Vector model construction method for heavy-load line

Abstract

The invention relates to the technical field of circuit model construction, and particularly discloses a vector model construction method for a heavy-load circuit, which comprises the following steps of: accurately matching model complexity, ensuring the practicability and accuracy of a model, setting a plurality of detection position points on the heavy-load circuit, and acquiring and analyzing electrical data of the plurality of detection position points, so as to construct a vector model for the heavy-load circuit. According to the method, the complexity of the vector model can be scientifically determined, so that an initial model fitting the reality is constructed, and the method has another advantage of dynamic adjustment and optimization capability, after a training data set of the initial model is selected, whether data input batch values need to be adjusted or not can be judged in time by monitoring parameters of a training process, and the accuracy of the data input batch values is improved. The simulation data of each monitoring position point is collected and compared with the actual input data for analysis, and the deviation in the model is found and corrected in time, so that the model is more perfect, and the accuracy and practicability of the model are ensured.

Description

Technical Field

[0001] The present invention relates to the technical field of circuit model construction, and in particular to a vector model construction method for a heavy-load circuit. Background Art

[0002] With the development of the economy, the scale of industrial production continues to expand, and the demand for electricity in cities continues to rise. Heavy-load lines play a key role in power transmission in modern power systems, and accurate control of their operating status is crucial.

[0003] For example, the invention patent with publication number CN116451630A discloses a method for constructing a broadband equivalent circuit model of a grounding grid based on the vector matching method. First, a broadband equivalent circuit model of the main equipment of a GIS station suitable for engineering calculations is established, including a circuit model of the GIS busbar port voltage and line current under high frequency, an actual model of the SF6 / Air high-voltage bushing, and a broadband equivalent model of the grounding grid composed of actual models of the grounding grid and grounding pillars. Then, based on the established dynamic arcing model of the disconnector and the broadband equivalent circuit model of the main GIS equipment, the fast transient overvoltage generated by the switch operation in the GIS test circuit is simulated and calculated.

[0004] For example, the invention patent with publication number CN114169115A discloses a method for rapid and automated modeling of transmission lines. It realizes rapid and automated modeling of transmission lines based on the pole tower model library and line ledger information. According to the pole tower model recorded in the line ledger information and the height information contained therein, a typical pole tower three-dimensional model library file is established, and the hanging information is stored in the library. The pole tower orientation and proportion are automatically calculated according to the pole tower number and spatial position coordinates, the ground wire vector information is generated, and the three-dimensional model of the transmission line is constructed.

[0005] However, in the process of implementing the embodiments of the present application, the present application found that the above-mentioned technology has at least the following technical problems: the existing technology often follows a relatively direct and linear process in the model construction and simulation application process, namely the model construction stage, the simulation execution stage and the result application stage. In the actual application scenario of the large-load line model, since the model needs to be continuously revised according to the actual situation, this linear process of the existing technology limits the flexibility and adaptability of the model. Summary of the Invention

[0006] In view of the deficiencies in the prior art, the present invention provides a method for constructing a vector model for a heavy-load line, which can effectively solve the problems involved in the above-mentioned background technology.

[0007] To achieve the above objectives, the present invention is implemented through the following technical solutions: a method for constructing a vector model for a heavy-load line, comprising: step 1, setting each detection location point on the heavy-load line, collecting and analyzing electrical data at each detection location point, thereby matching the model complexity of the heavy-load line vector model; step 2, constructing a heavy-load line vector model based on the model complexity of the heavy-load line vector model, marking it as an initial model, selecting each training data set of the initial model, training the initial model, collecting training process parameters of the initial model, and determining whether to adjust the data input batch value of the initial model; step 3, collecting simulation data from each monitoring location point, comparing and analyzing it with the input data of each monitoring location point, and determining whether to modify the initial model.

[0008] As a further method, the model complexity of the heavy-load line vector model is matched, and the specific matching process is: analyzing the electrical data of each detection location point to obtain the data fluctuation index of the heavy-load line; comparing the data fluctuation index of the heavy-load line with the data fluctuation index reference interval; if the data fluctuation index of the heavy-load line is greater than the maximum value of the data fluctuation index reference interval, the model complexity of the heavy-load line vector model matched from the database is the first model complexity; if the data fluctuation index of the heavy-load line belongs to the data fluctuation index reference interval, the model complexity of the heavy-load line vector model matched from the database is the second model complexity; if the data fluctuation index of the heavy-load line is less than the minimum value of the data fluctuation index reference interval, the model complexity of the heavy-load line vector model matched from the database is the third model complexity.

[0009] As a further method, the data input batch value of the initial model is adjusted, and the specific adjustment process is: obtaining the model complexity of the initial model, and matching the data input minimum batch value of the initial model from the database; obtaining the performance set of the operating equipment to which the initial model belongs in the training sub-cycle, and matching the data input maximum batch value of the initial model from the database; performing difference processing on the data reception anomaly index of the initial model in the training sub-cycle and the data reception anomaly interval, and performing ratio processing on the difference processing result and the data reception anomaly interval to obtain the data reception anomaly index of the initial model in the training sub-cycle; obtaining the data input batch value of the initial model, if the data input batch value of the initial model is less than the data input minimum batch value of the initial model, then The data input batch value is increased and adjusted according to the data reception anomaly index of the initial model in the training sub-cycle; if the data input batch value of the initial model is greater than the data input maximum batch value of the initial model, the data input batch value is decreased and adjusted according to the data reception anomaly index of the initial model in the training sub-cycle; the data input minimum batch value is marked as the minimum value of the data input reference batch value interval, and the data input maximum batch value is marked as the maximum value of the data input reference batch value interval. If the data input batch value of the initial model belongs to the data input reference batch value interval, the data input batch value adjustment value is matched from the database according to the data reception anomaly index of the initial model in the training sub-cycle, and the data input batch value of the initial model is adjusted.

[0010] As a further method, the initial model is corrected, and the specific correction process is: obtaining the simulation data execution rate of the initial model and comparing it with the simulation data execution threshold; if the simulation data execution rate of the initial model is less than or equal to the simulation data execution threshold, then obtaining the data distribution density of each detection module belonging to the initial model, locating each detection module whose data distribution density is less than the data distribution density threshold, marking it as each abnormal module, and increasing the number of nodes of each abnormal module according to the electrical simulation deviation factor of the initial model; if the simulation data execution rate of the initial model is greater than the simulation data execution threshold, then obtaining the harmonic frequency deviation value of the initial model and comparing it with the harmonic frequency deviation threshold; if the harmonic frequency deviation value of the initial model is greater than the harmonic frequency deviation threshold, enabling the frequency change parameter module; if the harmonic frequency deviation value of the initial model is less than or equal to the harmonic frequency deviation threshold, then increasing the number of detection position points according to the electrical simulation deviation factor of the initial model, thereby completing the correction of the initial model.

[0011] Compared with the prior art, the embodiments of the present invention have at least the following advantages or beneficial effects:

[0012] (1) The present invention provides a method for constructing a vector model for a heavy-load line. By accurately matching the complexity of the model, the practicality and accuracy of the model are ensured. By setting multiple detection position points on the heavy-load line and collecting and analyzing the electrical data of the multiple detection position points, the complexity of the vector model can be scientifically determined, thereby constructing an initial model that fits the actual situation. Another major advantage of this method is its dynamic adjustment and optimization capabilities. After selecting the training data set of the initial model, by monitoring the training process parameters, it can be timely determined whether the data input batch value needs to be adjusted to ensure the best training effect. In addition, by collecting the simulation data of each monitoring position point and comparing and analyzing it with the actual input data, the deviation in the model can be timely discovered and corrected, making the model more perfect. Therefore, the present invention not only provides an effective method for constructing a heavy-load line vector model, but also ensures the accuracy and practicality of the model through a dynamic adjustment and optimization mechanism, providing strong support for the operation and maintenance of heavy-load lines.

[0013] (2) The present invention achieves a quantitative assessment of the degree of abnormality in model data reception by carefully collecting the training process parameters of the initial model. It can not only accurately grasp the data processing status of the model during the training process, but also provide a scientific basis for subsequent data input strategies. Based on this quantitative result, the data input batch value of the initial model can be timely determined and adjusted in a targeted manner to ensure the stability and efficiency of model training, thereby improving the accuracy and reliability of the model.

[0014] (3) The present invention achieves accurate correction of the initial model by comparing simulation data with input data. By quantifying the simulation data execution rate, data distribution density and harmonic frequency deviation value, it can accurately locate model anomalies and take corresponding measures, such as adding abnormal module nodes, enabling frequency-varying parameter modules or adding detection location points, which effectively improves the accuracy and adaptability of the model and ensures the reliable application of the model on high-load lines. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] The present invention is further described with reference to the accompanying drawings. However, the embodiments in the accompanying drawings do not constitute any limitation to the present invention. A person skilled in the art can obtain other drawings based on the following drawings without creative effort.

[0016] Figure 1 Schematic diagram of the method of the present invention. DETAILED DESCRIPTION

[0017] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0018] Reference Figure 1 As shown, the present invention provides a vector model construction method for a heavy-load line, comprising: step 1, setting each detection position point on the heavy-load line, collecting and analyzing the electrical data of each detection position point, and thus matching the model complexity of the heavy-load line vector model.

[0019] Specifically, the model complexity of the heavy load line vector model is matched, and the specific matching process is: analyze the electrical data of each detection position point to obtain the data fluctuation index of the heavy load line; compare the data fluctuation index of the heavy load line with the data fluctuation index reference interval. If the data fluctuation index of the heavy load line is greater than the maximum value of the data fluctuation index reference interval, the model complexity of the heavy load line vector model matched from the database is the first model complexity; if the data fluctuation index of the heavy load line belongs to the data fluctuation index reference interval, the model complexity of the heavy load line vector model matched from the database is the second model complexity; if the data fluctuation index of the heavy load line is less than the minimum value of the data fluctuation index reference interval, the model complexity of the heavy load line vector model matched from the database is the third model complexity; the above-mentioned data fluctuation index reference interval is a preset interval in the database, which is a numerical range for measuring the degree of data fluctuation of the heavy load line; the above-mentioned first model complexity, the specific matching process is: when the heavy load When the data fluctuation index of the heavy-load line is greater than the maximum value of the data fluctuation index reference interval, it is identified that the data fluctuation is extremely severe and a higher-precision model is required to capture the line characteristics. Therefore, the first model complexity with the highest complexity is automatically selected from the database as the model complexity of the heavy-load line vector model to ensure that the model can accurately reflect the dynamic changes of the heavy-load line; the second model complexity and the third model complexity matching process are consistent with the first model complexity matching process. When the data fluctuation index of the heavy-load line belongs to the data fluctuation index reference interval, it is judged that the line data fluctuation degree is moderate, and the second model complexity with moderate complexity is selected from the database as the model complexity of the heavy-load line vector model. When the data fluctuation index of the heavy-load line is less than the minimum value of the data fluctuation index reference interval, it is identified that the line data fluctuation is small and the change is relatively stable. Therefore, the third model complexity with the smallest complexity is selected from the database as the model complexity of the heavy-load line vector model to meet basic description requirements while reducing computational costs.

[0020] It needs to be explained that the vector model corresponding to the first model complexity has the strongest descriptive ability, the most parameters and the most complex structure; the vector model corresponding to the second model complexity has moderate complexity and can better balance accuracy and computational efficiency; the vector model corresponding to the third model complexity has a simpler structure and less computational complexity.

[0021] Specifically, the data fluctuation index of the heavy-load line has a specific analysis process as follows: the electrical data of each detection location point includes the voltage fluctuation rate of each detection location point within the data detection period, the average drift rate of the harmonic impedance phase angle of each detection location point within the data detection period, and the total current harmonic content of each detection location point within the data detection period; the above-mentioned data detection period refers to the time period for collecting the electrical data of each detection location point, which is formulated by electrical engineers; the above-mentioned voltage fluctuation rate refers to the degree of fluctuation of the voltage value of each detection location point within the data detection period, and the real-time voltage of each detection location point within the data detection period is obtained through the voltage monitoring system of the heavy-load line, and the maximum voltage and minimum voltage are obtained from the real-time voltage. As well as the average voltage, the maximum voltage and the minimum voltage are differenced, and the processing result is ratioed with the average voltage to finally obtain the voltage fluctuation rate of each detection position point within the data detection period; the above-mentioned average drift rate of the harmonic impedance phase angle refers to the average rate at which the harmonic impedance phase angle changes with time within the data detection period. The harmonic impedance phase angle reflects the phase relationship between the voltage and current in the circuit. Its drift rate can reveal the change of the harmonic components in the circuit. It is an important indicator for evaluating the stability of large-load lines and can be obtained from the voltage monitoring system to which the large-load lines belong; the above-mentioned total current harmonic content indicates the total content of harmonic components contained in the current within the data detection period, which can be obtained from the voltage monitoring system to which the large-load lines belong.

[0022] The above-mentioned voltage monitoring system includes sensors, data acquisition modules and analysis software, which can realize functions such as remote monitoring, data analysis and alarm. Sensors are used to collect voltage signals in the circuit and convert them into electrical signals for transmission to the data acquisition module. Data acquisition module is responsible for receiving electrical signals from sensors and converting them into digital signals for storage and processing. Analysis software analyzes the collected voltage data to obtain parameters such as real-time voltage, average drift rate of harmonic impedance phase angle, and total current harmonic content, providing decision support for the operation and management of the power system.

[0023] The data fluctuation index is finally obtained by summarizing and averaging the influence of the ratio of the voltage fluctuation rate of each detection location point to the defined voltage fluctuation rate on the data fluctuation index, the influence of the ratio of the average drift rate of the harmonic impedance phase angle of each detection location point to the defined harmonic impedance phase angle average drift rate on the data fluctuation index, and the influence of the ratio of the total current harmonic content of each detection location point to the defined current harmonic total content on the data fluctuation index. The data fluctuation index of the heavy load line is used to quantify the degree of data fluctuation of the heavy load line, and the specific expression is:

[0024]

[0025] Where HLC is the data fluctuation index of the heavy load line, RV p is the voltage fluctuation rate of the pth detection position within the data detection period, J_RV is the defined voltage fluctuation rate preset in the database, ADH p is the average drift rate of the harmonic impedance phase angle at the pth detection position within the data detection period, J_ADH is the average drift rate of the defined harmonic impedance phase angle preset in the database, THC p is the total current harmonic content of the pth detection position point within the data detection period, J_THC is the defined total current harmonic content preset in the database, sa1 is the influence weight value corresponding to the voltage fluctuation rate preset in the database, sa2 is the influence weight value corresponding to the average drift rate of the harmonic impedance phase angle preset in the database, sa3 is the influence weight value corresponding to the total current harmonic content preset in the database, p is the number of each detection position point, p = {1, 2, 3, ..., f}, f is the total number of detection position points.

[0026] The above definition of voltage fluctuation rate indicates the maximum allowable value of voltage fluctuation rate; the above definition of average drift rate of harmonic impedance phase angle indicates the maximum allowable value of average drift rate of harmonic impedance phase angle; the above definition of total current harmonic content indicates the maximum allowable value of total current harmonic content.

[0027] The influence weight value corresponding to the above-mentioned voltage fluctuation rate represents the numerical value of the influence of the unit value of the voltage fluctuation rate on the data fluctuation index; the influence weight value corresponding to the above-mentioned harmonic impedance phase angle average drift rate represents the numerical value of the influence of the unit value of the harmonic impedance phase angle average drift rate on the data fluctuation index; the influence weight value corresponding to the above-mentioned total current harmonic content represents the numerical value of the influence of the unit value of the total current harmonic content on the data fluctuation index; the database stores the correspondence between the voltage fluctuation rate, the harmonic impedance phase angle average drift rate and the total current harmonic content and their corresponding influence weight values. For example, when the voltage fluctuation rate, the harmonic impedance phase angle average drift rate and the total current harmonic content are input into the database, the database can match the influence weight value corresponding to the voltage fluctuation rate, the influence weight value corresponding to the harmonic impedance phase angle average drift rate and the influence weight value corresponding to the total current harmonic content, and the value range is between 0 and 1.

[0028] It should be explained that the voltage fluctuation rate reflects the stability of the voltage of a large-load line, and its size is directly affected by factors such as load changes, power supply fluctuations and line impedance. When the voltage fluctuation rate increases, it means that the line voltage stability decreases, which may lead to an increase in the drift rate of the harmonic impedance phase angle, because the unstable voltage will aggravate the harmonic components in the line, thereby affecting the phase angle of the harmonic impedance. At the same time, voltage fluctuations may also cause the harmonic content in the current to increase, because unstable voltage will excite the nonlinear elements in the line and generate more harmonic currents. The average drift rate of the harmonic impedance phase angle reflects the speed of change of the line harmonic characteristics. It is closely related to the harmonic source, line impedance and harmonic response of the load in the line. When the harmonic impedance phase angle drift rate increases, it means that the harmonic components in the line are changing rapidly, which may be due to drastic changes in the load or harmonic source. This change will further affect the voltage fluctuation rate, because the change of harmonic components will be superimposed on the fundamental voltage, resulting in increased voltage fluctuation. The total current harmonic content is the sum of the harmonic currents in the line, which directly reflects the degree of harmonic pollution in the line. When the total current harmonic content increases, it means that the harmonic current in the line increases, which will lead to increased voltage waveform distortion and thus increase the voltage fluctuation rate. At the same time, the increase in harmonic current will also change the harmonic impedance characteristics in the line, resulting in changes in the drift rate of the harmonic impedance phase angle. In summary, there is a close connection and mutual influence between the voltage fluctuation rate, the average drift rate of the harmonic impedance phase angle and the total current harmonic content. They work together in the calculation of the data fluctuation index, and their different contributions to the data fluctuation degree are reflected through their respective weight values, thereby comprehensively reflecting the data fluctuation degree of the heavy-load line.

[0029] The above database is used to store parameters involved in a method for constructing a vector model for a heavy-load line.

[0030] Furthermore, the detection location points are set on the heavy-load line. The specific setting process is as follows: obtaining the total length of the heavy-load line and matching it with the detection location point spacing corresponding to each total line length interval in the database to obtain the detection location point spacing of the heavy-load line; marking each detection location point on the heavy-load line in sequence according to the detection location point spacing of the heavy-load line; the above-mentioned total line length of the heavy-load line refers to the total distance from the starting point to the end point of the heavy-load line, which reflects the scale and scope of the line and is extracted from the design plan of the heavy-load line; the above-mentioned detection location point spacing of the heavy-load line represents the distance between each detection location point on the heavy-load line, and the specific matching process is as follows: querying the total line length interval in the database to which the total line length of the heavy-load line belongs, and the detection location point spacing corresponding to the total line length interval in the database is the detection location point spacing of the heavy-load line; the above-mentioned marking of each detection location point specifically transmits the determined heavy-load line detection location point spacing information to the voltage monitoring system. This step is to allow the voltage monitoring system to identify the specific spacing of the detection location points that need to be set. In the voltage monitoring system, each detection location point on the heavy-load line is marked one by one based on this spacing information. These marking points will be the key nodes for subsequent voltage monitoring and data collection. Finally, the system obtains the electrical data of the corresponding locations through these marking points, providing a basis for subsequent data analysis and processing.

[0031] Step 2: Based on the model complexity of the heavy-load line vector model, a heavy-load line vector model is constructed and marked as an initial model. The training data sets of the initial model are selected and the initial model is trained. The training process parameters of the initial model are collected to determine whether the data input batch value of the initial model should be adjusted.

[0032] In a specific embodiment, the present invention realizes a quantitative assessment of the degree of abnormality in model data reception by carefully collecting the training process parameters of the initial model. It can not only accurately grasp the data processing status of the model during the training process, but also provide a scientific basis for subsequent data input strategies. Based on this quantitative result, the data input batch value of the initial model can be timely determined and adjusted in a targeted manner to ensure the stability and efficiency of model training, thereby improving the accuracy and reliability of the model.

[0033] In an example embodiment, assuming that the model complexity of the heavy-load line vector model is the first model complexity, the above-mentioned construction of the heavy-load line vector model is specifically carried out as follows: when constructing the initial model, all kinds of information related to the heavy-load line are comprehensively sorted out, and the resistance value of each section of the line is accurately measured from the basic line electrical parameters. For example, a four-wire resistance meter is used to measure multiple times at different temperatures and take the average value to obtain accurate resistance data; the inductance value is measured by a professional inductance meter based on factors such as the material, number of turns, and geometric shape of the line; the capacitance value is measured using equipment such as a capacitance bridge, combined with parameters such as the insulation medium characteristics of the line. For load characteristics, with the help of high-precision power analyzers, the active power and reactive power in different time periods are monitored in real time, and their change curves over time are recorded. The periodic characteristics of load changes and the law of peak occurrence are analyzed. At the same time, geographic information system technology is used to accurately draw the line topology, clarify the direction of the line, branching and the connection relationship between each node. Based on the above detailed data, advanced mathematical modeling theory is used to abstract the line into a vector form. For example, complex number representation is used to describe the electrical quantities such as voltage and current in the line, and a complex mathematical model containing the parameters of each part of the line is constructed, thereby completing the construction of the initial model.

[0034] The aforementioned selected training data sets for the initial model specifically refer to the training data sets corresponding to the model complexity stored in the database. Based on the model complexity of the heavy-load line vector model, the corresponding training data sets are obtained from the database and marked as the training data sets for the initial model. The training data sets are primarily derived from historical data of the heavy-load line operation process. In an example embodiment, the training data sets include, but are not limited to: current data: covering real-time current values at different locations on the line, including the current magnitude and direction during steady-state operation, as well as transient process data of the current under dynamic conditions such as load mutations and faults, which can accurately reflect the changing characteristics of the current under different operating conditions; voltage data: It contains the voltage amplitude and phase information of each node of the line, including not only the rated voltage data during normal operation, but also the voltage change data of the line under abnormal conditions such as overload, undervoltage, and voltage fluctuation. By analyzing the voltage data, it can help the model learn the operating characteristics of the line under different voltage conditions; Power data: Active power and reactive power data record the power transmission of the line in detail. Active power reflects the actual power consumed by the line to do work on the load, and reactive power reflects the scale of energy exchange between energy storage elements such as inductance and capacitance in the line and the power supply. These data can provide the model with key information about the efficiency and stability of line energy transmission. Load characteristic data: Load size: different The power consumption values of different load types (such as industrial load, residential load, commercial load, etc.) at different time points. Industrial load may have periodic high-power consumption periods, while residential load has obvious power changes during peak power consumption in the morning and evening. These load size data can help the model understand the impact of different load characteristics on the line. Load change trend: load growth or decay trend data over time, including seasonal changes, differences between weekdays and weekends, etc. For example, in summer, the overall load will increase due to increased use of air conditioners; industrial load is higher during the day on weekdays, and the proportion of residential load increases at night. Understanding the load change trend will help the model predict the load conditions of the line at different time periods. Environmental factor data: temperature data Data: Ambient temperature significantly affects line resistance, which in turn affects the line's electrical performance. The training dataset contains real-time temperature data at different locations along the line, as well as temperature curves over time, allowing the model to learn the relationship between temperature and line parameters. Humidity data: Humidity affects line insulation performance. A humid environment can cause problems such as line leakage. The dataset records the real-time values of ambient humidity and humidity changes, helping the model consider the potential impact of humidity on line operation. Wind speed data: Strong winds can cause lines to sway, affecting line spacing and even causing line failures. Wind speed data includes real-time wind speed and direction information, which the model uses to analyze the mechanical effects of wind on line operation.Line topology and geographic information data: Topology data: Detailed description of line connections, including line branches, node information, and the electrical connections between each line segment. This data enables the model to understand the flow path and distribution patterns of current throughout the line network. Geographic coordinate data: The line's location coordinates in geographic space, accurate to longitude and latitude. This helps analyze the impact of the geographical environment on line operation by combining geographic information such as topography. For example, line operating characteristics may vary in special geographical areas such as mountainous areas and near water. Surrounding environment data: Information about the line's surrounding environment, obtained through line monitoring cameras and geographic information technology, such as whether it is near buildings and trees, and whether it is in areas of strong electromagnetic interference. This data allows the model to consider potential interference from surrounding environmental factors on line operation.

[0035] Specifically, the determination of whether to adjust the data input batch value of the initial model is carried out as follows: by analyzing the training process parameters of the initial model, a data reception anomaly index of the initial model in a training sub-cycle is obtained; the training sub-cycle refers to a training time period divided according to a time period preset by a data analyst during the entire training process of the initial model. In each training sub-cycle, a portion of the training data set is sequentially input according to the batch value preset by the initial model to train the initial model; by analyzing the training process parameters of the initial model in each training sub-cycle, a data reception anomaly index of the initial model in the training sub-cycle is calculated; based on the data reception anomaly index, it is determined whether the data input batch value of the initial model in the next adjacent training sub-cycle needs to be adjusted; if an adjustment is determined, the data input batch value is modified accordingly, and the next adjacent training sub-cycle is entered; in the next adjacent training sub-cycle, the training process parameters of the initial model are analyzed again, a new data reception anomaly index is calculated, and the above determination process is repeated. This process will continue until all training data sets are input and training is completed. Therefore, there are several training sub-cycles in the training process of the initial model, and the specific number is determined by the actual training process.

[0036] The data reception anomaly interval is extracted from the database, and the data reception anomaly index of the initial model in the training sub-cycle is compared with the data reception anomaly interval. If the data reception anomaly index of the initial model in the training sub-cycle does not fall within the data reception anomaly interval, it is determined that the data input batch value of the initial model does not need to be adjusted; if the data reception anomaly index of the initial model in the training sub-cycle does not fall within the data reception anomaly interval, it is determined that the data input batch value of the initial model needs to be adjusted; the above-mentioned data reception anomaly interval represents a reasonable range of the data reception anomaly index, which is used to determine whether to adjust the data input batch value of the initial model. If the data reception anomaly index falls within the data reception anomaly interval, it indicates that the data reception anomaly index is within a reasonable range and no adjustment is required. Among them, the data reception anomaly index falls within the data reception anomaly interval, and also includes the maximum value of the data reception anomaly index falling within the data reception anomaly interval and the minimum value of the data reception anomaly index falling within the data reception anomaly interval.

[0037] Specifically, the data reception anomaly index of the initial model in the training sub-cycle is analyzed as follows: the training process parameters of the initial model include the data reception buffer occupancy rate of the initial model in the training sub-cycle, the average training dataset reception time of the initial model in the training sub-cycle, and the average training dataset loading rate of the initial model in the training sub-cycle; the data reception buffer occupancy rate refers to the average proportion of the data reception buffer of the initial model occupied in the training sub-cycle. The buffer is used to temporarily store data awaiting processing to ensure that data can be input into the model in an orderly manner according to a preset batch value. It is obtained by monitoring the memory usage of the initial model during training, especially the memory usage of the data reception buffer. Most deep learning frameworks and machine learning libraries provide corresponding tools or APIs (such as memory allocation monitoring functions) to view the memory usage of the model; the average training dataset reception time refers to the average time required for the initial model to receive a complete training dataset in the training sub-cycle, which can be monitored using time monitoring tools (such as time monitoring functions) provided by the machine learning library; and the average training dataset loading rate refers to the amount of data that the initial model can load per unit time in the training sub-cycle, which can be monitored using data loading rate monitoring tools (such as data loading rate monitoring functions) provided by the machine learning library.

[0038] Quantify and summarize the impact of the data fluctuation index on the data reception anomaly index, the impact of the deviation between the data reception buffer occupancy rate and the reference data reception buffer occupancy rate on the data reception anomaly index, the impact of the deviation between the average reception time of the training dataset and the average reception time of the reference training dataset on the data reception anomaly index, and the impact of the deviation between the average loading rate of the training dataset and the average loading rate of the reference training dataset on the data reception anomaly index, and finally obtain the data reception anomaly index; the data reception anomaly index of the initial model in the training sub-cycle is used to quantify the degree of data reception anomaly of the initial model in the training sub-cycle, and the specific expression is:

[0039]

[0040] Where DAI is the data reception anomaly index of the initial model in the training sub-cycle, HLC is the data fluctuation index of the heavy-load line, RFB is the data reception buffer occupancy of the initial model in the training sub-cycle, DTD is the average reception time of the training dataset of the initial model in the training sub-cycle, LRT is the average loading rate of the training dataset of the initial model in the training sub-cycle, ΔRFB is the reference data reception buffer occupancy preset in the database, ΔDTD is the average reception time of the reference training dataset preset in the database, ΔLRT is the average loading rate of the reference training dataset preset in the database, Q is the influence weight value corresponding to the data fluctuation index preset in the database, mo1 is the influence weight value corresponding to the data reception buffer occupancy preset in the database, mo2 is the influence weight value corresponding to the average reception time of the training dataset preset in the database, and mo3 is the influence weight value corresponding to the average loading rate of the training dataset preset in the database.

[0041] The above-mentioned reference data receiving buffer occupancy rate represents the reference value of the data receiving buffer occupancy rate; the above-mentioned reference training data set average receiving time represents the reference value of the average receiving time of the training data set; the above-mentioned reference training data set average loading rate represents the reference value of the average loading rate of the training data set.

[0042] The influence weight value corresponding to the above-mentioned data fluctuation index is used to de-unitize the data fluctuation index and represents the numerical value of the influence of the unit value of the data fluctuation index on the data reception anomaly index; the influence weight value corresponding to the above-mentioned data reception buffer occupancy rate represents the numerical value of the influence of the unit value of the data reception buffer occupancy rate on the data reception anomaly index; the influence weight value corresponding to the above-mentioned average reception time of the training dataset represents the numerical value of the influence of the unit value of the average reception time of the training dataset on the data reception anomaly index; the influence weight value corresponding to the above-mentioned average loading rate of the training dataset represents the numerical value of the influence of the unit value of the average loading rate of the training dataset on the data reception anomaly index; the database stores the correspondence between the data fluctuation index, the data reception buffer occupancy rate, the average reception time of the training dataset, and the average loading rate of the training dataset and their corresponding influence weight values. For example, by inputting the data fluctuation index, the data reception buffer occupancy rate, the average reception time of the training dataset, and the average loading rate of the training dataset into the database, the database can match the influence weight value corresponding to the data fluctuation index, the influence weight value corresponding to the data reception buffer occupancy rate, the influence weight value corresponding to the average reception time of the training dataset, and the influence weight value corresponding to the average loading rate of the training dataset, and the value range is between 0 and 1.

[0043] It needs to be explained that during the training of the initial model, there is a close relationship between the data receiving buffer occupancy rate, the average receiving time of the training data set and the average loading rate of the training data set, which jointly affect the data receiving anomaly index. Specifically, when the average loading rate of the training data set is lower than a certain reference value (that is, the loading is too slow), the amount of data loaded into the data receiving buffer per unit time will decrease. If the data processing capability of the model remains unchanged or is relatively slow at this time, the data receiving buffer occupancy rate will gradually increase. As the buffer occupancy rate increases, the model needs to wait longer to receive the complete data set for processing, thereby increasing the average receiving time of the training data set, deviating from the normal receiving time reference value. This deviation indicates that there is a bottleneck in the data receiving process, which may be that the data loading speed cannot keep up with the model processing. The processing speed or data processing capacity needs to be improved. Conversely, if the average loading rate of the training dataset is too high and the data processing capacity of the model fails to increase accordingly, then although the buffer occupancy rate may temporarily decrease, the excessively high loading rate may lead to data loss or untimely processing, which will also increase the data reception anomaly index. In addition, if the average reception time is too short, it may also mean that there is a mismatch between data loading and processing, that is, data loading is too fast but processing is too slow, or data loading and processing are both too fast but exceed the system's carrying capacity. Therefore, the balance between the data reception buffer occupancy rate, the average reception time of the training dataset, and the average loading rate of the training dataset is crucial. Abnormal changes in any parameter (whether too high or too low) may affect other parameters and lead to abnormal changes in the data reception anomaly index.

[0044] Furthermore, the data input batch value of the initial model is adjusted, and the specific adjustment process is: obtaining the model complexity of the initial model, and matching the data input minimum batch value of the initial model from the database; the model complexity of the above-mentioned initial model is the model complexity of the heavy-load line vector model; the data input minimum batch value of the above-mentioned initial model is specifically matched in the following process: storing the data input minimum batch value corresponding to each model complexity in the database, querying the model complexity stored in the database corresponding to the model complexity of the initial model, and the data input minimum batch value corresponding to the model complexity stored in the database is the data input minimum batch value of the initial model.

[0045] Obtain the performance set of the running equipment to which the initial model belongs within the training sub-cycle, and match the maximum batch value of data input for the initial model from the database; the performance set of the running equipment to which the initial model belongs within the training sub-cycle refers to a set of performance parameters of the running equipment to which the initial model belongs within the training sub-cycle, specifically including the real-time memory utilization and real-time running speed of the running equipment to which the initial model belongs within the training sub-cycle, and the maximum batch value of data input for the initial model. The specific matching process is: the maximum batch value of data input corresponding to each performance set is stored in the database, and the performance set of the running equipment to which the initial model belongs within the training sub-cycle is compared with each performance set stored in the database in turn for similarity (the comparison can be performed through the cosine similarity algorithm), and the performance set stored in the database that has the highest similarity with the performance set of the running equipment to which the initial model belongs within the training sub-cycle is selected. The maximum batch value of data input corresponding to the performance set stored in the database is the matched maximum batch value of data input.

[0046] The above-mentioned minimum data input batch value represents the minimum allowable value of the data input batch value of the initial model; the above-mentioned maximum data input batch value represents the maximum allowable value of the data input batch value of the initial model.

[0047] Obtaining a data input batch value for the initial model. If the data input batch value for the initial model is less than the minimum data input batch value for the initial model, then increasing the data input batch value based on the data reception anomaly index of the initial model during the training sub-cycle. The data input batch value for the initial model refers to the number of data samples that the initial model can receive and process at one time during each iteration or each data processing phase of the model training process. Before the initial model begins training, its data input batch value is determined based on the model's predefined settings. Once the initial model enters the training phase, its data input batch value is adjusted and optimized based on the actual training situation and needs. The specific adjustment process for the increase and adjustment is as follows: a data input batch value adjustment value (e.g., +32) is matched from the database based on the data reception anomaly index of the initial model during the training sub-cycle, and added to the data input batch value of the initial model. If the result of the addition is greater than or equal to the minimum data input batch value, the adjustment is complete. If the result of the addition is less than the minimum data input batch value, then the data input batch value of the initial model is directly adjusted to the minimum data input batch value.

[0048] If the data input batch value of the initial model is greater than the maximum data input batch value of the initial model, the data input batch value is reduced and adjusted according to the data reception anomaly index of the initial model in the training sub-cycle. The specific adjustment process of the above reduction adjustment is: according to the data reception anomaly index of the initial model in the training sub-cycle, the data input batch value adjustment value (such as -48) is matched from the database, and added to the data input batch value of the initial model (with a negative sign for addition). If the addition result is less than or equal to the maximum data input batch value, the adjustment is completed. If the addition result is greater than the maximum data input batch value, the data input batch value of the initial model is directly adjusted to the maximum data input batch value.

[0049] The minimum batch value of data input is marked as the minimum value of the data input reference batch value interval, and the maximum batch value of data input is marked as the maximum value of the data input reference batch value interval. If the data input batch value of the initial model belongs to the data input reference batch value interval, the data input batch value adjustment value is matched from the database according to the data reception anomaly index of the initial model in the training sub-cycle, and the data input batch value of the initial model is adjusted; the above-mentioned data input reference batch value interval is a closed interval; the above-mentioned data input batch value adjustment value refers to the specific value for increasing or decreasing the data input batch value of the initial model. The specific matching process is: the data input corresponding to each data reception anomaly index interval is stored in the database Batch value adjustment value, query the data reception anomaly index interval stored in the database to which the data reception anomaly index of the initial model in the training sub-cycle belongs, the data input batch value adjustment value corresponding to the data reception anomaly index interval stored in the database, that is, the matched data input batch value adjustment value. It should be explained that if the data reception anomaly index of the initial model in the training sub-cycle is less than the minimum value of the data reception anomaly interval, then the data input batch value adjustment value is to increase the data input batch value of the initial model; if the data reception anomaly index of the initial model in the training sub-cycle is greater than the maximum value of the data reception anomaly interval, then the data input batch value adjustment value is to decrease the data input batch value of the initial model.

[0050] Step 3: Collect the simulation data of each monitoring location point, compare and analyze it with the input data of each monitoring location point, and determine whether to modify the initial model.

[0051] In a specific embodiment, the present invention achieves accurate correction of the initial model by comparing simulation data with input data. By quantifying the simulation data execution rate, data distribution density and harmonic frequency deviation value, it can accurately locate model anomalies and take corresponding measures, such as adding abnormal module nodes, enabling frequency-varying parameter modules or adding detection location points, effectively improving the accuracy and adaptability of the model and ensuring the reliable application of the model on high-load lines.

[0052] Specifically, the determination of whether to correct the initial model is as follows: by comparing and analyzing the simulation data of each monitoring position point with the input data of each monitoring position point, the electrical simulation deviation factor of the initial model is obtained; the electrical simulation deviation threshold is extracted from the database, and the difference processing is performed on the electrical simulation deviation factor of the initial model, and the difference processing result is compared with the electrical simulation deviation threshold to obtain the electrical simulation deviation rate of the initial model, and compared with the electrical simulation deviation threshold stored in the database; if the electrical simulation deviation rate of the initial model is less than or equal to the electrical simulation deviation threshold, it is determined that the initial model is not corrected, and if the electrical simulation deviation rate of the initial model is greater than the electrical simulation deviation threshold, it is determined that the initial model is corrected; the above-mentioned electrical simulation deviation threshold represents the minimum value of the reasonable range of the electrical simulation deviation factor; the above-mentioned monitoring position point refers to a node randomly positioned in the initial model (which can be randomly positioned through a Poisson distribution model).

[0053] Specifically, the electrical simulation deviation factor of the initial model, the specific analysis process is: the simulation data of each monitoring position point includes the simulated harmonic frequency of each monitoring position point within the simulation period, the simulated voltage fluctuation rate of each monitoring position point within the simulation period and the simulated line loss value of each monitoring position point within the simulation period; the above-mentioned simulation period refers to the time period for monitoring the simulation of the initial model, and the specific duration is determined by the electrical engineer; the above-mentioned simulated harmonic frequency refers to the frequency corresponding to the component whose frequency is an integer multiple of the fundamental frequency in the voltage waveform of the monitoring position point; the above-mentioned simulated voltage fluctuation rate has the same meaning as the voltage fluctuation rate of each detection position point within the data detection period, that is, the degree of fluctuation of the simulated voltage value of each monitoring position point; the above-mentioned simulated line loss value represents the line loss heat value of the monitoring position point, wherein the simulated harmonic frequency, simulated voltage fluctuation rate and simulated line loss value can all be extracted from the simulation data of the initial model.

[0054] The input data of each monitoring location point includes the actual harmonic frequency of each monitoring location point during the simulation period, the actual voltage fluctuation rate of each monitoring location point during the simulation period, and the actual line loss value of each monitoring location point during the simulation period; the above-mentioned actual harmonic frequency refers to the harmonic frequency that each monitoring location point should have, which can be obtained from each training data set; the above-mentioned actual voltage fluctuation rate refers to the voltage fluctuation rate that each monitoring location point should have, which can be obtained from each training data set; the above-mentioned actual line loss value refers to the line loss value that each monitoring location point should have, which can be obtained from each training data set.

[0055] According to the model complexity of the heavy-load line vector model, an electrical simulation deviation factor correction value is matched, which represents the degree of correction of the electrical simulation deviation factor of the initial model. The specific matching process is: the electrical simulation deviation factor correction value corresponding to each model complexity is stored in the database, and the model complexity stored in the database corresponding to the model complexity of the heavy-load line vector model is queried. The electrical simulation deviation factor correction value corresponding to the model complexity stored in the database is the matched electrical simulation deviation factor correction value.

[0056] Obtaining the data reception anomaly index of the initial model in the training cycle refers to accumulating the data reception anomaly index of the initial model in several training sub-cycles and performing mean processing to obtain the index.

[0057] Quantify and summarize the influence of the deviation degree between the simulated harmonic frequency and the actual harmonic frequency on the electrical simulation deviation factor, the influence of the deviation degree between the simulated voltage fluctuation rate and the actual voltage fluctuation rate on the electrical simulation deviation factor, the influence of the deviation degree between the simulated line loss value and the actual line loss value on the electrical simulation deviation factor, and the influence of the data reception abnormality index on the electrical simulation deviation factor. Finally, the electrical simulation deviation factor is obtained by correcting the summary results. The electrical simulation deviation factor of the initial model is used to quantify the electrical simulation deviation degree of the initial model. The specific expression is:

[0058]

[0059] Where DFM is the electrical simulation deviation factor of the initial model, d is the number of each monitoring location, d = {1, 2, 3, ..., w}, w is the total number of detection locations, TF is the electrical simulation deviation factor correction value, SUM_DAI is the data reception anomaly index of the initial model during the training cycle, B is the influence weight value corresponding to the data reception anomaly index preset in the database, and YR d is the simulated harmonic frequency of the dth monitoring position in the simulation cycle, ΔYR d is the actual harmonic frequency of the dth monitoring position in the simulation period, GF d is the simulated voltage fluctuation rate of the dth monitoring point in the simulation cycle, ΔGF d is the actual voltage fluctuation rate of the dth monitoring point during the simulation period, HS d is the simulated line loss value of the dth monitoring point in the simulation cycle, ΔHS d is the actual line loss value of the dth monitoring position during the simulation period, ki1 is the impact weight value corresponding to the simulated harmonic frequency preset in the database, ki2 is the impact weight value corresponding to the simulated voltage fluctuation rate preset in the database, and ki3 is the impact weight value corresponding to the simulated line loss value preset in the database.

[0060] The influence weight value corresponding to the above-mentioned data reception anomaly index is used to de-unitize the data reception anomaly index, and represents the numerical value of the influence of the unit value of the data reception anomaly index on the electrical simulation deviation factor; the influence weight value corresponding to the above-mentioned simulated harmonic frequency represents the numerical value of the influence of the unit value of the simulated harmonic frequency on the electrical simulation deviation factor; the influence weight value corresponding to the above-mentioned simulated voltage fluctuation rate represents the numerical value of the influence of the unit value of the simulated voltage fluctuation rate on the electrical simulation deviation factor; the influence weight value corresponding to the above-mentioned simulated line loss value represents the numerical value of the influence of the unit value of the simulated line loss value on the electrical simulation deviation factor; the database stores the correspondence between the data reception anomaly index, simulated harmonic frequency, simulated voltage fluctuation rate and simulated line loss value and their corresponding influence weight values. For example, the data reception anomaly index, simulated harmonic frequency, simulated voltage fluctuation rate and simulated line loss value are input into the database, and the database can match the influence weight value corresponding to the data reception anomaly index, the influence weight value corresponding to the simulated harmonic frequency, the influence weight value corresponding to the simulated voltage fluctuation rate and the influence weight value corresponding to the simulated line loss value, and the value range is between 0 and 1.

[0061] It's important to explain that the data reception anomaly index (DRI) is an important metric for measuring data accuracy during simulation. A high DRI indicates significant uncertainty or error in data reception and processing during the simulation, which directly leads to increased simulation anomalies. Specifically, if the simulated harmonic frequency deviates significantly from the actual harmonic frequency, this deviation not only reflects the inaccuracy of the simulation model in predicting harmonic characteristics but also triggers a chain reaction. Since changes in harmonic frequency are often closely related to voltage fluctuations and line losses in the power system, significant deviations in the simulated harmonic frequency will also result in significant deviations from the actual voltage fluctuation. This, in turn, will cause the simulated line loss values to deviate significantly from the actual values. The accumulation and interaction of these deviations ultimately manifest themselves in the degree of deviation in the electrical simulation, casting doubt on the accuracy and reliability of the overall simulation results. Therefore, it can be said that there is a close correlation between the DRI, the deviation between the simulated harmonic frequency and the actual harmonic frequency, the deviation between the simulated voltage fluctuation and the actual voltage fluctuation, and the deviation between the simulated line loss and the actual line loss values, all of which collectively impact the accuracy and reliability of the electrical simulation.

[0062] Furthermore, the initial model is corrected, and the specific correction process is: obtain the simulation data execution rate of the initial model, and compare it with the simulation data execution threshold. If the simulation data execution rate of the initial model is less than or equal to the simulation data execution threshold, then obtain the data distribution density of each detection module belonging to the initial model, locate the detection modules whose data distribution density is less than the data distribution density threshold, mark them as abnormal modules, and increase the number of nodes of each abnormal module according to the electrical simulation deviation factor of the initial model; the simulation data execution rate of the above-mentioned initial model refers to the ratio of the data actually executed by the initial model during the simulation process to the input data, which is extracted from the data execution log of the initial model; the simulation data execution threshold represents the minimum value of the reasonable range of the simulation data execution rate of the initial model, which is extracted from the database; the data distribution density of each detection module belonging to the above-mentioned initial model refers to the ratio of the data actually executed by each detection module during the simulation process of the initial model. The degree of aggregation of the processed data within the simulation cycle is extracted from the data log of the initial model; the above-mentioned data distribution density threshold refers to the minimum value of the reasonable range of data distribution density of each detection module belonging to the initial model, which is extracted from the database; the above-mentioned detection modules refer to a set of modules with similar number of nodes, and the specific degree of similarity is determined by the data analyst and marked as each detection module. The above-mentioned increase in the number of nodes of each abnormal module refers to the electrical simulation deviation factor of the initial model matching the node number increase value from the database, and uniformly increasing the node number of each abnormal module according to the node number increase value; the node number increase value, the specific matching process is: the node number increase value corresponding to each electrical simulation deviation factor interval is stored in the database, and the electrical simulation deviation factor interval stored in the database to which the electrical simulation deviation factor of the initial model belongs is queried, and the node number increase value corresponding to the electrical simulation deviation factor interval stored in the database is the matched node number increase value.

[0063] If the simulation data execution rate of the initial model is greater than the simulation data execution threshold, the harmonic frequency deviation value of the initial model is obtained and compared with the harmonic frequency deviation threshold. If the harmonic frequency deviation value of the initial model is greater than the harmonic frequency deviation threshold, the frequency conversion parameter module is enabled; the above harmonic frequency deviation value, that is, The above-mentioned harmonic frequency deviation threshold represents the minimum value of the reasonable range of the harmonic frequency deviation value of the initial model and is extracted from the database; the above-mentioned frequency-varying parameter module is mainly used to process and analyze parameters related to frequency changes. In the power system, the dynamic changes of large loads will cause system frequency fluctuations, and the electrical parameters of the line such as resistance and inductance will change with the frequency. The primary task of the frequency-varying parameter module is to accurately capture these parameter characteristics that change with frequency. It collects the operating data of the line at different frequencies and uses special algorithms to calculate and update the frequency-varying parameters. For example, using signal processing technologies such as Fourier transform, the complex current and voltage signals in the line are decomposed into different frequency components, and then the changes in parameters under each frequency component are analyzed.

[0064] If the harmonic frequency deviation value of the initial model is less than or equal to the harmonic frequency deviation threshold, the number of detection position points is increased according to the electrical simulation deviation factor of the initial model, thereby completing the correction of the initial model; the above-mentioned increase in the number of detection position points specifically refers to matching the number of new detection position points from the database according to the electrical simulation deviation factor of the initial model, and evenly distributing them on the heavy-load line, thereby increasing the number of detection position points. The specific matching process of the number of new detection position points is: the number of new detection position points corresponding to each electrical simulation deviation factor interval is stored in the database, and the electrical simulation deviation factor interval stored in the database to which the electrical simulation deviation factor of the initial model belongs is queried. The number of new detection position points corresponding to the electrical simulation deviation factor interval stored in the database is the number of new detection position points matched.

[0065] It needs to be explained that the increase in the number of nodes and the number of newly added detection location points are matched by the electrical simulation deviation factor. The increase in the number of nodes corresponding to each electrical simulation deviation factor interval and the number of newly added detection location points corresponding to each electrical simulation deviation factor interval are divided into different intervals. The specific corresponding rules are formulated by electrical engineers.

[0066] In a specific embodiment, the present invention provides a vector model construction method for a heavy-load line. By accurately matching the model complexity, the practicality and accuracy of the model are ensured. By setting multiple detection location points on the heavy-load line and collecting and analyzing the electrical data of the multiple detection location points, the complexity of the vector model can be scientifically determined, thereby constructing an initial model that fits the reality. Another major advantage of this method is its dynamic adjustment and optimization capabilities. After selecting the training data set of the initial model, by monitoring the training process parameters, it can promptly determine whether the data input batch value needs to be adjusted to ensure the best training effect. In addition, by collecting simulation data from each monitoring location point and comparing and analyzing it with the actual input data, deviations in the model can be promptly discovered and corrected, making the model more perfect. Therefore, the present invention not only provides an effective method for constructing a heavy-load line vector model, but also ensures the accuracy and practicality of the model through a dynamic adjustment and optimization mechanism, providing strong support for the operation and maintenance of heavy-load lines.

[0067] The above content is merely an example and explanation of the structure of the present invention. Those skilled in the art may make various modifications or additions to the described specific embodiments or replace them in a similar manner. As long as they do not deviate from the structure of the invention or exceed the scope defined by the present invention, they should all fall within the scope of protection of the present invention.

Claims

1. A vector model construction method for heavy-load lines, characterized in that: include: Step 1: Set various detection points on the heavy-load line, collect and analyze electrical data from each detection point, and thus match the model complexity of the heavy-load line vector model; Step 2: Based on the model complexity of the heavy-load line vector model, a heavy-load line vector model is constructed and marked as an initial model. Training data sets for the initial model are selected and the initial model is trained. Training process parameters of the initial model are collected, and it is determined whether the data input batch value of the initial model should be adjusted. Step 3: Collect the simulation data of each monitoring location point, compare and analyze it with the input data of each monitoring location point, and determine whether to modify the initial model.

2. The method for constructing a vector model for a heavy-load line according to claim 1, characterized in that: The specific setting process of setting each detection position point on the heavy load line is as follows: Obtain the total length of the heavy-load line and match it with the detection position point spacing corresponding to each line total length interval in the database to obtain the detection position point spacing of the heavy-load line; Mark each detection position point on the heavy-load line in sequence according to the detection position point spacing of the heavy-load line.

3. The method for constructing a vector model for a heavy-load line according to claim 1, characterized in that: The model complexity of the heavy-load line vector model is matched, and the specific matching process is as follows: Analyze the electrical data of each detection location and obtain the data fluctuation index of the heavy-load line; Comparing the data fluctuation index of the heavy-load line with the data fluctuation index reference interval, if the data fluctuation index of the heavy-load line is greater than the maximum value of the data fluctuation index reference interval, the model complexity of the heavy-load line vector model matched from the database is the first model complexity; If the data fluctuation index of the heavy-load line belongs to the data fluctuation index reference interval, the model complexity of the heavy-load line vector model matched from the database is the second model complexity; If the data fluctuation index of the heavy-load line is less than the minimum value of the data fluctuation index reference interval, the model complexity of the heavy-load line vector model matched from the database is the third model complexity.

4. The method for constructing a vector model for a heavy-load line according to claim 3, characterized in that: The specific analysis process of the data fluctuation index of the heavy-load line is as follows: The electrical data of each detection location point includes the voltage fluctuation rate of each detection location point within the data detection period, the average drift rate of the harmonic impedance phase angle of each detection location point within the data detection period, and the total current harmonic content of each detection location point within the data detection period; The data fluctuation index is finally obtained by summarizing and averaging the influence of the ratio of the voltage fluctuation rate at each detection location to the defined voltage fluctuation rate, the influence of the ratio of the average drift rate of the harmonic impedance phase angle at each detection location to the defined harmonic impedance phase angle, and the influence of the ratio of the total current harmonic content at each detection location to the defined current harmonic content. The data fluctuation index of the heavy-load line is used to quantify the degree of data fluctuation of the heavy-load line.

5. The method for constructing a vector model for a heavy-load line according to claim 1, characterized in that: The specific process of determining whether to adjust the data input batch value of the initial model is as follows: By analyzing the training process parameters of the initial model, the data reception anomaly index of the initial model in the training sub-cycle is obtained; Extract the data reception anomaly interval from the database, compare the data reception anomaly index of the initial model in the training sub-cycle with the data reception anomaly interval, and if the data reception anomaly index of the initial model in the training sub-cycle does not fall within the data reception anomaly interval, determine that the data input batch value of the initial model does not need to be adjusted; If the data reception anomaly index of the initial model in the training sub-cycle does not belong to the data reception anomaly interval, it is determined that the data input batch value of the initial model should be adjusted.

6. The method for constructing a vector model for a heavy-load line according to claim 5, characterized in that: The data input batch value of the initial model is adjusted, and the specific adjustment process is as follows: Obtain the model complexity of the initial model and match the data input minimum batch value of the initial model from the database; Obtain the performance set of the running device to which the initial model belongs within the training sub-cycle, and match the maximum batch value of the data input of the initial model from the database; Obtain the data input batch value of the initial model. If the data input batch value of the initial model is less than the data input minimum batch value of the initial model, increase the data input batch value according to the data reception anomaly index of the initial model in the training sub-cycle. If the data input batch value of the initial model is greater than the maximum data input batch value of the initial model, the data input batch value is reduced and adjusted according to the data reception anomaly index of the initial model in the training sub-cycle; The data input minimum batch value is marked as the minimum value of the data input reference batch value interval, and the data input maximum batch value is marked as the maximum value of the data input reference batch value interval. If the data input batch value of the initial model belongs to the data input reference batch value interval, the data input batch value adjustment value is matched from the database according to the data reception anomaly index of the initial model in the training sub-cycle, and the data input batch value of the initial model is adjusted.

7. The method for constructing a vector model for a heavy-load line according to claim 6, characterized in that: The data reception anomaly index of the initial model in the training sub-cycle is analyzed in detail as follows: The training process parameters of the initial model include the data receiving buffer occupancy rate of the initial model in the training sub-cycle, the average receiving time of the training data set of the initial model in the training sub-cycle, and the average loading rate of the training data set of the initial model in the training sub-cycle; Quantify and summarize the impact of the data fluctuation index on the data reception anomaly index, the impact of the deviation between the data reception buffer occupancy rate and the reference data reception buffer occupancy rate on the data reception anomaly index, the impact of the deviation between the average reception time of the training dataset and the average reception time of the reference training dataset on the data reception anomaly index, and the impact of the deviation between the average loading rate of the training dataset and the average loading rate of the reference training dataset on the data reception anomaly index, and finally obtain the data reception anomaly index; The data reception anomaly index of the initial model in the training sub-period is used to quantify the degree of data reception anomaly of the initial model in the training sub-period.

8. The method for constructing a vector model for a heavy-load line according to claim 1, characterized in that: The specific determination process of whether to modify the initial model is as follows: By comparing and analyzing the simulation data of each monitoring location point with the input data of each monitoring location point, the electrical simulation deviation factor of the initial model is obtained; Extracting an electrical simulation deviation threshold from a database, performing difference processing on the electrical simulation deviation factor of the initial model, performing ratio processing on the difference processing result and the electrical simulation deviation threshold, obtaining an electrical simulation deviation rate of the initial model, and comparing the result with the electrical simulation deviation threshold stored in the database; If the electrical simulation deviation rate of the initial model is less than or equal to the electrical simulation deviation threshold, it is determined that the initial model is not corrected; if the electrical simulation deviation rate of the initial model is greater than the electrical simulation deviation threshold, it is determined that the initial model is corrected.

9. The method for constructing a vector model for a heavy-load line according to claim 8, characterized in that: The initial model is corrected, and the specific correction process is as follows: Obtain the simulation data execution rate of the initial model and compare it with the simulation data execution threshold. If the simulation data execution rate of the initial model is less than or equal to the simulation data execution threshold, obtain the data distribution density of each detection module belonging to the initial model, locate the detection modules whose data distribution density is less than the data distribution density threshold, mark them as abnormal modules, and increase the number of nodes of each abnormal module according to the electrical simulation deviation factor of the initial model; If the simulation data execution rate of the initial model is greater than the simulation data execution threshold, the harmonic frequency deviation value of the initial model is obtained and compared with the harmonic frequency deviation threshold. If the harmonic frequency deviation value of the initial model is greater than the harmonic frequency deviation threshold, the frequency variable parameter module is enabled; If the harmonic frequency deviation value of the initial model is less than or equal to the harmonic frequency deviation threshold, the number of detection position points is increased according to the electrical simulation deviation factor of the initial model, thereby completing the correction of the initial model.

10. The method for constructing a vector model for a heavy-load line according to claim 8, characterized in that: The electrical simulation deviation factor of the initial model is analyzed in the following way: The simulation data of each monitoring location point includes the simulated harmonic frequency of each monitoring location point within the simulation period, the simulated voltage fluctuation rate of each monitoring location point within the simulation period, and the simulated line loss value of each monitoring location point within the simulation period; The input data of each monitoring location point includes the actual harmonic frequency of each monitoring location point in the simulation period, the actual voltage fluctuation rate of each monitoring location point in the simulation period, and the actual line loss value of each monitoring location point in the simulation period; According to the model complexity of the heavy-load line vector model, the electrical simulation deviation factor correction value is matched; Obtain the data reception anomaly index of the initial model during the training cycle; Quantify and summarize the impact of the deviation between the simulated harmonic frequency and the actual harmonic frequency on the electrical simulation deviation factor, the impact of the deviation between the simulated voltage fluctuation rate and the actual voltage fluctuation rate on the electrical simulation deviation factor, the impact of the deviation between the simulated line loss value and the actual line loss value on the electrical simulation deviation factor, and the impact of the data reception anomaly index on the electrical simulation deviation factor. Finally, the electrical simulation deviation factor is obtained by correcting the summarized results. The electrical simulation deviation factor of the initial model is used to quantify the degree of electrical simulation deviation of the initial model.

Images