A carbon emission peak early warning method based on discrete diffusion model and dynamic regularization

Abstract

The present application relates to carbon emission peak early warning technical field, especially to a kind of carbon emission peak early warning method based on discrete diffusion model and dynamic regularization, the time-sharing power consumption of service area charging station, pile state information, charging station power supply side capacity upper limit parameter and time-varying power carbon emission factor are obtained according to preset time period, according to observation data, full load cut-off characteristic section is generated, the amount of electricity that can be accepted in each period in prediction window is calculated and is dispersed according to minimum electricity;Full load cut-off characteristic section and future acceptance capacity discrete sequence are used to construct the condition state of discrete denoising diffusion probability model, monotone alignment is carried out to candidate potential electricity discrete section sequence and future acceptance capacity discrete sequence, the potential electricity discrete section sequence is attributed to time period, and time period is summarized, charging carbon emission sequence is calculated, and carbon emission peak early warning result is output, and the present application realizes the pre-position identification and effective early warning of subsequent carbon emission peak risk caused by full load cut-off of service area charging station.

Description

Technical Field

[0001] This invention relates to the field of carbon emission peak early warning technology, and in particular to a carbon emission peak early warning method based on discrete diffusion model and dynamic regularization. Background Technology

[0002] With the continuous growth of new energy vehicle ownership, highway service area charging stations have become a key node in the cross-regional travel energy replenishment system. As an important support for the refined operation of highway service areas, existing technical solutions usually revolve around time-of-use data collection, load status identification, future demand estimation, and carbon emission calculation. By collecting charging pile power readings at fixed statistical periods and summarizing them to form a time-of-use power consumption sequence, and combining it with time-varying electricity carbon emission factors, the carbon emissions for each period are calculated. Furthermore, time series models, machine learning models, or queuing inference models are used to predict future charging demand, thereby outputting load scheduling suggestions, peak hour reminders, and carbon emission analysis results, providing data support for service area charging operation and energy consumption management.

[0003] However, existing technologies have significant shortcomings in scenarios where service areas operate at full capacity. Current carbon emission analyses can only be based on past electricity consumption results, making it difficult to reflect the delayed release of electricity caused by capacity cutoff on the power supply side. This makes it impossible to effectively identify potential charging demand not reflected in real-time metering results. For the arrangement of delayed charging demand into future periods, existing solutions mostly rely on fixed rules, simple sorting, or experience-based allocation methods. There is a lack of a detailed correspondence between future available capacity and potential released electricity that can maintain time sequence while taking into account capacity constraints, resulting in unclear carbon emission release paths and ambiguous time period attributions in subsequent periods. In addition, existing solutions have weak coupling between the generated results and time period attributions during the generation and alignment process, making it difficult to improve the consistency between candidate sequences and future available capacity while maintaining release order constraints. As a result, the risk of peak carbon emissions from subsequent charging caused by full-load cutoff in service area scenarios cannot be effectively warned during actual operation. At present, there is a need for a carbon emission peak warning method based on discrete diffusion models and dynamic regularization. Summary of the Invention

[0004] To address the technical problem that existing charging carbon emission early warning methods cannot provide advance warning of subsequent peak carbon emission periods when the power supply capacity of the service area is cut off at full load, and that operation scheduling lacks a prior basis, this invention provides a carbon emission peak early warning method based on a discrete diffusion model and dynamic regularization.

[0005] This invention provides a carbon emission peak early warning method based on a discrete diffusion model and dynamic regularization, employing the following technical solution: A peak carbon emission early warning method based on a discrete diffusion model and dynamic regularization includes: The system acquires time-of-use power consumption, charging pile status information, charging station power supply capacity limit parameters, and time-varying electricity carbon emission factors of charging stations in the service area according to preset time periods, forming observation data. Based on the observation data, determine the continuous time periods when the time-of-use power consumption reaches the capacity limit and calculate the drop slope after the full load is released to generate the full load cutoff characteristic segment. Based on the full-load truncation characteristic segment, the electricity that can be received in each time period within the prediction window is calculated and discretized according to the minimum electricity, so as to obtain the discrete sequence of future capacity. The conditional state of the discrete denoising diffusion probability model is constructed by using the full-load truncation feature segment and the discrete sequence of future capacity, and a candidate potential power discrete segment sequence is generated. Monotonic alignment is performed between the candidate potential power discrete segment sequence and the future capacity discrete sequence to obtain the time period belonging path identifier, which is then written into the candidate potential power discrete segment sequence to form the time period belonging potential power discrete segment sequence. The discrete sequence of potential electricity generation time periods is aggregated by time period to form a delayed release distribution; The charging carbon emission sequence is calculated based on the delayed release distribution and time-varying electricity carbon emission factor, and the peak carbon emission warning result is output.

[0006] Furthermore, the formation of observation data specifically includes: The power consumption readings are obtained from the metering interface of each charging pile in the service area charging station according to the preset time period, and the power consumption readings of each charging pile in the same preset time period are summed to form a time-sharing power consumption sequence according to the order of the preset time period. The operating status is obtained from the status interface of each charging pile in the service area charging station according to a preset time period, and the operating status is converted into the number of available piles, the number of occupied piles, and the number of faulty piles to form pile status information. Obtain the upper limit parameter of the power supply side of the charging station corresponding to the charging station in the service area, and receive the time-varying power carbon emission factor corresponding to the preset time period to form capacity emission corresponding information; The time-of-use power consumption sequence, charging pile status information, and capacity emission corresponding information are time-correlated according to the same preset time period. The time-of-use power consumption, charging pile status information, charging station power supply side capacity limit parameters, and time-varying electricity carbon emission factors within the same preset time period are combined into time period records to form observation data.

[0007] Furthermore, the generation of the full-load truncation feature segment includes extracting time-sharing power consumption and pile status information, comparing the time-sharing power consumption with the upper limit parameter of the charging station's power supply side capacity in each time period, and determining the time period sequence that reaches the upper limit when the time-sharing power consumption reaches the upper limit parameter of the charging station's power supply side capacity and the pile status information meets the preset full-load conditions. The time-series data are continuously merged according to time sequence to determine the consecutive time periods when the time-of-use power consumption reaches the capacity limit; Extract the time-of-use power consumption within the consecutive preset time period after the end of the continuous time period, calculate the change rate of time-of-use power consumption between adjacent time periods, and determine the drop slope after the full load is released based on the relationship between the change rate of time-of-use power consumption and the time period position. The full-load truncation feature segment is generated by combining the continuous time period, its continuous duration, and the drop slope after the full load is released.

[0008] Furthermore, the calculation of the electricity that can be received in each time period within the prediction window includes constructing a prediction window time period sequence based on the full load truncation feature segment and a series of preset time periods after the end position of the continuous time period, wherein each time period in the prediction window time period sequence is arranged in an ascending order of time. Based on the continuous duration L and the descent slope Calculate the decline process of each time period in the prediction window time series relative to the end position of the continuous time period. And accumulate them according to the time period position to form a time period decline sequence; Based on the time period decline sequence and the upper limit parameter of the charging station's power supply side, the upper limit of the capacity corresponding to the end position of the continuous time period is taken as the recovery starting point, and the recovery power consumption is calculated time period by time along the prediction window time period sequence to form a recovery power consumption sequence; The candidate acceptable power consumption sequence is obtained by calculating the time-period difference between the capacity limit parameter and the recovery power consumption sequence. The calculation formula for the fallback process is as follows: , in, For the first The decline process within a prediction window period, For window position index, To eliminate the drop slope after full load, This refers to the intermediate time period number in the cumulative calculation of the decline process. For the preset time period length, This refers to the upper limit parameter for the power supply capacity of the charging station.

[0009] Furthermore, obtaining the discrete sequence of future carrying capacity includes extracting pile state information corresponding to the prediction window time period sequence from the observation data and setting a zero threshold. and preset threshold The pile status information is compared with the set threshold. and By comparing the data, a sequence of acceptable power volumes can be formed; The available power capacity sequence is filtered by time period order and rearranged according to the time sequence of the prediction window time period sequence, and the time period position is written for each reserved time period. and the order of time periods This forms the result of time period succession arrangement; The available electricity capacity for each time period in the time period arrangement results is ranked by minimum electricity capacity. The data is split sequentially, and the source time period position and source time period order are written to form the minimum power arrangement result with time period position; The minimum power consumption results with time period positions are concatenated from front to back according to the source time period position, while maintaining the arrangement order within the same source time period, so that each discrete position retains the corresponding time period position and time period sequence, generating a discrete sequence of future capacity.

[0010] Furthermore, the generation of the candidate potential energy discrete segment sequence includes extracting continuous durations. The drop slope after the full load is removed Time period and location and the order of time periods After sequential encoding, the conditional state vector is constructed by concatenating the data in a fixed order. ; according to and Calculate the number of discrete segments Based on minimum power Assign discrete segment positions to each discrete segment to form an initial potential energy discrete segment sequence, and determine the number of discrete segments. Greater than the number of effective discrete positions The excess portion is sequentially compressed to form a sequence of discrete segments of limited potential power. Will The reverse denoising process of the discrete denoising diffusion probability model, which is input together with the constrained potential power discrete segment sequence, limits the generated object to the potential power discrete segment sequence. After denoising update and sorting constraints, a constrained potential power discrete segment sequence is formed. For each discrete segment in the sequence of constrained potential electrical quantities, determine the discrete segment electrical quantity value, and combine it with... , and Write a release priority identifier for each discrete segment to form a sequence of candidate potential power discrete segments; in, For continuous duration, To eliminate the drop slope after full load, In chronological order of time periods This is the conditional state vector. The number of discrete segments. Minimum power consumption This indicates the time period and location.

[0011] Furthermore, the formation of the potential power discrete segment sequence for the time period includes inputting the candidate potential power discrete segment sequence and the future capacity discrete sequence into the candidate sequence receiving stage, and calculating the monotonic alignment cost sequence based on the matching difference between the power value of each discrete segment and the power capacity that can be received at each discrete location and the order constraint corresponding to the release priority identifier. Based on the monotonic alignment cost sequence, the alignment weights corresponding to each discrete position of each discrete segment are calculated, and the alignment weights are kept to change monotonically along the arrangement order of the future capacity discrete sequence to form a monotonic alignment weight sequence. Perform differentiable dynamic time warping monotonic alignment on the monotonically aligned weight sequence to map each discrete segment to a continuous discrete position in the future capacity discrete sequence, thus obtaining the time period belonging path identifier; The time period-assigned path identifier is written into the candidate potential electricity discrete segment sequence to form the time period-assigned potential electricity discrete segment sequence, and then written into the inverse denoising update state of the discrete denoising diffusion probability model.

[0012] Furthermore, the process of forming a discrete segment sequence of potential electricity for a given time period also includes mapping the electricity value of each discrete segment to a corresponding discrete location according to the time period-assigned path identifier, and performing cumulative calculation on the electricity values ​​of discrete segments mapped to the same discrete location. The cumulative calculation result is compared with the capacity of the corresponding discrete position. If the cumulative calculation result is greater than the corresponding capacity, the capacity of the discrete segment located later is adjusted to the subsequent consecutive discrete positions according to the order of the release priority identifier. If the cumulative calculation result is not greater than the corresponding capacity, the current time period's belonging path identifier is retained. The time period attribution path identifier is updated according to the adjusted discrete location, and the updated time period attribution path identifier is written into the candidate potential power discrete segment sequence to form the time period attribution potential power discrete segment sequence.

[0013] Furthermore, the formation of the delayed release distribution includes extracting the discrete segment power value, release priority identifier and corresponding time period position of each discrete segment from the discrete segment sequence of potential power segments belonging to the time period, grouping discrete segments with the same time period position into the same time period group, and arranging the discrete segment power values ​​in each time period group according to the order of the release priority identifier to form the time period power grouping result; The cumulative calculation is performed on the discrete segment electricity values ​​within each time period in the time period electricity grouping results. The electricity values ​​of multiple discrete segments within the same time period are accumulated to form the electricity released in the corresponding time period. The electricity released in the time period that does not match a discrete segment is set to zero, thus forming a time period electricity release sequence. The time-series electricity release sequence is reconstructed into a complete time series result according to the chronological order of each time period within the prediction window, thus obtaining the delayed release distribution.

[0014] Furthermore, the output carbon emission peak warning result includes matching the released electricity in each time period of the delayed release distribution with the factor value of the corresponding time period in the time-varying power carbon emission factor according to the same time period position, multiplying them one by one to obtain the carbon emission amount of each time period, and arranging them in the order of time periods to form a charging carbon emission sequence. The carbon emission difference between adjacent time periods in the charging carbon emission sequence is calculated to obtain the carbon emission change sequence. Based on the comparison condition that the change direction of the previous time period is upward and the change direction of the next time period is downward, the peak candidate time period is screened. The carbon emission corresponding to the peak candidate time period is combined with the time period position to form the peak determination information. The carbon emissions in the peak determination information are compared with the preset warning threshold. When the carbon emissions meet the comparison condition of the preset warning threshold, the carbon emission peak warning result containing the location of the corresponding time period is output.

[0015] In summary, the present invention has the following beneficial technical effects: 1. This invention determines continuous full-load periods by comparing time-of-use power consumption with the upper limit of power supply capacity on a time-by-time basis, and constructs a full-load cutoff characteristic segment by combining the drop slope after the full load is released. Then, the available space within the prediction window is discretized into a discrete sequence of future capacity with the smallest power granularity. The generated object is limited to a discrete sequence of potential power consumption segments by using a discrete denoising diffusion probability model. This realizes an explicit characterization of the delayed charging power that is not reflected in the real-time metering results under the full-load cutoff condition of the power supply capacity of the charging station in the service area. This expands the subsequent carbon emission analysis from only focusing on the observed power consumption to include the potential release process of the cutoff power.

[0016] 2. This invention introduces differentiable dynamic time warping monotonic alignment in the candidate sequence acceptance stage, performs continuous mapping on the discrete segments of candidate potential power capacity and the discrete sequences of future capacity to maintain the non-reversible time period order, and applies sequential constraints to the alignment cost in combination with the release priority identifier, and writes the time period belonging path identifier back to the reverse denoising update state of the discrete denoising diffusion probability model, thereby realizing the ordered correspondence and closed-loop constraint between the discrete segments of potential power capacity and the future available time periods, avoiding the ambiguity of time period belonging and the distortion of release order caused by fixed rule allocation or empirical proportion amortization.

[0017] 3. This invention introduces threshold screening of the number of available charging piles and the number of faulty charging piles in the future capacity calculation stage, and corrects the release order by combining the comparison relationship between the number of occupied charging piles and the number of available charging piles in the release priority identifier writing stage. At the same time, after the generation of the time period belonging path identifier, the cumulative electricity of the same discrete location is checked and adjusted in order. This invention realizes the dynamic adaptation of the delayed release distribution with the availability status of charging station equipment and the capacity constraints, and improves the response accuracy of carbon emission peak warning results to changes in the on-site operating status of the service area.

[0018] 4. This invention forms a delayed release distribution by summarizing the discrete segments of potential electricity at different time periods according to their time period positions. It then multiplies the released electricity at each time period with the time-varying electricity carbon emission factor according to the unified time period position to obtain the charging carbon emission sequence. Based on the direction of carbon emission change in adjacent time periods, it selects peak candidate time periods and compares them with preset warning thresholds. This enables the early identification and effective warning of the subsequent carbon emission peak risk caused by the full load cut-off of charging stations in the service area. It provides a decision-making basis that can be implemented down to specific time periods for operation scheduling, peak-shifting guidance, and carbon emission risk control. Attached Figure Description

[0019] Figure 1 This is a schematic diagram of the overall process of a carbon emission peak early warning method based on a discrete diffusion model and dynamic regularization according to an embodiment of the present invention.

[0020] Figure 2 This is a schematic diagram illustrating the process of generating observation data according to an embodiment of the present invention.

[0021] Figure 3 This is a schematic diagram of the process for generating the fully loaded truncated feature segment according to an embodiment of the present invention.

[0022] Figure 4 This is a diagram of the input layer architecture of the discrete denoising diffusion probability model according to an embodiment of the present invention.

[0023] Figure 5 This is a diagram of the intermediate layer architecture of the discrete denoising diffusion probability model in an embodiment of the present invention.

[0024] Figure 6 This is a diagram of the output layer architecture of the discrete denoising diffusion probability model according to an embodiment of the present invention.

[0025] Figure 7 This is a flowchart illustrating the generation of a discrete sequence of future capacity in an embodiment of the present invention.

[0026] Figure 8 This is a flowchart illustrating the generation process of candidate potential energy discrete segment sequences according to an embodiment of the present invention.

[0027] Figure 9 This is a flowchart of the time period-attributed path identifier generation process according to an embodiment of the present invention.

[0028] Figure 10 This is a flowchart illustrating the delayed release distribution formation process according to an embodiment of the present invention.

[0029] Figure 11 This is a schematic diagram of the dashboard-style results of the carbon peak warning output in an embodiment of the present invention.

[0030] Figure 12 This is a raster chart showing the predicted mean and uncertainty in an embodiment of the present invention. Detailed Implementation

[0031] The present invention will be further described in detail below with reference to the accompanying drawings.

[0032] Example 1 Reference Figure 1 This embodiment of a carbon emission peak early warning method based on a discrete diffusion model and dynamic regularization includes: like Figure 1 , Figure 2 As shown, S1, the preset time period is determined based on the observation start and end times of the early warning task for peak carbon emissions from a single charge. , This represents the fixed time interval between two adjacent discrete statistical points; the observation start time is denoted as... The observations will be processed from the start time to the end time according to... The number of time periods obtained by segmentation is denoted as , No. The time position of each preset time period is denoted as ,in The set of all charging piles within the service area charging station is denoted as . , This indicates the total number of charging piles in the service area's charging stations.

[0033] For any charging station Read timestamped metering records from the metering interface. ,in Indicates the first The first charging pile The timestamp of each measurement record This indicates the corresponding battery level reading. Indicates the first [number]th ... The number of metering records for each charging station. For each preset time period. The metering records within the time period are merged: when the metering interface directly returns the time-of-use electricity consumption, the last valid electricity consumption reading within the current preset time period is selected as the first... The charging pile is at the first Power consumption readings for a preset time period When the metering interface returns the cumulative energy consumption, the most recent cumulative energy consumption reading before the observation start time is first read as the initial cumulative reading. Among them, the initial cumulative reading Used to determine the starting boundary of the observation and to read the last valid cumulative energy reading within that time period. When n=0, Compared with the initial cumulative reading Perform a score check; when n>0, The effective cumulative electricity reading compared with the previous preset time period Perform a difference analysis and determine the difference result as follows: If there are no valid metering records in the current preset time period, then Recorded as zero, the power consumption readings of all charging piles within the same preset time period are summed to form a time-of-use power consumption sequence. ,in , Indicates the first The time-of-use power consumption of charging stations in the service area within a preset time period.

[0034] For the same set Each charging station reads its timestamped operating status record from the status interface and processes it according to the same preset time period. For each charging pile's operating status within a preset time period, the last valid operating status within that period is selected for statistical analysis. If no valid operating status record exists for the current preset time period, the corresponding charging pile is counted as a faulty pile. Operating statuses capable of accepting charging tasks are counted as available piles. The running status of charging tasks will be included in the number of occupied charging piles. The number of faulty, offline, or non-recorded operating states is included in the number of fault stubs. After arranging the piles by time period and location, the pile status information is written as a three-dimensional vector sequence. ,in , , They represent the first The number of available piles, occupied piles, and faulty piles for each preset time period.

[0035] Read the upper limit parameter of the power supply side capacity of the charging station from the main power supply and distribution configuration of the service area charging station, and record the upper limit parameter of the power supply side capacity of the charging station as a scalar. , This indicates the upper limit of the power supply capacity that a charging station in the current service area is allowed to carry within the same observation period, and it receives a time-varying electricity carbon emission factor sequence corresponding to a unified time axis. , Indicates the first The time-varying carbon emission factor of electricity for a preset time period will and Information corresponding to capacity emissions is combined according to the same preset time period. .

[0036] Then follow the unified index The time-of-use power consumption, charging pile status information, charging station power supply capacity limit parameters, and time-varying electricity carbon emission factor are implemented accordingly, and the first The time period records for each preset time period are written as follows ,in Indicates the first Observation data records for a preset time period, with all time period records categorized as follows: Increasing arrangement forms a length of Observational data sequence Each time period is recorded as a six-dimensional record vector. The six fields correspond to the time-of-use power consumption, the number of available charging piles, the number of occupied charging piles, the number of faulty charging piles, the upper limit parameter of the charging station's power supply side capacity, and the time-varying power carbon emission factor, respectively. Since all fields share the same time location index... The determination of continuous time periods when the power consumption reaches the capacity limit, the calculation of the fallback slope after the full load is released, the construction of the discrete sequence of future capacity, and the calculation of the charging carbon emission sequence can all be directly read according to the same preset time period without the need to perform asynchronous alignment again.

[0037] like Figure 3 As shown, S2, first, start from the observation data according to a unified time position Extracting time-sharing power consumption sequences Available Stake Quantity Sequence Sequence of the number of occupied piles Faulty pile number sequence and the upper limit parameters of the power supply side of the charging station .symbol This indicates the total number of preset time periods covered by the observation period, which is pre-written into the service area charging station operation configuration by the capacity determination threshold. Available stake threshold Occupied stake threshold Fault pile threshold Shortest continuous time period length and fallback analysis window length .

[0038] For each time location Calculate time-of-use power consumption Parameters of the upper limit of the power supply capacity of the charging station The absolute difference, and compare the absolute difference with The pre-set full-load condition is determined by comparing three aspects of the pile state information: , and At that time, determine the location of the time. The preset full-load conditions are met. The absolute difference is no greater than... And time and location When the preset full load condition is met, the time position will be... Write time series that have reached capacity limit And arranged in ascending order of time, the time series that have reached the capacity limit. This corresponds to the time period when the capacity limit is reached in the original steps.

[0039] For time series that have reached their capacity limit Performing continuous merging involves grouping consecutive positions with a time difference of 1 into the same continuous interval, resulting in a continuous interval sequence. ,symbol Indicates the number of consecutive intervals, consecutive intervals of Indicates the starting time period position. Indicates the location of the end time period. According to... and Calculate continuous duration Continuous duration This indicates the preset number of time periods covered by the continuous interval. and In comparison, Preserve continuous intervals The reserved continuous intervals are defined as the continuous periods during which time-of-use power consumption reaches the capacity limit; The corresponding continuous intervals are removed. All retained continuous time periods are arranged in ascending order of their start time position, and the positions of the continuous time periods, the end time periods, and the continuous duration can all be read directly.

[0040] For each consecutive period of retention , with the end time period position Construct a continuous preset time window based on the base point . Window The time-of-use power consumption arranged in ascending order of time is written as: ,symbol Display window The number of time locations that can be compared, excluding the termination time period. Convert the time-of-use electricity consumption of adjacent time locations into a time-of-use electricity consumption change rate sequence. The calculation rule is as follows ,symbol This indicates the preset time period length. Each rate of change... With the corresponding window position index The sample pairs are used to characterize the relationship between the time-of-use power consumption rate and the location of the time period.

[0041] exist At that time, the descent slope after full load will be released. Set to zero; At that time, the descent slope after full load will be released. Set to zero; At that time, a first-order linear least squares fitting is performed on the sample pairs, and the slope of the fitted line is determined as the fallback slope after unloading. The fitting process is written as: , in, To fit the intercept of the straight line, To eliminate the drop slope after full load, This refers to the number of time locations that can be compared within a continuously preset time window. Let be the rate of change of time-of-use electricity consumption in the j-th time period. For window position index, The slope of the fitted line, This represents the baseline intercept component of the linear trend at the beginning of the window. After calculating the drop slope for all continuous time periods, the continuous time periods when the time-of-use power consumption reaches the capacity limit, the continuous duration, and the drop slope after the full load is released are combined into a full load cutoff characteristic segment. Each fully loaded truncated feature segment is written as a four-dimensional feature vector. The first two dimensions of the four-dimensional feature vector together represent the continuous period during which time-of-use power consumption reaches the capacity limit; the third dimension represents the continuous duration; and the fourth dimension represents the drop-off slope after the full load is released. The feature segment truncated under full load is... Increasing arrangement, end time period position Continuous duration and the drop slope after unloading full load It can be directly used for predictive window construction, future capacity discrete sequence generation, and discrete denoising diffusion probability model conditional state construction.

[0042] S3. Record the end position of the continuous period in the full-load truncation characteristic segment where the time-sharing power consumption reaches the capacity limit as... The continuous duration is recorded as The slope of the fall after the full load is removed is denoted as The sequence of available stakes was extracted from the observation data at a uniform time location. and the sequence of the number of faulty piles ,in This represents the total number of preset time periods covered by the observation data. The upper limit parameter of the charging station's power supply side capacity is denoted as... The preset time period length is recorded as The prediction window length is denoted as , zero threshold Fixed as The preset threshold corresponding to the number of faulty piles is denoted as The minimum charge is recorded as In a feasible fixed example, the prediction window length... Pick Minimum battery capacity Pre-configured to be compatible with The corresponding fixed discrete granularity ensures that the number of actually generated discrete locations within a prediction window does not exceed [a certain value]. .

[0043] Construct a prediction window time series based on consecutive preset time periods following the end of a continuous time period, and then... The position of each prediction window period is written as ,in and .according to Increasing arrangement This constitutes a prediction window time series. For each time period in the prediction window time series, the decline process is calculated, and the [number]th [period] is [calculated / determined]. The decline process within each prediction window period is denoted as... The retracement process is calculated using the following rules: , In the formula, Indicates the first The decline process within a single prediction window period; Indicates the prediction window time period number; This indicates the sequence number of the intermediate time period in the cumulative calculation of the retracement process; This indicates the descent slope after the load is released; Indicates the preset time period length; This indicates the upper limit parameter of the power supply capacity of the charging station; Indicates continuous duration; This indicates that the fall direction component is extracted from the fall slope after the load is released; Indicates the distance from the start point of the prediction window to the... The cumulative decline weight for each prediction window period; This indicates that the pullback process will be limited to no more than [a certain value]. Within the interval. According to this calculation rule, the continuous duration The larger the value, the slower the cumulative decline weight increases in the early part of the prediction window; the decline slope after unloading the full load. The larger the directional component of the pullback, the faster the pullback process. When At that time, all Takes a value of zero; when hour, Follow Increasing but monotonically undecreasing. (According to...) Increasing arrangement This constitutes a time-limited decline sequence.

[0044] A recovery power consumption sequence is constructed based on the time-period decline sequence and the charging station's power supply capacity limit parameter. The capacity limit is... As the power consumption level corresponding to the recovery starting point, each prediction window period is calculated using a proportional coefficient. Capacity limit Perform a reduction and record the reduction result as the restored power consumption. ;when At that time, the corresponding power consumption will be restored. Write it as zero. Press Increasing arrangement This constitutes a power consumption recovery sequence. Subsequently, the upper limit parameter of the charging station's power supply side capacity will be... With restoration of power consumption Calculate the difference for each time period and write the difference as the candidate available power capacity. Each With zero threshold In comparison, The corresponding difference is retained as the space to be released. The corresponding difference will be written as zero.

[0045] like Figure 7 As shown, after completing the calculation of candidate available electricity capacity, the electricity is distributed according to the predicted window time period. Read the corresponding number of available stakes from the observation data. and number of faulty piles .Will With zero threshold Comparison, will Preset threshold corresponding to the number of faulty piles Comparison; in and At that time, the filtered acceptable electricity volume is written as and order ;exist or At that time, Write it as zero. Delete in ascending order of time. For a prediction window period of zero, the reserved location index sequence is obtained. ,in Indicates the number of retention periods, and satisfies... For each retention period Write the corresponding capacity as Write the corresponding time period position as The time periods generated according to the retention order are written as follows: .according to Increasingly sorted records The results of the time period arrangement are as follows, among which The three fields represent the available power capacity, time period location, and time period sequence, respectively.

[0046] The time period sequence results are split sequentially based on the minimum electricity consumption. For any record... From the power capacity Repeatedly deducting the minimum power Each time a deduction is completed, a discrete location record is generated, and the corresponding available electricity value for that discrete location is written as follows: At the same time, the source time location is retained. and the order of the source time period When the remaining battery power is greater than the zero threshold and less than At that time, the remaining electricity is written as the last electricity value of the same source time period into a new discrete location record. All discrete location records are concatenated from front to back according to their source time period positions, maintaining the generation order within the same source time period. Each discrete position is written as ,in Indicates the first The amount of electricity that can be received at each discrete location. Indicates the first The time period corresponding to each discrete location. Indicates the first The chronological order of time periods corresponding to each discrete location. , This indicates the actual number of discrete locations generated. (By...) Increasing arrangement This constitutes a discrete sequence of future capacity.

[0047] In a feasible fixed example, the discrete sequence of future capacity is organized into a maximum length of A sequence of discrete positions. When the actual number of discrete positions generated... Less than At that time, Write fill value The first field The second field indicates that discrete locations are not receiving electrical charges. The third field indicates the location code for an invalid time period. This represents the sequence code for an invalid time period. Therefore, the length is... The original input for the future capacity discrete sequence consists of three fields: the capacity to be received, the time period location, and the time period sequence. When constructing the conditional state in the discrete denoising diffusion probability model, the time period location field... Encoded as Time-phase embedding features, thus forming The sequence features are input.

[0048] like Figure 8 As shown, S4 represents the currently processed full-load truncation feature segment as... ,in Indicates the full-load truncation feature segment number. This indicates the starting position of a continuous period during which time-of-use power consumption reaches the capacity limit. This indicates the end point of a continuous period in which time-of-use power consumption reaches its capacity limit. Indicates continuous duration. This represents the drop slope after the load is released from full capacity. The discrete sequence of future capacity is represented as... ,in Represents a discrete location index. , Indicates the first The amount of electricity that can be received at each discrete location. Indicates the first The time period corresponding to each discrete location. Indicates the first The sequence of time periods corresponding to each discrete location. The preset time period length is denoted as... The minimum charge is recorded as The preset threshold corresponding to the number of faulty piles is denoted as First, extract from the fully loaded truncation feature segment. and Then extract them one by one from the discrete sequence of future capacity. and The conditions for formation constitute information.

[0049] like Figure 4 , Figure 5 , Figure 6 As shown, in order to organize the conditional information into an input that can be directly received by the discrete denoised diffusion probability model, the following steps are taken: and Copy the sequence to all time points within a fixed prediction window length to form a sequence. ,in This represents the index of the time position within the prediction window. From all composition The input is the full-load truncation feature segment. The two channels correspond to the continuous duration and the fallback slope after the full load is released, respectively. For each discrete position in the discrete sequence of future capacity, the time period position is read first. Then, the lookup table is used to read the data from the location embedding lookup table. corresponding Dimensional position embedding; sequential reading of time periods Reading and using a sequential embedding lookup table corresponding Dimensional sequence embedding; then Dimensional position embedding and Dimensional sequence embedding is concatenated in a fixed order. Discrete location features. The location embedding lookup table is pre-built according to the location coding range of the time period supported by the system, and the sequential embedding lookup table is... to The time period sequence encoding range is pre-established; when or When the encoding is invalid, both the positional embedding lookup table and the sequential embedding lookup table return the preset invalid encoding embedding. (From all) discrete positions 3D discrete position feature composition The future capacity to accept discrete sequence inputs. Therefore, the original model inputs used to construct the conditional states are explicitly organized into a... Sequence and a sequence.

[0050] Will The fully loaded truncated feature segment is input into the first conditional coding branch, and... The future capacity of the discrete sequence input is fed into the second conditional coding branch. The first conditional coding branch uses a one-dimensional convolutional coding layer and a temporal pooling layer. The one-dimensional convolutional coding layer... Mapped to The time pooling layer converge into The second conditional coding branch also uses a one-dimensional convolutional coding layer and a temporal pooling layer. The one-dimensional convolutional coding layer... Mapped to The time pooling layer converge into Put two The encoded results are concatenated in a fixed order to obtain conditional state vector Conditional state vector The former The dimension is derived from the convolutional coding result of continuous duration and fallback slope after de-loading, and then... The dimension is derived from the convolutional encoding result of the time segment position and the order of the time segments. Conditional state vector. As the state input boundary of the discrete denoising diffusion probability model, it does not contain the time-sharing power consumption field, thus limiting the object generated by the reverse denoising process to a sequence of discrete segments of potential power consumption.

[0051] To ensure that the size of the potential energy discrete segment sequence is directly constrained by the continuous duration and the fallback slope after removing full load, the number of discrete segments is calculated before constructing the initial potential energy discrete segment sequence. Number of discrete segments The calculation rules are as follows: , In the formula, Indicates the first The number of discrete segments corresponding to a fully loaded truncated feature segment; Indicates the sequence number of the full-load truncation feature segment; This represents a discrete position index within a continuous time period, with a value range of [value range missing]. ; Indicates the first The continuous duration of each fully loaded truncated characteristic segment; Indicates the first The slope of the fallback after the full load is released from the full load characteristic segment; Indicates the preset time period length; Indicates the minimum charge level; This represents the directional component of the descent slope after the load is released; Indicates rounding down; This means limiting the number of discrete segments to no less than [a certain value]. ; This means limiting the number of discrete segments to no more than a fixed maximum number of discrete segments. According to this calculation rule, continuous duration Provide the base number of discrete segments for each full-load period, and the descent slope after removing full load. pass Converted to minimum power The quantity correction term with the same dimensions is then weighted by discrete place values. Adjusting the contribution of different full-load periods to the number of discrete segments. Number of discrete segments. After the calculation is completed, the previous Each discrete segment position is considered a valid discrete segment position.

[0052] Based on the number of discrete segments Construct an initial sequence of discrete segments representing potential electrical quantities. Denote the index of the discrete segment location as... .exist At that time, the first The discrete segment energy values ​​are initialized to the minimum energy value. and will the The release priority flag for each discrete segment is initialized to the middle placeholder code. ;exist At that time, the first The discrete segment electrical value of the first discrete segment is written as zero, and the first discrete segment is written as zero. The release priority flag for each discrete segment is also written as the middle placeholder code. .Depend on The length of each discrete segment position is fixed. The initial potential energy discrete segment sequence. The total effective discrete segment energy in the initial potential energy discrete segment sequence is composed of... It consists of a minimum energy unit, used to represent the initial discretization result of the truncated energy corresponding to the full-load truncation feature segment.

[0053] Subsequently, the discrete sequence of future capacity that satisfies... The number of discrete locations is calculated, and the statistical results are recorded as follows: .exist At that time, keep the previous one Each valid discrete segment and its corresponding position; in When the time limit is reached, sequential compression is performed on the excess portion. The sequential compression process is as follows: starting from the first... Starting with a valid discrete segment, from the last to the first valid discrete segment, segments exceeding the specified range are sequentially merged into the immediately preceding valid discrete segment, and the energy value of the merged discrete segment is updated to the accumulated value, until the number of valid discrete segments decreases to a certain value. After sequential compression, the length is... The positions of the discrete segments remain unchanged, and the number of effective discrete segments is limited to the number of effective discrete positions in the future capacity discrete sequence, thus forming a constrained potential power discrete segment sequence.

[0054] condition state vector The discrete segment sequence of limited potential energy is written together with the discrete denoising diffusion probability model in the inverse denoising process. The total number of rounds of inverse denoising is denoised as... and will the The reverse denoising state of the wheel is represented as ,in , , Indicates the first The discrete segment in the th... The discrete segment electrical value category of the wheel, Indicates the first The discrete segment in the th... The release priority of the wheel is identified by its category. The sequence embedding layer will... Depend on Class Embedded Wei, will Depend on Class Embedded Dimensions are then assembled into discrete segments in a fixed order. Wei said that by all A discrete segment is formed Input tensor. The main denoising network of the discrete denoising diffusion probability model adopts a structure of "sequence embedding layer - residual one-dimensional convolution stack - output head". The residual one-dimensional convolution stack includes... There are 10 residual blocks, each containing two convolutional layers with a total of 10 channels. The kernel size is A one-dimensional convolutional layer, with the residual block output dimension maintained as 1. Conditional state vector The residual block is generated after two fully connected transformations. Dimensional scaling vector sum The bias vector is used to perform channel-wise scaling and channel-wise translation on the convolutional output channels of the corresponding residual block. The output header contains two sets of fully connected layers, one output layer and one output layer. The discrete segment electric charge value category distribution, another set of outputs Release priority identifier category distribution. Reverse denoising recursion according to round. Execute round by round: The first round Wheel state With conditional state vector The data is input into the main denoising network, which reads the discrete segment power value category distribution and release priority identifier category distribution corresponding to each discrete segment position. Then, the discrete segment power value category with the highest probability and the release priority identifier category with the highest probability at each discrete segment position are written back to the first discrete segment. Round state, until the first Wheel state. (Number) After class decoding, the round states form a discrete sequence of intermediate potential energy segments. The length of the discrete sequence of intermediate potential energy segments is... Each discrete segment location simultaneously carries the discrete segment power value and release priority identifier.

[0055] The sequence of discrete segments representing intermediate potential energy quantities is subject to retention determination and sorting constraints. First, the energy value category of each discrete segment is decoded into the actual energy value of the discrete segment. Discrete segments with an energy value of zero are deleted. Then, the retained discrete segments are compressed forward according to their original position order, forming a constrained sequence of discrete segments representing potential energy quantities. To write the release priority flag, the truncation feature segment is extracted from the observation data. The corresponding pile status information, and for continuous time intervals The number of available piles, occupied piles, and faulty piles at each time location within the time frame are averaged to obtain the following results. , and ,in This indicates the average number of available stakes. This indicates the average number of stakes occupied. This represents the average number of faulty piles. The initial category of the release priority identifier has already been influenced by the conditional state vector during the reverse denoising process. Constraints, continuous duration and the drop slope after unloading full load The influence has been written into the initial category distribution; based on the initial categories, and then combined with , and Perform an overwrite correction. and At that time, the release priority identifier is uniformly written to the beginning of all valid discrete segments. ;exist At that time, the release priority identifier is uniformly written to the later part of all valid discrete segments. ;exist and At that time, retain the release priority flag category of the output of the reverse denoising process, and exclude those exceeding the limit. to The range category is written back as the median release priority flag. After writing the discrete segment power values ​​and release priority identifier, a length of [length missing] is formed. Candidate potential energy discrete segment sequence ,in , Indicates the first The discrete segment of electrical quantity value. Indicates the first Release priority identifier for each discrete segment.

[0056] like Figure 11 As shown, the horizontal grid sequence in the figure represents time grids and carbon intensity hot zones, with different gray levels representing different carbon intensity levels. Through the alignment and filling results obtained in step S3, a gradually deepening carbon intensity distribution is formed on these time grids, and peak periods are identified. The corresponding grids are highlighted with thick borders. The right-hand dashboard panel displays "Carbon Peak Warning Output," and the predicted peak period t8 is given in the warning period column. The text below explains that a warning signal is triggered and an alarm is issued at this time grid.

[0057] like Figure 9 As shown, in S5, during implementation, the discrete denoising diffusion probability model is implemented in the first... The candidate potential charge discrete segment sequence output at the end of the reverse denoising process is denoted as: ,in Indicates the full-load truncation feature segment number. Indicates the reverse denoising round. Indicates the discrete segment position index. , Indicates the first The discrete segment of electrical quantity value. Indicates the first The release priority identifier for each discrete segment. The discrete sequence of future capacity is denoted as... ,in , Indicates the first The amount of electricity that can be received at each discrete location. Indicates the first The time period corresponding to each discrete location. Indicates the first The sequential order of time periods corresponding to each discrete position. The candidate sequence acceptance stage is fixed between the candidate potential energy discrete segment sequence generation position and the reverse denoising update state writing position, in each round. Simultaneously receive and .

[0058] The candidate sequence acceptance process first deletes invalid discrete segments with a value of zero, and retains valid discrete segments according to their original positions, thus obtaining the number of valid discrete segments. Then, invalid discrete positions with zero capacity are deleted, and valid discrete positions are retained in their original order to obtain the sequence of valid discrete positions. ,in Indicates the number of valid discrete locations. , Indicates the first The amount of electricity that can be received at each effective discrete location. Indicates the first The time period position of each effective discrete location. Indicates the first The chronological order of time periods for each effective discrete location. or At that time, all The time-period attribution path identifier for each discrete segment is written as an invalid path code. The five fields of an invalid path code represent, in order, the start point of a continuous discrete location, the end point of a continuous discrete location, the discrete location, the time period location, and the order of the time periods; then all of them... The time-period attribution embedding features of each discrete segment are written as: A zero-dimensional vector, and the corresponding discrete segment of the current energy value. Dimensional power embedding and current release priority identifier corresponding to Dimension release prioritizes embedding and splicing to form The reverse denoising update state. Only in and Monotonic alignment is performed at that time.

[0059] To directly apply the release priority flag to the monotonic alignment process, the candidate sequence acceptance stage performs a stable sorting on all valid discrete segments. The stable sorting first arranges the segments from front to back according to their release priority flags, and then within the same release priority flag, arranges them from front to back according to their original discrete segment positions, resulting in a stable sorting index. ,in , Indicates the th order after stable sorting The original discrete segment indexes corresponding to each position. After stable sorting, the monotonic alignment cost is calculated for each pair of sorted discrete segments and effective discrete positions, resulting in a monotonic alignment cost sequence. : , In the formula, Indicates the first The fully loaded truncation feature segment in the first In the reverse denoising process, the stable sorting of the first... The discrete segment and the first The monotonic alignment cost between valid discrete locations; Indicates the sequence number of the full-load truncation feature segment; Indicates the reverse denoising round; This represents the index of the discrete segment after stable sorting; Indicates a valid discrete location index; This represents an intermediate index in the summation of cumulative discrete-segment energy values; This indicates an intermediate index in the summation of cumulative capacity. Indicates the th order after stable sorting Discrete segment electrical values; Indicates the th order after stable sorting The original discrete segment index corresponding to each position; Indicates the first The amount of electricity that can be received at each effective discrete location; Indicates the minimum charge level; Indicates the th order after stable sorting Release priority identifier for each discrete segment; Indicates the first The chronological order of the time periods of each effective discrete location; Indicates the chronological order of the last valid discrete position; Indicates the first The number of effective discrete segments in the wheel; Indicates the number of valid discrete locations; and This indicates a preset dimensionless cost weight; This indicates that cumulative summation is performed based on the index range; Represents absolute value; This indicates that the lower limit is set in the denominator. The larger of the values ​​is used for calculation. According to this calculation rule, the first item gives the main matching difference between the cumulative discrete segment power value and the cumulative load-bearing power value; the second item gives the order difference between the release priority identifier and the order of the effective discrete position time periods; and the third item gives the position difference between the stable sorting progress and the effective discrete position progress.

[0060] The candidate sequence acceptance process will Write differentiable, dynamically time-warped, monotonically aligned discrete segment indices sorted by stability. and effective discrete location index Construct a grid, allowing only three predecessor grids (left, top, and top-left) to enter the current position at each grid location. The cumulative cost at each grid location is determined by the monotonic alignment cost of the current location and the cumulative costs of the three predecessor grids. The three predecessor cumulative costs are exponentially smoothed using a preset smoothing coefficient to ensure the cumulative path remains differentiable. After constructing the cumulative cost table, backtrack from the bottom-right grid to the top-left grid to obtain an alignment path that satisfies the condition that the order of arrangement cannot be reversed. Then, construct an alignment weight matrix based on the backtracking path. ,in Indicates the first The discrete segment pairs of the first Alignment weights for each discrete position. For path units belonging to valid discrete segments and valid discrete positions, positive weights are first written, and then normalized within the row corresponding to the same discrete segment, so that the sum of all alignment weights in the same row is 1. For rows corresponding to invalid discrete segments and columns corresponding to invalid discrete positions, zero values ​​are uniformly written. This yields a monotonically aligned weight sequence whose arrangement order changes monotonically along the discrete sequence of future capacity.

[0061] Based on the monotonically aligned weight sequence, the candidate sequence acceptance stage extracts continuous non-zero weight intervals for each discrete segment. The first... The starting discrete position of each discrete segment of continuous non-zero weight interval is denoted as . The termination discrete position is denoted as The discrete position with the largest alignment weight in the same row is denoted as , and then Corresponding time period location Recorded as ,Bundle The corresponding time period sequence Recorded as The initial time period attribution path identifier is formed by five fields. Then sort the index according to stability. The order of the path identifiers belonging to all initial time periods is compared. The representative discrete position of the current discrete segment in the stable sorting is then determined. If the representative discrete position of the current discrete segment is earlier than the previous representative discrete position in the stable sorting, the current discrete segment is scanned backward along the direction of the effective discrete position, and the representative discrete position is updated to the first effective discrete position no earlier than the previous representative discrete position. Simultaneously, the... and Synchronous translation. After sequential comparison, all discrete segment power values ​​are mapped to representative discrete positions, and the power values ​​of discrete segments mapped to the same representative discrete position are accumulated. When the accumulated result at a certain representative discrete position is greater than the corresponding load-bearing power... At this time, the discrete segments with the highest release priority identifier are retained at the current position, and the discrete segments with the lowest release priority identifier are adjusted sequentially to the next consecutive valid discrete positions according to the last position in the stable sorting. After each adjustment, the cumulative results at the target valid discrete positions are re-compared with the corresponding available power capacity until all valid discrete positions meet the capacity constraints or the last valid discrete position is scanned. After completing the sequence comparison and capacity comparison, the time period belonging path identifier is obtained. .

[0062] Time period belonging path identifier When writing the candidate potential energy discrete segment sequence, first retain the potential energy discrete segment sequence with time period attribution in plaintext field form. ,in The three fields represent the discrete segment power value, release priority identifier, and time period belonging path identifier, respectively. Then, the encoded form required for writing the reverse denoising update state is constructed. For each valid discrete location... First, the time period location Input position embeds lookup table, and gets Position embedding; then chronological order of time periods. Input sequence embedding lookup table, get Dimensional order embedding; then put the two Dimensional embedding is assembled in a fixed order. 3D time-period embedding features Then, according to the alignment weight matrix... For all Perform row-weighted aggregation to obtain the first row. discrete segments Time-segmented embedding features When the first When each discrete segment is an invalid discrete segment Write it directly as Zero-dimensional vector.

[0063] To write the discrete segment sequence of potential electricity consumption at different time periods into the inverse denoising update state of the discrete denoising diffusion probability model, embedding is then performed on the discrete segment electricity values ​​and release priority identifiers. First, according to the steps already determined in S4... The coding rules for discrete segment electricity values ​​will Mapped to a battery category code, and then the corresponding code is retrieved through a battery embedding lookup table. Dimensional power embedding; then... As the release priority identifier category code, the corresponding value is read from the release priority embedding lookup table. Dimensional release prioritizes embedding. Then the first... discrete segments Dimensional power embedding, the first discrete segments Dimension release priority embedding and the first discrete segments The time-period attribution embedding features are concatenated in a fixed order to obtain the 1st time-period attribution embedding features. discrete segments Time period attribution characteristics From all discrete segments Dimensional Time Period Attribution Characteristics The reverse denoising update state. In the reverse denoising update state, each row explicitly contains three components: the previous... Dimension represents the embedding of discrete segment electrical values, in the middle The dimension indicates that the embedding is prioritized for release, followed by... The dimension represents the embedding of time periods.

[0064] In the In the reverse denoising process, the discrete denoising diffusion probability model is read. The reverse denoising update state is performed while maintaining the conditional state vector constructed in step S4. constant, The dimension is The main denoising network employs a "residual one-dimensional convolution stack-output head" operation mode to handle the reverse denoising update state. The residual one-dimensional convolution stack includes... There are 10 residual blocks, each containing two convolutional layers with a total of 10 channels. The kernel size is One-dimensional convolutional layer, conditional state vector Generated through two fully connected layers Dimensional scaling vector sum A bias vector is used to perform channel-wise scaling and channel-wise translation on the convolution output of each residual block. The output header holds two sets of outputs, one set of outputs and one set of outputs. The discrete segment electric charge value category distribution, another set of outputs The release priority identifier category distribution. The candidate sequence acceptance phase is in the... The second round reads the two sets of category distributions again, regenerates the candidate potential energy discrete segment sequence, and re-executes the following processes: effective discrete segment filtering, effective discrete position filtering, stable sorting, monotonically aligned cost sequence calculation, differentiable dynamic time warping monotonically aligned, initial time period belonging path identifier generation, order comparison, capacity comparison, writing of the time period belonging potential energy discrete segment sequence, and reverse denoising update status writing. The reverse denoising round decreases to the [number missing]th round. During the turn, retain and And retain the first Wheel corresponding Reverse denoising to update the status.

[0065] like Figure 10 As shown, S6, will the first The time period reserved by the round is written as the discrete sequence of potential electricity. The position of each discrete segment is recorded as follows: The prediction window time series is written as... ,in In a fixed example, Indexed by discrete segment position from arrive Read one by one , and The fourth field . Indicates the first The discrete segment of electrical quantity value. Indicates the first Release priority identifier for each discrete segment Indicates the first The time period position corresponding to each discrete segment. When When, no time period summary item is generated; when When, no time period summary item is generated; when and At that time, a time period-specific summary item is generated. Time Period Category Summary Item The three fields represent the discrete segment power value, release priority identifier, and time period location, respectively.

[0066] because The data is directly derived from the time period position field in the discrete sequence of future capacity, which has already been formed by splitting the prediction window time period sequence. Time period matching employs a strict correspondence rule. For each prediction window time period position... Check each of the generated time period category summary items one by one. , to satisfy All time period categories are summarized under the first category. Grouped by time period, forming . Indicates time period location The corresponding time period electricity grouping results. If no matching is found... The time period belongs to the summary item, then This is recorded as an empty group. If a certain time period belongs to a specific time period in the summary item... If a time period does not belong to any time period position in the forecast window time period sequence, the current time period's aggregation item will not be written into any time period group. The system generates time-period power consumption groupings based on the predicted window time period. Each time period group retains only the discrete segment power consumption value that needs to be accumulated in the current time period and the release priority identifier.

[0067] Grouping each non-empty time period Perform group sorting. Group sorting is first based on release priority identifier from front to back, with the release priority identifier as the first priority identifier. Priority release marker Priority flags to release Priority release marker The preceding text appears to be a fragmented list of keywords or tags, possibly related to a release priority flag. It doesn't translate coherently without further context or clarification. A more accurate translation would require the original, unrelated snippets. Arrange from smallest to largest. After permutation within the group, the [number]th... The discrete electricity values ​​in each time period group are arranged in an ordered manner, and the time period electricity grouping results are transformed from a collected state to an ordered state that can be directly accumulated. Empty groups are not sorted within the group and remain empty.

[0068] Before accumulating the electricity released during a time period, the electricity release sequence for each time period is initialized according to the position of the prediction window time period. and put all Write it as zero. Then group each non-empty time period. Read the discrete segment electricity values ​​item by item according to the sorting results within the group, and accumulate the read discrete segment electricity values ​​into the corresponding... In the case of multiple discrete energy values ​​within the same time period group, these values ​​are summed to represent the total energy released within the same time period. If the time period group is empty, the corresponding... Keep the value zero. After cumulative processing of all time periods, the time period release electricity sequence corresponding to the prediction window time period sequence is obtained. The first time period release electricity sequence is... The position corresponds only to the time period. The released power will not be used to perform any position rearrangement.

[0069] Will according to Arrange the data from smallest to largest to obtain the complete time series results. Each sequence position in the complete time series retains the original time order within the prediction window time period sequence; time periods with a time period aggregation item are written with the accumulated released electricity, while time periods without a time period aggregation item are written with zero values. The delayed release distribution is written as... and the prediction window time series Maintain the correspondence. In the fixed example, the delayed release distribution is as follows: dimensional sequence, the first Dimension represents the position of the time period within the prediction window. The amount of electricity released.

[0070] like Figure 12As shown, the left panel is a raster plot of the predicted mean. The space is divided into 10×10 grids with time as the horizontal axis and capacity level as the vertical axis. The gray level of each grid represents the predicted mean load or carbon intensity under the corresponding time-capacity combination; the darker the gray level, the higher the value. The "peak grid" marked with a thick border indicates the peak position in the predicted mean. The right panel is a plot of the predicted uncertainty, using the same time-capacity raster but encoding the uncertainty level with different texture densities: no texture indicates low uncertainty, sparse diagonal lines indicate medium uncertainty, and dense diagonal lines indicate high uncertainty.

[0071] S7. Write the delayed release distribution obtained in step S6 as follows: ,in Indicates the full-load truncation feature segment number. Indicates the index of the prediction window period. Indicates the first The released electricity within each prediction window period. The prediction window period sequence is written as follows: ,in Indicates the first The time period corresponding to each prediction window. In a fixed example, From the observation data, sorted by time period and location. Read the time-varying electricity carbon emission factor value corresponding to each prediction window period one by one, and record it as ,in Indicates time period location The corresponding time-varying electricity carbon emission factor value. Since the observation data has already been time-corresponded according to a unified preset time period in step S1... and The mapping is performed using locations with identical time intervals, without interpolation or cross-time interval compensation. After mapping is complete, the [number]th [time interval] will be [value]. The combination of electricity released, factor value, and time period location for each time period constitutes the carbon emission calculation item for that time period. ,in The three fields represent, in order, the released electricity volume, the time-varying electricity carbon emission factor value, and the time period. (By...) Arranged from smallest to largest This constitutes a sequence of carbon emission calculation items ordered chronologically by time period.

[0072] Carbon emission calculation items for each time period Perform time-by-time product calculation. and Multiply to get the first... Carbon emissions for each time period are denoted as ,in Indicates time period location The corresponding carbon emissions. For all Repeat this process to obtain the carbon emission sequence for charging. The carbon emission sequence for charging maintains a completely consistent index order with the prediction window time period sequence, the first... The position corresponds only to the time period. The carbon emissions are not rearranged. In the fixed example, the carbon emission sequence for charging is... dimensional time series, the first Dimension indicates the location of the time period. Carbon emissions; when When, corresponding Write it directly as zero; when When, corresponding It is also written as zero.

[0073] After the charging carbon emission sequence is generated, the difference is calculated for adjacent time periods. For each Using current carbon emissions Subtract carbon emissions from the previous period The change in carbon emissions is obtained and denoted as ,in Indicates the time period position Time period location The change in carbon emissions. (From all) according to Arranged from smallest to largest, forming a sequence of carbon emission changes. Index of each intermediate period in the carbon emission change sequence. Perform peak candidate filtering: when and At that time, index the time period. Indexes identified as peak candidate periods; when or At that time, do not index the time period. The indexes of the time periods selected as peak candidate periods are as follows. The first and last time periods, because they do not simultaneously possess both preceding and subsequent directions of change, are not included in the peak candidate selection. All time period indices that meet the selection criteria are arranged chronologically. ,in Indicates the candidate index of the peak. Indicates the number of candidate peak periods. Indicates the first The index position of each peak candidate period in the prediction window time series.

[0074] Index for each peak candidate period Read the corresponding carbon emissions and corresponding time period location The two pieces of information are combined into peak value determination information. ,in The first field represents the carbon emissions corresponding to the peak candidate period, and the second field represents the time period location corresponding to the peak candidate period. The preset warning threshold is denoted as... .

[0075] For each peak determination information Perform threshold comparison: when At that time, the warning sign will be written as and write the time period location field as ;when At that time, the warning sign will be written as And set the time period location field to an invalid location code. .

[0076] Then , and Combined into a peak carbon emission warning result ,in Indicates a warning sign. This indicates a time period location field. This indicates the corresponding carbon emissions. When... At that time, write directly This indicates that there are no carbon emission peak warning results within the current prediction window that meet the peak candidate screening criteria and include a valid time period location.

[0077] Arranged in chronological order This generates a peak carbon emission warning result corresponding to the current full-load cutoff characteristic segment.

[0078] The above are all preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Therefore, all equivalent changes made in accordance with the structure, shape and principle of the present invention should be covered within the scope of protection of the present invention.

Claims

1. A method for early warning of carbon emission peaks based on a discrete diffusion model and dynamic regularization, characterized in that, include: The system acquires time-of-use power consumption, charging pile status information, charging station power supply capacity limit parameters, and time-varying electricity carbon emission factors of charging stations in the service area according to preset time periods, forming observation data. Based on the observation data, determine the continuous time periods when the time-of-use power consumption reaches the capacity limit and calculate the drop slope after the full load is released to generate the full load cutoff characteristic segment. Based on the full-load truncation characteristic segment, the electricity that can be received in each time period within the prediction window is calculated and discretized according to the minimum electricity, so as to obtain the discrete sequence of future capacity. The conditional state of the discrete denoising diffusion probability model is constructed by using the full-load truncation feature segment and the discrete sequence of future capacity, and a candidate potential power discrete segment sequence is generated. Monotonic alignment is performed between the candidate potential power discrete segment sequence and the future capacity discrete sequence to obtain the time period belonging path identifier, which is then written into the candidate potential power discrete segment sequence to form the time period belonging potential power discrete segment sequence. The discrete sequence of potential electricity generation time periods is aggregated by time period to form a delayed release distribution; The charging carbon emission sequence is calculated based on the delayed release distribution and time-varying electricity carbon emission factor, and the peak carbon emission warning result is output.

2. The carbon emission peak early warning method based on discrete diffusion model and dynamic regularization according to claim 1, characterized in that, The formation of observation data specifically includes: The power consumption readings are obtained from the metering interface of each charging pile in the service area charging station according to the preset time period, and the power consumption readings of each charging pile in the same preset time period are summed to form a time-sharing power consumption sequence according to the order of the preset time period. The operating status is obtained from the status interface of each charging pile in the service area charging station according to a preset time period, and the operating status is converted into the number of available piles, the number of occupied piles, and the number of faulty piles to form pile status information. Obtain the upper limit parameter of the power supply side of the charging station corresponding to the charging station in the service area, and receive the time-varying power carbon emission factor corresponding to the preset time period to form capacity emission corresponding information; The time-of-use power consumption sequence, charging pile status information, and capacity emission corresponding information are time-correlated according to the same preset time period. The time-of-use power consumption, charging pile status information, charging station power supply side capacity limit parameters, and time-varying electricity carbon emission factors within the same preset time period are combined into time period records to form observation data.

3. The carbon emission peak early warning method based on discrete diffusion model and dynamic regularization according to claim 1, characterized in that, The process of generating a full-load truncation feature segment includes extracting time-sharing power consumption and pile status information, comparing the time-sharing power consumption with the upper limit parameter of the charging station's power supply side capacity in each time period, and determining the time period sequence that reaches the upper limit when the time-sharing power consumption reaches the upper limit parameter of the charging station's power supply side capacity and the pile status information meets the preset full-load conditions. The time-series data are continuously merged according to time sequence to determine the consecutive time periods when the time-of-use power consumption reaches the capacity limit; Extract the time-of-use power consumption within the consecutive preset time period after the end of the continuous time period, calculate the change rate of time-of-use power consumption between adjacent time periods, and determine the drop slope after the full load is released based on the relationship between the change rate of time-of-use power consumption and the time period position. The full-load truncation feature segment is generated by combining the continuous time period, its continuous duration, and the drop slope after the full load is released.

4. The carbon emission peak early warning method based on discrete diffusion model and dynamic regularization according to claim 1, characterized in that, The calculation of the amount of electricity that can be received in each time period within the prediction window includes constructing a prediction window time period sequence based on the full load truncation feature segment and a series of consecutive preset time periods after the end position of the continuous time period. Each time period in the prediction window time period sequence is arranged in an ascending order of time. Based on the continuous duration L and the descent slope Calculate the decline process of each time period in the prediction window time series relative to the end position of the continuous time period. And accumulate them according to the time period position to form a time period decline sequence; Based on the time period decline sequence and the upper limit parameter of the charging station's power supply side, the upper limit of the capacity corresponding to the end position of the continuous time period is taken as the recovery starting point, and the recovery power consumption is calculated time period by time along the prediction window time period sequence to form a recovery power consumption sequence; The candidate acceptable power consumption sequence is obtained by calculating the time-period difference between the capacity limit parameter and the recovery power consumption sequence. The calculation formula for the fallback process is as follows: ， in, For the first The decline process within a prediction window period, For window position index, To eliminate the drop slope after full load, This refers to the intermediate time period number in the cumulative calculation of the decline process. For the preset time period length, This refers to the upper limit parameter for the power supply capacity of the charging station.

5. The carbon emission peak early warning method based on discrete diffusion model and dynamic regularization according to claim 4, characterized in that, The process of obtaining the discrete sequence of future carrying capacity includes extracting pile status information corresponding to the prediction window time period sequence from the observation data and setting a zero threshold. and preset threshold The pile status information is compared with the set threshold. and By comparing the data, a sequence of acceptable power volumes can be formed; The available power capacity sequence is filtered by time period order and rearranged according to the time sequence of the prediction window time period sequence, and the time period position is written for each reserved time period. and the order of time periods This forms the result of time period succession arrangement; The available electricity capacity for each time period in the time period arrangement results is ranked by minimum electricity capacity. The data is split sequentially, and the source time period position and source time period order are written to form the minimum power arrangement result with time period position; The minimum power consumption results with time period positions are concatenated from front to back according to the source time period position, while maintaining the arrangement order within the same source time period, so that each discrete position retains the corresponding time period position and time period sequence, generating a discrete sequence of future capacity.

6. The carbon emission peak early warning method based on discrete diffusion model and dynamic regularization according to claim 1, characterized in that, The generation of the candidate potential energy discrete segment sequence includes extracting continuous duration. The slope of the drop after the full load is removed Time period and location and the order of time periods After sequential encoding, the conditional state vector is constructed by concatenating the data in a fixed order. ; according to and Calculate the number of discrete segments Based on minimum power Assign discrete segment positions to each discrete segment to form an initial potential energy discrete segment sequence, and determine the number of discrete segments. Greater than the number of effective discrete positions The excess portion is sequentially compressed to form a sequence of discrete segments of limited potential power. Will The reverse denoising process of the discrete denoising diffusion probability model, which is input together with the constrained potential power discrete segment sequence, limits the generated object to the potential power discrete segment sequence. After denoising update and sorting constraints, a constrained potential power discrete segment sequence is formed. For each discrete segment in the sequence of constrained potential electrical quantities, determine the discrete segment electrical quantity value, and combine it with... , and Write a release priority identifier for each discrete segment to form a sequence of candidate potential power discrete segments; in, For continuous duration, To eliminate the drop slope after full load, In chronological order of time periods This is the conditional state vector. The number of discrete segments. Minimum power consumption This indicates the time period and location.

7. The carbon emission peak early warning method based on discrete diffusion model and dynamic regularization according to claim 1, characterized in that, The formation of the potential power discrete segment sequence for the time period includes inputting the candidate potential power discrete segment sequence and the future capacity discrete sequence into the candidate sequence receiving stage, and calculating the monotonic alignment cost sequence based on the matching difference between the power value of each discrete segment and the power capacity that can be received at each discrete location and the order constraint corresponding to the release priority identifier. Based on the monotonic alignment cost sequence, the alignment weights corresponding to each discrete position of each discrete segment are calculated, and the alignment weights are kept to change monotonically along the arrangement order of the future capacity discrete sequence to form a monotonic alignment weight sequence. Perform differentiable dynamic time warping monotonic alignment on the monotonically aligned weight sequence to map each discrete segment to a continuous discrete position in the future capacity discrete sequence, thus obtaining the time period belonging path identifier; The time period-assigned path identifier is written into the candidate potential electricity discrete segment sequence to form the time period-assigned potential electricity discrete segment sequence, and then written into the inverse denoising update state of the discrete denoising diffusion probability model.

8. The carbon emission peak early warning method based on discrete diffusion model and dynamic regularization according to claim 7, characterized in that, The process of forming a discrete segment sequence of potential electricity for a given time period also includes mapping the electricity value of each discrete segment to a corresponding discrete location according to the time period path identifier, and performing cumulative calculation on the electricity values ​​of discrete segments mapped to the same discrete location. The cumulative calculation result is compared with the capacity of the corresponding discrete position. If the cumulative calculation result is greater than the corresponding capacity, the capacity of the discrete segment located later is adjusted to the subsequent consecutive discrete positions according to the order of the release priority identifier. If the cumulative calculation result is not greater than the corresponding capacity, the current time period's belonging path identifier is retained. The time period attribution path identifier is updated according to the adjusted discrete location, and the updated time period attribution path identifier is written into the candidate potential power discrete segment sequence to form the time period attribution potential power discrete segment sequence.

9. The carbon emission peak early warning method based on discrete diffusion model and dynamic regularization according to claim 1, characterized in that, The process of forming a delayed release distribution includes extracting the discrete segment power value, release priority identifier, and corresponding time period position of each discrete segment from the discrete segment sequence of potential power segments belonging to the time period; grouping discrete segments with the same time period position into the same time period group; and arranging the discrete segment power values ​​in each time period group according to the order of the release priority identifier to form the time period power grouping result. The cumulative calculation is performed on the discrete segment electricity values ​​within each time period in the time period electricity grouping results. The electricity values ​​of multiple discrete segments within the same time period are accumulated to form the electricity released in the corresponding time period. The electricity released in the time period that does not match a discrete segment is set to zero, thus forming a time period electricity release sequence. The time-series electricity release sequence is reconstructed into a complete time series result according to the chronological order of each time period within the prediction window, thus obtaining the delayed release distribution.

10. The carbon emission peak early warning method based on discrete diffusion model and dynamic regularization according to claim 1, characterized in that, The output carbon emission peak warning result includes matching the released electricity in each time period of the delayed release distribution with the factor value of the corresponding time period in the time-varying power carbon emission factor according to the same time period position, multiplying them one by one to obtain the carbon emission amount of each time period, and arranging them in the order of time periods to form a charging carbon emission sequence. The carbon emission difference between adjacent time periods in the charging carbon emission sequence is calculated to obtain the carbon emission change sequence. Based on the comparison condition that the change direction of the previous time period is upward and the change direction of the next time period is downward, the peak candidate time period is screened. The carbon emission corresponding to the peak candidate time period is combined with the time period position to form the peak determination information. The carbon emissions in the peak determination information are compared with the preset warning threshold. When the carbon emissions meet the comparison condition of the preset warning threshold, the carbon emission peak warning result containing the location of the corresponding time period is output.

Images