Power distribution network load prediction system based on real-time power transaction

Abstract

The invention relates to a power distribution network load prediction system based on real-time power transaction, and belongs to the technical field of power distribution systems. Comprising an acquisition module for acquiring historical power transaction data and historical power load data; the data fusion module is used for extracting a power transaction feature sequence of historical power transaction data and a parameter association weight of the power transaction feature sequence, and performing multi-modal data fusion on the power transaction feature sequence and the parameter association weight of the power transaction feature sequence to generate transaction fusion data; the time alignment module performs time alignment on the transaction fusion data and historical power load data to obtain a time sequence set and construct a transaction-load data curve; the load sensitive factor fitting module is used for fitting a load sensitive factor and a transaction-load data curve to obtain a power distribution network load prediction model; and the real-time prediction module is used for determining a power load prediction result according to the power real-time transaction data and the power distribution network load prediction model. The method can reduce the prediction error, and improves the accuracy and efficiency of the power distribution network load prediction result.

Description

Technical Field

[0001] The present invention relates to the technical field of power distribution systems, and in particular to a distribution network load forecasting system based on real-time power trading. Background Art

[0002] With the deepening of power market reform, the power industry is gradually evolving from a traditional planned economy to a market-oriented and intelligent one. In this process, real-time power trading and intelligent management of distribution networks are key to improving economic benefits and power system operational efficiency. As a core component of the power system, distribution network load forecasting is crucial for optimizing power dispatch and improving power system stability.

[0003] However, current research on power trading and distribution network loads has focused solely on their respective fields, without establishing a correlation between the two. Furthermore, most current distribution network load forecasts fail to adequately consider load-sensitive factors (such as weather and holidays), while also ignoring the dynamic impact of the power trading market. Most distribution network load forecasting models rely solely on historical load time series, resulting in large prediction errors and inaccurate prediction results.

[0004] Therefore, the present invention provides a distribution network load forecasting system based on real-time power trading. Summary of the Invention

[0005] To solve the above technical problems, the present invention provides a distribution network load forecasting system based on real-time power trading. The technical solution of the present invention is as follows: Acquisition module: used to obtain historical power transaction data of the target power trading market and historical power load data of the target distribution network based on the preset power transaction interface and the preset distribution network interface of the target distribution network; Data fusion module: used to extract the power transaction feature sequence of each parameter type from historical power transaction data, obtain the parameter association weight of the power transaction feature sequence of each parameter type, and perform multimodal data fusion on the power transaction feature sequence of each parameter type and its parameter association weight to generate transaction fusion data; Time alignment module: used to time-align transaction fusion data with historical power load data to obtain a time series set, and construct a transaction-load data curve based on the time series set; Load sensitivity factor fitting module: used to fit the load sensitivity factors in the preset sensitivity factor table to the transaction-load data curve to obtain the distribution network load forecasting model; Real-time prediction module: used to obtain real-time power transaction data of the target power trading market based on the preset power transaction interface, and determine the current power load forecast result of the target distribution network based on the real-time power transaction data and the distribution network load forecast model.

[0006] Preferably, the acquisition module includes: Power transaction data acquisition unit: used to acquire historical power transaction data of the target power transaction market based on a preset power transaction interface according to a preset historical interval range and a preset parameter type of power transaction data; The power load data acquisition unit is used to acquire the historical power load data of the target distribution network based on the preset distribution network interface of the target distribution network according to the preset historical interval range.

[0007] Preferably, the data fusion module includes: Standardization unit: used to standardize the historical power transaction data of each parameter type through statistical methods to obtain standard power transaction data of each parameter type; A feature extraction unit is configured to, for standard power trading data of any parameter type, use a sliding window of a preset length, with the preset length as a sliding step, slide the sliding window from the start time node to the end time node of the standard power trading data of the parameter type within the standard power trading data of the parameter type, obtain the window feature of each time node in the sliding step of each sliding, and synthesize the window features of all time nodes to obtain a power trading feature sequence of the standard power trading data of the parameter type; Parameter association weight calculation unit: used to calculate the parameter association weight of the power transaction feature sequence of each parameter type relative to the power transaction feature sequences of other parameter types based on the power transaction feature sequence of each parameter type and its standard power transaction data; Multimodal data fusion unit: used to input the power transaction feature sequence of each parameter type and its parameter associated weights into the pre-trained data fusion recurrent neural network in sequence according to the preset parameter type order, and the data fusion recurrent neural network outputs the transaction fusion data; wherein, the power transaction feature sequence of the next parameter type specified by the single modality and the preset parameter type order output by the data fusion recurrent neural network last time and its parameter associated weights are the next input of the data fusion recurrent neural network, until the power transaction feature sequences of all parameter types are input into the data fusion recurrent neural network, the cycle ends and the transaction fusion data is obtained.

[0008] Preferably, the feature extraction unit is used to obtain the power transaction feature sequence H of the standard power transaction data of any parameter type through formula (1):

[0009] In formula (1), represents a cross-window aggregate function, represents the feature extraction function, represents the i1th time node, represents the standard power transaction data within the sliding window starting from time node t and ending at time node t+r, and , r represents the preset length of the sliding window, Indicates the total number of windows, Indicates the start time node of the standard power transaction data of the parameter type, Indicates the end time node of the standard power transaction data of the parameter type, Represents the time node obtained based on the sliding window Window characteristics.

[0010] Preferably, the parameter association weight calculation unit is used to calculate the parameter association weight of the power transaction feature sequence of the i-th parameter type relative to the power transaction feature sequences of other parameter types by formula (2): :

[0011] In formula (2), m represents the number of parameter types. represents the covariance function, represents the power transaction feature sequence of the i-th parameter type, represents the power transaction feature sequence of the jth parameter type, Indicates the first parameter type of the i-th Standard electricity transaction data at each time node, Indicates the jth parameter type Standard electricity transaction data at each time node, represents the mean of the standard power transaction data of the i-th parameter type, represents the mean value of the standard power transaction data of the jth parameter type, represents the correlation coefficient between the standard power transaction data of the i-th parameter type and the standard power transaction data of the j-th parameter type, Represents the normalized weight function, Indicates the preset influence coefficient.

[0012] Preferably, the time alignment module includes: Time alignment unit: used to standardize historical power load data to obtain standard power load data, compare the time granularity of transaction fusion data and standard power load data with the preset standard time granularity, perform linear interpolation processing on transaction fusion data and standard power load data with a time granularity greater than the preset standard time granularity, and upwardly aggregate transaction fusion data and standard power load data with a time granularity less than the preset standard time granularity based on statistical characteristics to obtain time-aligned aligned transaction fusion data and aligned standard power load data, and generate a time series set of the aligned transaction fusion data, aligned standard power load data and aligned time nodes; Curve construction unit: used to construct a transaction-load data curve based on a time series set, with the horizontal axis being aligned with transaction fusion data and the vertical axis being aligned with power load data.

[0013] Preferably, the load sensitive factor fitting module includes: Sensitive label adding unit: used to match the time interval corresponding to any type of load sensitive factor in the preset sensitive factor table with each aligned time node in the time series set. If any aligned time node successfully matches the time interval corresponding to the load sensitive factor of the type, a sensitive label of the load sensitive factor of the type is added to the aligned time node; a label time series set is generated based on all aligned time nodes and their sensitive labels, aligned transaction fusion data, and aligned standard power load data; a core weight determination unit configured to, for any alignment time node with a sensitive label, use the sensitive weight of the load sensitivity factor corresponding to the alignment time node with the sensitive label as the core weight of the alignment time node with the sensitive label; if the alignment time node with the sensitive label has multiple sensitive labels, use the maximum value of the sensitive weights corresponding to all the sensitive labels of the alignment time node with the sensitive label as the core weight of the alignment time node with the sensitive label; wherein the core weight of an alignment time node without a sensitive label is determined to be 0; The function fitting unit is used to use the transaction-load data curve as a constraint condition, fit the kernel weight and constraint condition of each aligned time node based on the kernel regression model, and obtain the distribution network load forecasting model.

[0014] Preferably, the real-time prediction module is used to: obtain the current real-time electricity trading data of the target electricity trading market based on a preset electricity trading interface, and input the real-time electricity trading data and the current timestamp into the distribution network load prediction model, and determine the current electricity load prediction result of the target distribution network according to the output of the distribution network load prediction model.

[0015] All the above optional technical solutions can be combined arbitrarily, and the present invention does not provide detailed descriptions of the structures after each combination.

[0016] By means of the above solution, the beneficial effects of the present invention are as follows: By obtaining historical power transaction data and historical power load data, and then performing multimodal data fusion on the power transaction feature sequence of each parameter type of the historical power transaction data and its parameter association weights, the transaction fusion data and the historical power load data are time-aligned and a transaction-load data curve is constructed based on the time series set generated by the time alignment. The load sensitivity factor is fitted into the transaction-load data curve to construct a distribution network load forecasting model. A model-parameterized distribution network load forecasting model is provided. The distribution network load forecasting model establishes a correlation between historical power transaction data and historical power load data, and introduces a time-related load sensitivity factor. When predicting the current power load of the target distribution network, the correlation between the power transaction data and the power load data and the load sensitivity factor are simultaneously considered, thereby reducing the prediction error and improving the accuracy and efficiency of the distribution network load forecasting results.

[0017] The above description is only an overview of the technical solution of the present invention. In order to more clearly understand the technical means of the present invention and implement it according to the contents of the specification, the following is a detailed description of the preferred embodiments of the present invention with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 It is a schematic diagram of the composition structure of a distribution network load forecasting system based on real-time power trading provided by an embodiment of the present invention.

[0019] Figure 2 It is a structural diagram of the data fusion recurrent neural network in the present invention. DETAILED DESCRIPTION

[0020] The following embodiments of the present invention are described in further detail with reference to the accompanying drawings and examples. The following examples are used to illustrate the present invention but are not intended to limit the scope of the present invention.

[0021] Example 1: Figure 1 As shown, the distribution network load forecasting system based on real-time power trading provided by the embodiment of the present invention includes: Acquisition module: used to obtain historical power transaction data of the target power trading market and historical power load data of the target distribution network based on the preset power transaction interface and the preset distribution network interface of the target distribution network; Data fusion module: used to extract the power transaction feature sequence of each parameter type from historical power transaction data, obtain the parameter association weight of the power transaction feature sequence of each parameter type, and perform multimodal data fusion on the power transaction feature sequence of each parameter type and its parameter association weight to generate transaction fusion data; Time alignment module: used to time-align transaction fusion data with historical power load data to obtain a time series set, and construct a transaction-load data curve based on the time series set; Load sensitivity factor fitting module: used to fit the load sensitivity factors in the preset sensitivity factor table to the transaction-load data curve to obtain the distribution network load forecasting model; Real-time prediction module: used to obtain real-time power transaction data of the target power trading market based on the preset power transaction interface, and determine the current power load forecast result of the target distribution network based on the real-time power transaction data and the distribution network load forecast model.

[0022] In this embodiment, the preset power trading interface refers to a pre-defined interface for reading power trading data of the target power trading market, which can read historical power trading data of the target power trading market and real-time power trading data of the target power trading market.

[0023] In this embodiment, the preset power distribution network interface refers to a predefined interface for reading power load data of a target power distribution network.

[0024] In this embodiment, the target distribution network refers to a specific distribution network area, which is a power distribution system corresponding to the target power trading market.

[0025] In this embodiment, the target power trading market refers to a market for power trading within a specific region or country. These markets are used for power trading between power suppliers, power demanders, and other related parties (such as grid operators, independent power producers, etc.).

[0026] In this embodiment, historical power transaction data refers to the relevant information and status records of power transactions that occurred in the target power transaction market within a period of time in the past, and the types include: transaction time, transaction volume, power price, transaction type, etc.

[0027] In this embodiment, the historical power load data refers to the specific data of the power load of the target distribution network recorded in the past period of time, usually in kilowatts (kW) or megawatts (MW).

[0028] In this embodiment, the parameter type refers to the type of historical power transaction data, including: transaction time, transaction volume, power price, transaction type, etc.

[0029] In this embodiment, the power transaction feature sequence refers to features of various parameter types with time series properties extracted from historical power transaction data.

[0030] In this embodiment, the parameter association weight refers to a value used to evaluate the relative importance of one parameter type to all other parameter types during the multimodal data fusion process.

[0031] In this embodiment, the transaction fusion data is a data set generated by performing multimodal data fusion on the power transaction feature sequence of each parameter type and its parameter association weight.

[0032] In this embodiment, time alignment refers to the process of matching and synchronizing different types of data in chronological order during data processing to ensure that their timestamps are consistent.

[0033] In this embodiment, the time series set is a set generated by time-aligning the transaction fusion data with the historical power load data.

[0034] In this embodiment, the transaction-load data curve is a data curve constructed based on a time series set, which shows the functional relationship between transaction fusion data and historical power load data.

[0035] In this embodiment, the load sensitivity factor refers to some parameters in the power system that affect the power load and are associated with time changes, such as holidays, weather, peak power consumption time, economic planning period, etc.

[0036] In this embodiment, the preset sensitivity factor table is a table that pre-stores a plurality of load sensitivity factors and a time interval corresponding to each load sensitivity factor.

[0037] In this embodiment, fitting is to combine the data in the preset sensitivity factor table with the transaction-load data curve through mathematical methods, with the purpose of finding the best matching relationship between the two.

[0038] In this embodiment, the distribution network load forecasting model is a model obtained by fitting the load sensitivity factor to the transaction-load data curve, which describes the relationship between the time-related load sensitivity factor, the target power trading market power trading data and the target distribution network power load data. It is a mathematical model that can predict the future target distribution network power load based on the real-time power trading data and time of the target power trading market.

[0039] In this embodiment, the real-time power transaction data refers to power transaction information generated during the real-time transaction process in the target power transaction market.

[0040] In this embodiment, the current power load forecast result of the target distribution network refers to the power load required on the target distribution network at a certain time point or time period in the future, which is predicted based on the distribution network load forecast model.

[0041] The working principle and beneficial effects of the above technical solution are: by obtaining historical electricity trading data and historical electricity load data, and then performing multimodal data fusion on the electricity trading feature sequence of each parameter type of the historical electricity trading data and its parameter association weights, and then time-aligning the transaction fusion data with the historical electricity load data and constructing a transaction-load data curve based on the time series set generated by the time alignment, and fitting the load sensitivity factor into the transaction-load data curve to construct a distribution network load forecasting model, providing a model-parameterized distribution network load forecasting model, which establishes a correlation between historical electricity trading data and historical electricity load data, and introduces a time-related load sensitivity factor, so that when predicting the current power load of the target distribution network, the correlation between the electricity trading data and the electricity load data and the load sensitivity factor are simultaneously considered, thereby reducing the prediction error and improving the accuracy and efficiency of the distribution network load forecasting results.

[0042] Embodiment 2: In a further embodiment of the present invention, the acquisition module includes: Power transaction data acquisition unit: used to acquire historical power transaction data of the target power transaction market based on a preset power transaction interface according to a preset historical interval range and a preset parameter type of power transaction data; The power load data acquisition unit is used to acquire the historical power load data of the target distribution network based on the preset distribution network interface of the target distribution network according to the preset historical interval range.

[0043] In this embodiment, the parameter types of the preset power transaction data are parameter types of historical power transaction data set in advance, including: transaction time, transaction volume, power price, transaction type, etc.

[0044] In this embodiment, the preset historical interval range refers to a specific time period, which is used to limit the historical time range for obtaining historical power transaction data and historical power load data.

[0045] The working principle and beneficial effects of the above technical solution are: by presetting the historical interval range and parameter type, the historical power trading data of the target power trading market and the historical power load data of the target distribution network are obtained through the preset power trading interface and the preset distribution network interface respectively, which can provide reference data for the subsequent construction of the distribution network load forecasting model.

[0046] Embodiment 3: In a further embodiment of the present invention, the data fusion module includes: Standardization unit: used to standardize the historical power transaction data of each parameter type through statistical methods to obtain standard power transaction data of each parameter type; A feature extraction unit is configured to, for standard power trading data of any parameter type, use a sliding window of a preset length, with the preset length as a sliding step, slide the sliding window from the start time node to the end time node of the standard power trading data of the parameter type within the standard power trading data of the parameter type, obtain the window feature of each time node in the sliding step of each sliding, and synthesize the window features of all time nodes to obtain a power trading feature sequence of the standard power trading data of the parameter type; Parameter association weight calculation unit: used to calculate the parameter association weight of the power transaction feature sequence of each parameter type relative to the power transaction feature sequences of other parameter types based on the power transaction feature sequence of each parameter type and its standard power transaction data; Multimodal data fusion unit: used to input the power transaction feature sequence of each parameter type and its parameter associated weights into the pre-trained data fusion recurrent neural network in sequence according to the preset parameter type order, and the data fusion recurrent neural network outputs the transaction fusion data; wherein, the power transaction feature sequence of the next parameter type specified by the single modality and the preset parameter type order output by the data fusion recurrent neural network last time and its parameter associated weights are the next input of the data fusion recurrent neural network, until the power transaction feature sequences of all parameter types are input into the data fusion recurrent neural network, the cycle ends and the transaction fusion data is obtained.

[0047] In this embodiment, when the data point X of the historical power transaction data of any parameter type is normalized by a statistical method, it can be achieved by formula (3): (3); In formula (3), Z represents the standard data point after the data point X is normalized, and J represents the mean value of all data points of the historical power transaction data of the parameter type. Represents the standard deviation of all data points in the historical electricity trading data of the parameter type.

[0048] In this embodiment, the standard power transaction data is a data set obtained by processing historical power transaction data using statistical methods. Standardization enables standard power transaction data of different parameter types to have unified standards and comparability.

[0049] In this embodiment, the preset length refers to a fixed time interval length of a sliding window set in advance in the feature extraction unit.

[0050] In this embodiment, the sliding window is a mathematical tool for dividing standard power transaction data of any parameter type into small subsets according to a sliding step size.

[0051] In this embodiment, the sliding step size refers to the time interval between each movement of the window when the sliding window is used for data processing.

[0052] In this embodiment, the start time node refers to the time point when the standard power transaction data starts.

[0053] In this embodiment, the end time node refers to the time point when the standard power transaction data ends.

[0054] In this embodiment, the window feature refers to the characteristic value of each time node of the standard power transaction data extracted within one sliding of the sliding window. The window feature types include: rate of change, difference, density, etc.

[0055] In this embodiment, for standard power trading data of any parameter type, when integrating the window features of all time nodes, the window features of all time nodes are spliced in chronological order to obtain the power trading feature sequence of the standard power trading data of the parameter type.

[0056] In this embodiment, the data fusion recurrent neural network is a recurrent neural network (RNN) variant that is pre-trained in the embodiment of the present invention for multimodal time series data fusion. Its core goal is to dynamically integrate power transaction feature sequences of different parameter types, and consider the parameter association weights between them, and finally output a comprehensive, high-dimensional fusion representation (transaction fusion data). For example: Assuming that there are three parameter types of power transaction feature sequences, the data fusion recurrent neural network is divided into three fusion stages, and the power transaction feature sequence ranked first in the preset parameter type sequence and its parameter association weights are input into the data fusion recurrent neural network, and then the single modality of the first fusion stage is output, the single modality of the first fusion stage and the power transaction feature sequence ranked second in the preset parameter type sequence and its parameter association weights are input into the data fusion recurrent neural network, and the second-order single modality is output, the single modality of the second fusion stage and the power transaction feature sequence ranked third in the preset parameter type sequence and its parameter association weights are input into the data fusion recurrent neural network, and the transaction fusion data is output. The example figure is as follows Figure 2 shown.

[0057] In this embodiment, unimodality refers to features involving only a single dimension output by the data fusion recurrent neural network.

[0058] In this embodiment, the preset parameter type order refers to the input order of the predefined power transaction feature sequences of various parameter types and their parameter associated weights in order to ensure smooth fusion during the multimodal data fusion process.

[0059] The working principle and beneficial effects of the above technical solution are: by extracting the power transaction feature sequence of standard power transaction data and obtaining the parameter association weights of the power transaction feature sequences of all parameter types, and then combining multimodal data fusion technology, all power transaction feature sequences and their parameter association weights are input into the data fusion recurrent neural network and then the transaction fusion data is output, which provides a solid data foundation for the subsequent construction of the distribution network load forecasting model.

[0060] Embodiment 4: In a further example of the present invention, the feature extraction unit is used to obtain the power transaction feature sequence H of the standard power transaction data of any parameter type through formula (1):

[0061] In formula (1), represents a cross-window aggregate function, represents the feature extraction function, represents the i1th time node, represents the standard power transaction data within the sliding window starting from time node t and ending at time node t+r, and , r represents the preset length of the sliding window, Indicates the total number of windows, Indicates the start time node of the standard power transaction data of the parameter type, Indicates the end time node of the standard power transaction data of the parameter type, Represents the time node obtained based on the sliding window Window characteristics.

[0062] In this embodiment, the cross-window aggregation function is used to aggregate the window features of multiple sliding windows. By aggregating the window features in each sliding window of standard power trading data of any parameter type, a power trading feature sequence of the standard power trading data of the parameter type can be obtained.

[0063] In this embodiment, the feature extraction function is used to extract window features from each sliding window.

[0064] In this embodiment, formula (1) decomposes the feature extraction of standard power transaction data into two steps: local feature extraction and global aggregation through a sliding window, taking into account both computational efficiency and feature expression capability.

[0065] The working principle and beneficial effects of the above technical solution are: through cross-window aggregation function and feature extraction function, the power transaction feature sequence of standard power transaction data can be accurately extracted, thereby improving the robustness of the analysis of standard power transaction data.

[0066] Embodiment 5: In a further embodiment of the present invention, the parameter association weight calculation unit is used to calculate the parameter association weight of the power transaction feature sequence of the i-th parameter type relative to the power transaction feature sequences of other parameter types by formula (2): : In formula (2), m represents the number of parameter types. represents the covariance function, represents the power transaction feature sequence of the i-th parameter type, represents the power transaction feature sequence of the jth parameter type, Indicates the first parameter type of the i-th Standard electricity transaction data at each time node, Indicates the jth parameter type Standard electricity transaction data at each time node, represents the mean of the standard power transaction data of the i-th parameter type, represents the mean value of the standard power transaction data of the jth parameter type, represents the correlation coefficient between the standard power transaction data of the i-th parameter type and the standard power transaction data of the j-th parameter type, Represents the normalized weight function, Indicates the preset influence coefficient.

[0067] In this embodiment, the covariance function is used to calculate the covariance between two power trading feature sequences of different parameter types. If the changing trends of the two variables are consistent, the covariance is positive; if their changing trends are opposite, the covariance is negative; if their changes have no significant correlation, the covariance is close to zero.

[0068] In this embodiment, the correlation coefficient is an indicator used to measure the strength of the relationship between power transaction feature sequences of different parameter types.

[0069] In this embodiment, the normalized weight function is used to uniformly standardize different weight values so that they can be compared and operated on the same scale. The sum of the parameter association weights of the power transaction feature sequences of all parameter types after processing by the normalized weight function is 1.

[0070] In this embodiment, formula (2) realizes the quantification of parameter association weights and adaptive adjustment of weights through a three-step design of covariance statistics + standardization + dynamic normalization, which can highlight key influencing factors so as to extract more critical features from historical power trading data while ignoring less important features.

[0071] The working principle and beneficial effects of the above technical solution are: through the covariance function, the correlation coefficients of the power trading feature sequences of different parameter types are obtained, and then normalization processing is performed to obtain the parameter association weights of the power trading feature sequences of any parameter type to the power trading feature sequences of all other parameter types. This can accurately evaluate the impact of the power trading feature sequences of each parameter type on the overall situation, improve the accuracy of subsequent multimodal data fusion, and enhance the robustness of the distribution network load forecasting model constructed subsequently.

[0072] Embodiment 6: In a further example of the present invention, the time alignment module includes: Time alignment unit: used to standardize historical power load data to obtain standard power load data, compare the time granularity of transaction fusion data and standard power load data with the preset standard time granularity, perform linear interpolation processing on transaction fusion data and standard power load data with a time granularity greater than the preset standard time granularity, and upwardly aggregate transaction fusion data and standard power load data with a time granularity less than the preset standard time granularity based on statistical characteristics to obtain time-aligned aligned transaction fusion data and aligned standard power load data, and generate a time series set of the aligned transaction fusion data, aligned standard power load data and aligned time nodes; Curve construction unit: used to construct a transaction-load data curve based on a time series set, with the horizontal axis being aligned with transaction fusion data and the vertical axis being aligned with power load data.

[0073] In this embodiment, the preset standard time granularity is a comparison value used as a unified standard to compare the time granularity of transaction fusion data and standard power load data. Time granularity refers to the smallest unit of transaction fusion data and standard power load data in the time dimension, which determines the frequency and density of data recording and describes the time interval between data points.

[0074] In this embodiment, the linear interpolation process is to fill in the data points of the transaction fusion data and the standard power load data with a time granularity smaller than the preset standard time granularity with the assumed values of the linear hypothesis. Assuming that the assumed values need to be inserted in the middle of y1 and y2, the assumed value y0 is: ;in, Indicates the time node corresponding to y1, Indicates the time node corresponding to y2.

[0075] In this embodiment, statistical features refer to representative features extracted from transaction fusion data and standard power load data according to statistical methods, which can describe data distribution and change trends, such as mean, median, standard deviation, maximum value, etc.

[0076] In this embodiment, upward aggregation refers to compressing transaction fusion data and standard power load data with smaller time granularity into data with larger time granularity. For example, if the time granularity of the standard power load data is 1 minute, the total time length is 1 hour, and the preset standard time granularity is 15 minutes, then when the standard power load data is upward aggregated, the 60 data points in the standard power load data are divided into four areas, and the aggregation value of each area is obtained according to statistical characteristics. The aggregation value is used as the value of the data point in the corresponding area to obtain aligned standard power load data with a time granularity of 15 minutes.

[0077] In this embodiment, aligned transaction fusion data refers to a data set obtained after time alignment processing of the transaction fusion data, aligned standard power load data refers to a data set obtained after time alignment processing of the standard power load data, and aligned time node refers to the unified time node of the aligned transaction fusion data and aligned power load data after time alignment processing.

[0078] In this embodiment, the transaction-load data curve uses the aligned transaction fusion data in the time series set as the x-axis and the aligned standard power load data as the y-axis, and draws a one-to-one scatter plot of the aligned transaction fusion data and the aligned standard power load data in the standard coordinate system, and then performs function fitting to generate the transaction-load data curve.

[0079] The working principle and beneficial effects of the above technical solution are: a time series set of standard power load data and transaction fusion data is obtained through time alignment unit processing, and finally a transaction-load data curve is generated based on the time series set, providing accurate time series support for distribution network power load prediction and analysis, and realizing the curvilinearization of the correlation function between standard power load data and transaction fusion data, which provides a prerequisite for subsequently introducing load sensitivity factors into the transaction-load data curve to obtain a distribution network load prediction model.

[0080] Example 7: In a further example of the present invention, the load sensitivity factor fitting module includes: Sensitive label adding unit: used to match the time interval corresponding to any type of load sensitive factor in the preset sensitive factor table with each aligned time node in the time series set. If any aligned time node successfully matches the time interval corresponding to the load sensitive factor of the type, a sensitive label of the load sensitive factor of the type is added to the aligned time node; a label time series set is generated based on all aligned time nodes and their sensitive labels, aligned transaction fusion data, and aligned standard power load data; a core weight determination unit configured to, for any alignment time node with a sensitive label, use the sensitive weight of the load sensitivity factor corresponding to the alignment time node with the sensitive label as the core weight of the alignment time node with the sensitive label; if the alignment time node with the sensitive label has multiple sensitive labels, use the maximum value of the sensitive weights corresponding to all the sensitive labels of the alignment time node with the sensitive label as the core weight of the alignment time node with the sensitive label; wherein the core weight of an alignment time node without a sensitive label is determined to be 0; The function fitting unit is used to use the transaction-load data curve as a constraint condition, fit the kernel weight and constraint condition of each aligned time node based on the kernel regression model, and obtain the distribution network load forecasting model.

[0081] In this embodiment, matching is to determine whether any aligned time node is within a time interval corresponding to any type of load sensitivity factor.

[0082] In this embodiment, a sensitivity label is an identifier added to an alignment time node after a successful match between the alignment time node and the load sensitivity factor. This identifier matches the load sensitivity factor one-to-one. For example, if the load sensitivity factor type is high temperature weather, the time interval is 07-01-08-31, and the alignment time node is 06-01-15:30, the load sensitivity factor of the alignment time node is determined to be high temperature weather, and the sensitivity label added to the alignment time node is high temperature weather.

[0083] In this embodiment, the sensitive weight is a value preset for the load sensitive factor to represent the degree of influence on the power load of the distribution network. In the preset stage, if the load value of the historical power load data is large within the corresponding time interval of any load sensitive factor, a larger weight value is configured for the load sensitive factor.

[0084] In this embodiment, the kernel weight refers to an indicator assigned to each aligned time node, indicating the importance of each aligned time node in the entire kernel regression model.

[0085] In this embodiment, the constraint condition refers to the restriction requirement that the kernel regression model must comply with during the fitting process.

[0086] In this embodiment, the kernel regression model is a regression model based on a nonlinear kernel method. It maps the kernel weight of each aligned time node and the transaction-load data curve to a high-dimensional space, so that the linearly inseparable problem in the original space becomes linearly separable in the high-dimensional space. By fitting the kernel weight of each aligned time node with the transaction-load data curve, the output is a model in which the independent variables are the aligned time nodes and aligned transaction fusion data, and the dependent variable is the aligned standard power load data.

[0087] The working principle and beneficial effects of the above technical solution are: by matching the load sensitive factors in the preset sensitive factor table and the aligned time nodes of the time series set, a label time series set with sensitive labels is generated, and the transaction-load data curve is used as a constraint condition. The constraint condition and the kernel weight of each aligned time node are then fitted to the kernel regression model to obtain a distribution network load forecasting model. A function model is constructed with the input of aligned time nodes and aligned transaction fusion data and the output of aligned standard power load data. The distribution network load forecasting model establishes a correlation between power transaction data and power load data, and introduces load sensitive factors, so that when power load forecasting is performed based on the distribution network load forecasting model in the future, the accuracy and robustness of the power load forecasting results can be improved.

[0088] Example 8: In a further example of the present invention, the real-time prediction module is used to: obtain the current real-time electricity trading data of the target electricity trading market based on a preset electricity trading interface, and input the real-time electricity trading data and the current timestamp into the distribution network load prediction model, and determine the current electricity load prediction result of the target distribution network according to the output of the distribution network load prediction model.

[0089] In this embodiment, the timestamp is a mark indicating the current time point, and is usually stored in a standard format, such as year-month-day hour:minute:second.

[0090] The working principle and beneficial effects of the above technical solution are: by inputting the current real-time electricity transaction data of the target electricity trading market and its current timestamp into the distribution network load forecasting model, the distribution network load forecasting model outputs the current electricity load forecast result of the target distribution network, thereby realizing real-time and accurate prediction of the electricity load.

[0091] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the technical principles of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A distribution network load forecasting system based on real-time power trading, characterized by: include: Acquisition module: used to obtain historical power transaction data of the target power trading market and historical power load data of the target distribution network based on the preset power transaction interface and the preset distribution network interface of the target distribution network; Data fusion module: used to extract the power transaction feature sequence of each parameter type from historical power transaction data, obtain the parameter association weight of the power transaction feature sequence of each parameter type, and perform multimodal data fusion on the power transaction feature sequence of each parameter type and its parameter association weight to generate transaction fusion data; Time alignment module: used to time-align transaction fusion data with historical power load data to obtain a time series set, and construct a transaction-load data curve based on the time series set; Load sensitivity factor fitting module: used to fit the load sensitivity factors in the preset sensitivity factor table to the transaction-load data curve to obtain the distribution network load forecasting model; Real-time prediction module: used to obtain real-time power transaction data of the target power trading market based on the preset power transaction interface, and determine the current power load forecast result of the target distribution network based on the real-time power transaction data and the distribution network load forecast model.

2. The distribution network load forecasting system based on real-time power trading according to claim 1 is characterized in that: The acquisition module includes: Power transaction data acquisition unit: used to acquire historical power transaction data of the target power transaction market based on a preset power transaction interface according to a preset historical interval range and a preset parameter type of power transaction data; The power load data acquisition unit is used to acquire the historical power load data of the target distribution network based on the preset distribution network interface of the target distribution network according to the preset historical interval range.

3. The distribution network load forecasting system based on real-time power trading according to claim 1 is characterized in that: The data fusion module includes: Standardization unit: used to standardize the historical power transaction data of each parameter type through statistical methods to obtain standard power transaction data of each parameter type; A feature extraction unit is configured to, for standard power trading data of any parameter type, use a sliding window of a preset length, with the preset length as a sliding step, slide the sliding window from the start time node to the end time node of the standard power trading data of the parameter type within the standard power trading data of the parameter type, obtain the window feature of each time node in the sliding step of each sliding, and synthesize the window features of all time nodes to obtain a power trading feature sequence of the standard power trading data of the parameter type; Parameter association weight calculation unit: used to calculate the parameter association weight of the power transaction feature sequence of each parameter type relative to the power transaction feature sequences of other parameter types based on the power transaction feature sequence of each parameter type and its standard power transaction data; Multimodal data fusion unit: used to input the power transaction feature sequence of each parameter type and its parameter associated weights into the pre-trained data fusion recurrent neural network in sequence according to the preset parameter type order, and the data fusion recurrent neural network outputs the transaction fusion data; wherein, the power transaction feature sequence of the next parameter type specified by the single modality and the preset parameter type order output by the data fusion recurrent neural network last time and its parameter associated weights are the next input of the data fusion recurrent neural network, until the power transaction feature sequences of all parameter types are input into the data fusion recurrent neural network, the cycle ends and the transaction fusion data is obtained.

4. The distribution network load forecasting system based on real-time power trading according to claim 3 is characterized in that: The feature extraction unit is used to obtain the power transaction feature sequence H of standard power transaction data of any parameter type through formula (1): In formula (1), represents a cross-window aggregate function, represents the feature extraction function, represents the i1th time node, represents the standard power transaction data within the sliding window starting from time node t and ending at time node t+r, and , r represents the preset length of the sliding window, Indicates the total number of windows, Indicates the start time node of the standard power transaction data of the parameter type, Indicates the end time node of the standard power transaction data of the parameter type, Represents the time node obtained based on the sliding window Window characteristics.

5. The distribution network load forecasting system based on real-time power trading according to claim 3 is characterized in that: The parameter association weight calculation unit is used to calculate the parameter association weight of the power transaction feature sequence of the i-th parameter type relative to the power transaction feature sequences of other parameter types through formula (2): : In formula (2), m represents the number of parameter types. represents the covariance function, represents the power transaction feature sequence of the i-th parameter type, represents the power transaction feature sequence of the jth parameter type, Indicates the first parameter type of the i-th Standard electricity transaction data at each time node, Indicates the jth parameter type Standard electricity transaction data at each time node, represents the mean of the standard power transaction data of the i-th parameter type, represents the mean value of the standard power transaction data of the jth parameter type, represents the correlation coefficient between the standard power transaction data of the i-th parameter type and the standard power transaction data of the j-th parameter type, Represents the normalized weight function, Indicates the preset influence coefficient.

6. The distribution network load forecasting system based on real-time power trading according to claim 1 is characterized in that: The time alignment module includes: Time alignment unit: used to standardize historical power load data to obtain standard power load data, compare the time granularity of transaction fusion data and standard power load data with the preset standard time granularity, perform linear interpolation processing on transaction fusion data and standard power load data with a time granularity greater than the preset standard time granularity, and upwardly aggregate transaction fusion data and standard power load data with a time granularity less than the preset standard time granularity based on statistical characteristics to obtain time-aligned aligned transaction fusion data and aligned standard power load data, and generate a time series set of the aligned transaction fusion data, aligned standard power load data and aligned time nodes; Curve construction unit: used to construct a transaction-load data curve based on a time series set, with the horizontal axis being aligned with transaction fusion data and the vertical axis being aligned with power load data.

7. The distribution network load forecasting system based on real-time power trading according to claim 6 is characterized in that: The load sensitivity factor fitting module includes: Sensitive label adding unit: used to match the time interval corresponding to any type of load sensitive factor in the preset sensitive factor table with each aligned time node in the time series set. If any aligned time node successfully matches the time interval corresponding to the load sensitive factor of the type, a sensitive label of the load sensitive factor of the type is added to the aligned time node; a label time series set is generated based on all aligned time nodes and their sensitive labels, aligned transaction fusion data, and aligned standard power load data; a core weight determination unit configured to, for any alignment time node with a sensitive label, use the sensitive weight of the load sensitivity factor corresponding to the alignment time node with the sensitive label as the core weight of the alignment time node with the sensitive label; if the alignment time node with the sensitive label has multiple sensitive labels, use the maximum value of the sensitive weights corresponding to all the sensitive labels of the alignment time node with the sensitive label as the core weight of the alignment time node with the sensitive label; wherein the core weight of an alignment time node without a sensitive label is determined to be 0; The function fitting unit is used to use the transaction-load data curve as a constraint condition, fit the kernel weight and constraint condition of each aligned time node based on the kernel regression model, and obtain the distribution network load forecasting model.

8. The distribution network load forecasting system based on real-time power trading according to claim 1 is characterized in that: The real-time prediction module is used to: obtain the current real-time electricity trading data of the target electricity trading market based on a preset electricity trading interface, input the real-time electricity trading data and the current timestamp into the distribution network load prediction model, and determine the current electricity load prediction result of the target distribution network according to the output of the distribution network load prediction model.

Images