single value obtained by combining the power load deduction intelligent agent with hard unrealizable indicators, abnormal morphological indicators and global relaxation compression value through weighted or logical summation. It is used to comprehensively determine the metering rationality of power bills.

[0074] The lower limit of the loss product for all time periods, the electricity consumption for each time period, and the duration of each time period are calculated separately to obtain the relevant value for the entire cycle. This means aggregating the extrapolated data at the time period level into overall data for the settlement cycle, achieving a dimensional transformation from local to global, and providing a unified calculation basis for subsequent overall feasibility assessments. For example, the lower limit of the loss product for the peak, flat, and valley periods of a certain electricity transfer ticket are 1,620,000, 1,440,000, and 500,000, respectively. After summing them, the lower limit of the loss product for the entire cycle is 3,560,000, which reflects the lower limit of the power square integral for the entire settlement cycle.

[0075] The energy slack for a time period is calculated by dividing the difference between the upper and lower limits of the loss product by the upper limit. This quantifies the margin of the loss product for each time period using a relative value. The smaller the value, the closer the lower limit of the loss product is to the upper limit, and the closer the load distribution is to the maximum demand constraint. For example, if the upper limit of the loss product for a certain time period is 2,700,000 and the lower limit is 1,620,000, the calculated energy slack for that time period is 0.4, which reflects a 40% margin in the loss product for that time period.

[0076] Based on the exponential decay transformation of the natural constant combined with the saturation sensitivity coefficient, the global relaxation compression value is obtained by averaging the energy relaxation of each time period. This means that the characteristics of the edge-adjacent time periods with small relaxation are amplified by the nonlinear exponential decay transformation, and then the characteristics of each time period in the whole cycle are integrated by averaging, making the abnormal edge-adjacent time periods more prominent in the global index. For example, if the relaxation of a certain bill in three time periods is 0.1, 0.5 and 0.6 respectively, and the saturation sensitivity coefficient is 0.05, the numerical difference is amplified after the exponential decay transformation, and the global relaxation compression value obtained by averaging can significantly reflect the edge-adjacent characteristics of the first time period.

[0077] The load pattern amplification index is calculated by dividing the product of the total settlement cycle duration and the lower limit of the full-cycle loss product by the square of the total electricity consumption during the settlement cycle. This index quantifies the unevenness of the load structure by utilizing the correlation between the square integral of power and electricity consumption. A higher value indicates a more pronounced peak-valley difference in load and a higher potential loss driven by the square law. For example, for a bill with a total settlement cycle duration of 200 hours, a lower limit of the full-cycle loss product of 3,560,000, and a total electricity consumption of 26,000 kWh, the calculated load pattern amplification index is 1.06, reflecting a relatively uniform load pattern for this bill.

[0078] Introducing indicator functions to quantify hard-to-achieve and abnormal morphological conditions means transforming qualitative logical judgments into quantitative 0 or 1 values, allowing non-numerical judgment conditions to participate in subsequent numerical calculations and achieving the integration of qualitative and quantitative indicators. For example, if the lower limit of the full-cycle loss product is greater than the upper limit, the indicator function takes the value of 1; otherwise, it takes the value of 0. This value can be directly summed with other indicators.

[0079] Combining hard unrealizable indicators, abnormal load morphology indicators, and global relaxation compression values, a single power supply logic discriminant is generated through weighted or logical summation. This means normalizing multi-dimensional audit indicators into a comprehensive value, achieving collaborative judgment based on multiple features, avoiding the one-sidedness of single-indicator judgment, and making the audit results more comprehensive. For example, directly summing the values ​​of two indicators with the global relaxation compression value yields a power supply logic discriminant that can simultaneously reflect both unrealizableness and load morphology abnormalities.

[0080] The specific rules for determining the saturation sensitivity coefficient are as follows: the default value is 0.05. If the quantization step size of the electricity bill is 0.01 kWh, the value is 0.01. If the quantization step size is greater than 0.01 kWh, the value is taken as 5 times the quantization step size. All values ​​are limited to between 0.01 and 0.10. Values ​​outside this range are taken as boundary values. This rule ensures that the coefficient is compatible with the metering resolution and takes into account both feature amplification effect and numerical stability.

[0081] The specific calculation details of the exponential decay transformation are as follows: only the values ​​of energy relaxation greater than 0 in the time period are transformed. If the relaxation is less than or equal to 0, the negative infinity power of the natural constant is directly taken, which is 0.

[0082] The specific triggering conditions for the indicator function are as follows: when the lower limit of the full-cycle loss product is greater than the upper limit, the lower limit of the root mean square power is greater than the maximum demand value, and the load amplification index is greater than the amplification threshold, the corresponding indicator function is triggered respectively. The tolerance rule allows a relative deviation of 0.5% when making the judgment. If the numerical difference is within the deviation range, it is considered that the condition is not met, so as to avoid false triggering caused by rounding error of the bill value.

[0083] The load amplification threshold is based on statistical data of electricity bills in power transfer scenarios across the country. 95% of compliant bills have a load amplification index of less than 4.0. Bills that exceed this value often have anomalies such as time period mapping errors or metering clock drift. This threshold is a statistically reasonable anomaly dividing point.

[0084] The specific method for weighting or summing the power supply logic discrimination quantities is to use direct logical summation by default, which adds the value of each indicator to the global relaxation compression value directly. If weighted summation is required, the weight of the hard-to-achieve indicator is 0.5, the weight of the abnormal shape indicator is 0.3, and the weight of the global relaxation compression value is 0.2. The weight allocation is set according to the degree of influence of each indicator on the audit result.

[0085] The range of energy relaxation for a given period is greater than or equal to 0 and less than or equal to 1. If the calculation result exceeds this range, the boundary value is taken directly. 0 indicates that there is no loss product margin for that period, and 1 indicates that the lower limit of the loss product for that period is 0.

[0086] The abnormal threshold for the global relaxation compression value is 0.8. If the calculated global relaxation compression value is greater than or equal to 0.8, it indicates that the bill has edge characteristics in multiple time periods, which is an abnormal situation of load form. This threshold is determined by the statistical average of the global relaxation compression values ​​of compliant bills and is a reasonable threshold for abnormal judgment.

[0087] Preferably, the step of outputting an audit failure conclusion when the lower limit of the loss product is greater than the upper limit of the loss product or the power supply logic discrimination quantity exceeds a preset range specifically includes: The power load simulation agent calculates the upper limit of the full-cycle loss product according to the following formula. : ; in, This represents the total number of time period types. This is the upper limit of the loss product; The power load projection agent performs the following logical judgment; if any one of the conditions is met, it outputs a conclusion that the audit is unqualified: Condition 1: The lower limit of the full-cycle loss product Greater than the upper limit of the total cycle loss product ,Right now: ; Condition 2: The lower limit of the root mean square power calculated according to the following formula is greater than the maximum demand value. : ; in, The total duration of the settlement cycle; Condition 3: The power supply logic discrimination quantity Greater than or equal to the preset discrimination threshold : ; If none of the above conditions are met, the power load simulation agent will output a qualified conclusion.

[0088] The discrimination threshold is a preset critical value used to determine whether the power supply logic discrimination quantity is abnormal, and it is a custom parameter. It is preferably 1.60 to adapt to the requirement of integrating hard judgment items and form judgment items in power bill review. When there is no hard failure or form abnormality, the power supply logic discrimination quantity is in the range of 0 to 1. After adding the form abnormality indicator, it increases by a minimum of 1. This value can stably distinguish between compliant and abnormal bills, covering the review requirements of most power transfer scenarios.

[0089] The root mean square power lower limit is a value obtained by taking the square root of the ratio of the lower limit of the full-cycle loss product calculated by the power load projection agent to the total duration of the settlement cycle. It reflects the minimum possible value of the root mean square power consumption within the settlement cycle.

[0090] The upper limit of the loss product for all time periods is accumulated to obtain the upper limit of the total cycle loss product. This upper limit is compared with the lower limit of the total cycle loss product as the first hard criterion. This means checking whether the lower limit of the power square integral exceeds the upper limit from a full-cycle perspective. If it does, it indicates that the time-of-use electricity structure of the ticket is mathematically impossible to realize. This is the most basic physical feasibility check, and an anomaly can be directly determined without relying on other features. For example, if the upper limits of the loss product for peak, flat, and valley periods for a certain ticket are 2,700,000, 3,600,000, and 1,500,000 respectively, the accumulated upper limit of the total cycle loss product is 7,800,000, and the lower limit is 3,560,000. If the lower limit is less than the upper limit, no disqualification is triggered. However, if the lower limit of the total cycle loss product is 8,000,000, which is greater than the upper limit, then it is directly determined to be disqualified.

[0091] The second, hard criterion is to calculate the lower limit of the root mean square (RMS) power and compare it with the maximum demand value. This means that, based on the mathematical relationship between RMS power and its maximum value, the maximum power must be greater than or equal to the RMS power. If the lower limit of the RMS power exceeds the maximum demand value, it indicates that the bill data violates basic mathematical laws and is direct evidence that it is physically impossible. For example, if a bill has a lower limit of RMS power of 298 kW and a maximum demand of 300 kW, it will not trigger an invalidation. However, if the lower limit of the RMS power is 305 kW, which is greater than the maximum demand value, it will be directly judged as invalid.

[0092] The third comprehensive judgment is to compare the power supply logic discrimination quantity with the preset discrimination threshold. This means combining the results of the first two hard judgments, abnormal load patterns, and edge-fitting characteristics throughout the entire cycle, and performing a comprehensive verification using a single value. This covers abnormal pattern and edge-fitting scenarios beyond those that are considered impossible to achieve, thus compensating for the verification blind spots of the first two hard judgments. For example, if a bill has no impossible conditions, but the load pattern amplification index exceeds the threshold and the load is edge-fitting for multiple periods, the power supply logic discrimination quantity is calculated to be 1.7, which is greater than the discrimination threshold of 1.60, so it is judged as unqualified.

[0093] The rule employing a three-level judgment process, where a document is deemed unqualified if any one condition is met, and qualified if none are met, prioritizes the hard judgment of physical feasibility. This eliminates documents that are mathematically and physically impossible to fulfill. A comprehensive judgment then identifies hidden morphological and edge-fitting anomalies, ensuring the hierarchical and rigorous nature of the judgment logic and guaranteeing the uniqueness of the audit conclusion without ambiguous intermediate judgments. For example, if a document triggers the first hard judgment, it is directly deemed unqualified without further judgments; if neither of the first two judgments is triggered, the third comprehensive judgment is then executed.

[0094] The tolerance rule for comparing the lower limit of the root mean square power with the maximum demand value is to allow a relative deviation of 0.5%. If the proportion of the lower limit of the root mean square power exceeding the maximum demand value is within 0.5%, it is considered that the numerical difference is caused by the rounding error of the invoice and will not trigger the second judgment of non-compliance. If the proportion exceeds 0.5%, it will directly trigger non-compliance. This rule avoids misjudgment caused by the rounding of values ​​in measurement and invoicing.

[0095] The discrimination threshold is based on statistical data from 100,000 sets of compliant and abnormal invoices in the power transfer scenario. The power supply logic discrimination value of compliant invoices is less than 1.60, while the power supply logic discrimination value of abnormal invoices with time period mapping errors, metering clock drift, and missing sampling filling errors is greater than or equal to 1.60. This threshold is the statistical dividing point between compliance and abnormality. The scenario adaptation rule is that if it is applied to the review of invoices in a park with a single business type, the threshold can be slightly adjusted to between 1.50 and 1.70, and kept at around 1.60 overall.

[0096] The specific execution order of the three-level judgment is that the first judgment takes precedence over the second judgment, and the second judgment takes precedence over the third judgment. This execution order is a fixed and unadjustable rule to ensure that the hard verification of feasibility is completed first, and then the comprehensive feature verification is performed, which conforms to the hierarchical nature of the audit logic. If the preceding judgment triggers an unqualified result, the subsequent judgment does not need to be executed, and the audit conclusion is directly output, thereby improving audit efficiency.

[0097] The specific format for the audit conclusion output is a structured text containing the conclusion type, trigger judgment type, and core anomaly indicator values. The conclusion type only includes two categories: qualified and unqualified. If the output is unqualified, it is necessary to clearly indicate whether the first judgment, second judgment, or third judgment was triggered. If the third judgment is triggered, the specific value of the power supply logic judgment quantity should also be marked to facilitate subsequent invoice review and anomaly cause investigation.

[0098] like Figure 2 As shown, Figure 2 This is a schematic diagram illustrating a specific implementation of the present invention, showing the entire process from electricity bill data to output audit results.

[0099] It should be noted that the interval and threshold sizes are set for ease of comparison. The size of the threshold depends on the amount of sample data and the base number set by those skilled in the art for each set of sample data, as long as it does not affect the proportional relationship between the parameter and the quantized value. Furthermore, the above formulas are all dimensionless calculations, and the formulas are derived from software simulations using a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0100] The embodiments of this example have been described above. However, this example is not limited to the specific implementation methods described above. The specific implementation methods described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms based on the guidance of this example, and all of them are within the protection scope of this example.

Claims

1. A multi-agent collaborative verification system for electricity bills, characterized in that, This includes an intelligent agent for extracting ticket elements based on communication connections and an intelligent agent for extrapolating electricity load; The bill element extraction agent is configured to parse the electricity bill data object to extract time-of-use electricity values, maximum demand values, and billing cycle data. Based on the settlement cycle determined by the billing cycle data, it divides the data into multiple metering time slots to generate an electricity consumption time-series dataset containing time-period electricity consumption and time-period duration. The electricity load projection agent is configured to calculate the lower limit of the loss product of the power square integral within each time period based on the electricity consumption time-series dataset. Based on the maximum demand value, it performs constrained discrete projection of the energy distribution within each time period to construct an upper limit of the loss product that satisfies the maximum demand boundary. The system generates a power supply logic discrimination quantity based on the numerical inclusion relationship between the lower limit and the upper limit of the loss product. If the lower limit of the loss product is greater than the upper limit or the power supply logic discrimination quantity exceeds a preset range, it outputs a conclusion of audit failure.

2. The multi-agent collaborative review system for electricity bills according to claim 1, characterized in that, The bill element extraction agent parses the electricity bill data object to extract time-of-use electricity consumption values, maximum demand values, and billing cycle data. Based on the settlement cycle determined by the billing cycle data, it divides the data into multiple metering time slots, generating an electricity consumption time-series dataset containing time-period electricity consumption and time-period duration, including: The bill element extraction agent identifies the billing start and end dates of the electricity bill data object to determine the settlement period, uses a preset thesaurus to map the original labels of the time-of-use electricity values ​​to a preset set of standard time periods, and extracts the cumulative electricity corresponding to each standard time period as the time period electricity; the bill element extraction agent identifies the maximum demand field, contract demand field, or transformer capacity field from the electricity bill data object according to a preset priority order, and uses the identified value as the maximum demand value.

3. The multi-agent collaborative review system for electricity bills according to claim 2, characterized in that, The bill element extraction agent parses the electricity bill data object to extract time-of-use electricity consumption values, maximum demand values, and billing cycle data. Based on the settlement cycle determined by the billing cycle data, it divides the data into multiple metering time slots to generate an electricity consumption time-series dataset containing time-period electricity consumption and time-period duration. It also includes: The bill element extraction agent sets the demand period as the smallest metering unit and divides the settlement period into multiple consecutive metering time slots. Based on the preset time-of-use electricity price template and calendar library, the bill element extraction agent determines the time period type of each metering time slot and counts the number of metering time slots belonging to the same time period type. The bill element extraction agent calculates the time period duration of the time period type by multiplying the quantity by the demand period, and combines the time period electricity consumption with the corresponding time period duration to form the electricity consumption time series dataset.

4. The multi-agent collaborative review system for electricity bills according to claim 3, characterized in that, The power load projection agent calculates the lower limit of the loss product of the power square integral in each time period based on the power consumption time series dataset, and performs constrained discrete projection of the energy distribution in each time period based on the maximum demand value, constructing an upper limit of the loss product that satisfies the maximum demand boundary, including: For each time period type, the power load projection agent calculates the ratio of the square of the power consumption during that time period to the duration of that time period, and uses the calculated ratio as the lower limit of the loss product for that time period type. For each time period type, the power consumption projection agent divides the power consumption during that time period type by the demand period to obtain the total power of the time slots for that time period type, and divides the total power of the time slots by the maximum demand value and takes the integer part to obtain the number of full-load time slots.

5. The multi-agent collaborative verification system for electricity bills according to claim 4, characterized in that, The power load projection agent calculates the lower limit of the loss product of the power square integral in each time period based on the power consumption time series dataset, and performs constrained discrete projection of the energy distribution in each time period based on the maximum demand value to construct the upper limit of the loss product that satisfies the maximum demand boundary. It also includes: The power load projection agent calculates the difference between the sum of the time slot power and the product of the number of fully loaded time slots and the maximum demand value, and uses this difference as the remaining power value; the power load projection agent calculates the product of the square of the maximum demand value and the number of fully loaded time slots, and calculates the square of the remaining power value, and multiplies the sum of the two by the demand period to obtain the upper limit of the loss product for this time period type.

6. The multi-agent collaborative review system for electricity bills according to claim 5, characterized in that, The system generates power supply logic discrimination quantities based on the numerical inclusion relationship between the lower limit of the loss product and the upper limit of the loss product, including: The power load simulation agent accumulates the lower limit values ​​of the loss product for all time periods to obtain the lower limit value of the loss product for the entire cycle, and accumulates the electricity consumption and duration for each time period to obtain the total electricity consumption and total duration for the settlement cycle. For each time period, the agent calculates the difference between the upper and lower limit values ​​of the loss product, and divides this difference by the upper limit value to obtain the energy relaxation for that time period. The power load simulation agent is based on the natural constant and... A preset saturation sensitivity coefficient is used to perform exponential decay transformation and mean calculation on the energy relaxation of the time period for all time periods to obtain a global relaxation compression value. The power load projection agent calculates the product of the total duration of the settlement cycle and the lower limit of the full-cycle loss product, and divides the product by the square of the total electricity consumption of the settlement cycle to obtain the load pattern amplification index. Based on the lower limit of the full-cycle loss product, the load pattern amplification index, and the global relaxation compression value, the power supply logic discrimination quantity is generated by weighted or logical summation.

7. The multi-agent collaborative review system for electricity bills according to claim 6, characterized in that, When the lower limit of the loss product is greater than the upper limit of the loss product or the power supply logic discrimination value exceeds the preset range, the audit failure conclusion is output, including: First determination: The power load simulation agent accumulates the upper limit of the loss product for all time periods to obtain the upper limit of the full-cycle loss product; The power load simulation agent determines whether the lower limit of the full-cycle loss product is greater than the upper limit of the full-cycle loss product. If so, it outputs a conclusion that the audit is not qualified. Second determination: The power load simulation agent calculates the ratio of the lower limit of the full-cycle loss product to the total duration of the settlement cycle, and performs a square root operation on the ratio to obtain the lower limit of the root mean square power; The power load simulation agent determines whether the lower limit of the root mean square power is greater than the maximum demand value. If so, it outputs a conclusion that the audit is not qualified. Third determination: The power load estimation agent determines whether the power supply logic discrimination quantity is greater than or equal to the preset discrimination threshold. If so, it outputs a conclusion that the audit is not qualified. If the first, second, and third criteria are not met, the power load simulation agent outputs a qualified conclusion.

Images