Power load prediction method and system based on big data
By clustering and error rate optimization of the power load metering nodes, only some node data are obtained for prediction, the problem of excessive data acquisition in power load prediction is solved, and efficient resource utilization and prediction efficiency are achieved.
Patent Information
- Application Number
- CN202411080165.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-07
- Publication Date
- 2024-12-06
- Estimated Expiration
- 2044-08-07
AI Technical Summary
In power load prediction, as the number of metering nodes increases, the pressure for data acquisition and transmission increases significantly, resulting in the problem of excessive acquisition of metering data and excessive resource consumption during the prediction process.
By clustering the metrology nodes, only the data of some nodes are obtained for prediction, and the data of other nodes is predicted by deep learning models, reducing the resource pressure for data acquisition. The specific steps include obtaining the grid model to configure the metering node, synchronously calculating the instantaneous load, counting and comparing the instantaneous load, randomly selecting nodes that meet the conditions for prediction, and updating the node conditions based on the error rate to optimize the prediction process.
It effectively reduces the resource pressure of the measurement node data acquisition, reduces the amount of data acquisition, improves prediction efficiency, and significantly reduces resource consumption only when the accuracy is reduced.
Smart Images

Figure CN119093331A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of power load forecasting, and in particular to a power load forecasting method and system based on big data. Background Art
[0002] Electricity load forecasting includes two meanings: it refers to various electrical equipment installed at users such as government agencies, enterprises, and residents, and it can also be used to describe the value of the power consumed by the above-mentioned electrical equipment. Electricity load forecasting is a series of forecasting work based on electricity load. From the perspective of the forecast object, electricity load forecasting includes the forecast of future electricity demand (power), the forecast of future electricity consumption (energy), and the forecast of load curves.
[0003] The power load forecasting process requires the use of a large number of data collection devices. One of the prediction methods is to directly obtain the data collected by these collection devices, process the collected data based on deep learning and other methods, and train a prediction model. In practical applications, it is only necessary to collect data from all collection devices at regular intervals to predict the power load in the future. This is a global prediction solution that relies on the data of all metering nodes. In fact, when the number of metering nodes is large, the pressure of data acquisition and transmission is very high. How to further optimize and reduce the amount of metering data acquired during the prediction process is the technical problem that the technical solution of the present invention wants to solve. Summary of the invention
[0004] The purpose of the present invention is to provide a method and system for power load forecasting based on big data to solve the problems raised in the above background technology.
[0005] To achieve the above object, the present invention provides the following technical solutions:
[0006] A method for predicting power load based on big data, the method comprising:
[0007] Obtain the power grid model, configure the metering nodes based on the power grid model and number them;
[0008] Obtain power load based on metering nodes and calculate instantaneous load synchronously;
[0009] Count the instantaneous load within a preset time period, compare the counted instantaneous load, and cluster the metering nodes;
[0010] When a prediction instruction is received, a metering node is randomly selected until the selected metering node meets the preset node condition, the instantaneous load of the selected metering node is obtained, and the instantaneous load of all metering nodes is predicted based on the clustering result;
[0011] Synchronously verifying the error rate of the previous prediction process, and updating the node condition based on the error rate;
[0012] The node condition is a quantity threshold. When the number of any type of metering nodes among the selected metering nodes is greater than the quantity threshold, the node condition is considered to be met.
[0013] As a further solution of the present invention: the step of acquiring the power load based on the metering node and synchronously calculating the instantaneous load includes:
[0014] For each metering node, the power load is obtained regularly according to the preset frequency;
[0015] The acquired power load is sorted based on time, and a discrete Fourier transform is performed on the sorted power load to determine the frequency distribution characteristics;
[0016] A preset number of frequencies are selected from the frequency distribution characteristics, and a frequency-based trigonometric function and a frequency-based trigonometric function are created as a reference function for the fitting process;
[0017] The sorted power load is fitted by the benchmark function to obtain the fitting function of each metering node;
[0018] Deriving the fitting function to obtain a load derivative function, and determining the instantaneous load at each moment based on the load derivative function;
[0019] The benchmark function is: Where N is the preset number; a0, a n and b n are all randomly determined constants, w n It is the nth frequency value after sorting the selected N frequencies.
[0020] As a further solution of the present invention: the step of counting instantaneous loads within a preset time period, comparing the counted instantaneous loads, and clustering the metering nodes includes:
[0021] Receive the time period input by the management party, and intercept the load derivative function within the time period;
[0022] For any two metering nodes, calculate the difference function of the load derivative function, and calculate the integral of the difference function over the time period as the node distance;
[0023] The metering nodes are clustered based on the node distances.
[0024] As a further solution of the present invention: the step of clustering the metering nodes based on the node distance comprises:
[0025] The number of categories entered by the receiving management party;
[0026] Randomly select the number of metering nodes of the classification as the category center among the metering nodes;
[0027] For any metering node, it is classified according to its distance from the center of each category;
[0028] After all metering nodes are classified once, the category center is updated;
[0029] The loop is executed until a preset exit condition is reached; the exit condition includes that the number of iterations reaches a threshold number and the update amount of the category center is less than a preset update amount threshold;
[0030] For any metering node, the rule for classifying it according to its distance from the center of each category is: label i =argmin 1≤j≤k ||x i -μ j ||; label i is the classification result of the i-th metering node, x i represents the i-th metering node, μ j represents the jth category center, and k represents the total number of category centers;
[0031] The updated category center is the center of the same type of metering nodes in space.
[0032] As a further solution of the present invention: when a prediction instruction is received, a metering node is randomly selected until the selected metering node meets a preset node condition, the instantaneous load of the selected metering node is obtained, and the step of predicting the instantaneous load of all metering nodes based on the clustering result includes:
[0033] When a prediction instruction is received, a metering node is randomly selected to obtain the category to which it belongs;
[0034] Calculate the number of metering nodes of each category, and when the number does not exceed a preset number threshold, continue to select, and when the number exceeds the preset number threshold, stop selecting;
[0035] Obtain the instantaneous load of the selected metering node and calculate its ratio to the instantaneous load at the previous moment;
[0036] For other metering nodes of the same category, the product of the instantaneous load at the previous moment and the ratio is calculated as the predicted instantaneous load.
[0037] As a further solution of the present invention: the step of synchronously verifying the error rate of the previous prediction process and updating the node condition based on the error rate includes:
[0038] Read the calculated ratio, and calculate the absolute value of the difference between the ratio and 1 as the error rate;
[0039] The scaling ratio is determined according to the direct proportion of the error rate, and the scaling ratio is multiplied by the quantity threshold and rounded to an integer to be used as the updated node condition.
[0040] The technical solution of the present invention also provides a power load forecasting system based on big data, the system comprising:
[0041] A node configuration module is used to obtain a power grid model, configure metering nodes based on the power grid model, and number them;
[0042] The instantaneous load acquisition module is used to obtain the power load based on the metering node and synchronously calculate the instantaneous load;
[0043] The node clustering module is used to count the instantaneous load within a preset time period, compare the counted instantaneous load, and cluster the metering nodes;
[0044] The node selection application module is used to randomly select a metering node when receiving a prediction instruction until the selected metering node meets the preset node condition, obtain the instantaneous load of the selected metering node, and predict the instantaneous load of all metering nodes based on the clustering result;
[0045] A node condition updating module, used for synchronously verifying the error rate of the previous prediction process and updating the node condition based on the error rate;
[0046] The node condition is a quantity threshold. When the number of any type of metering nodes among the selected metering nodes is greater than the quantity threshold, the node condition is considered to be met.
[0047] As a further solution of the present invention: the instantaneous load acquisition module includes:
[0048] The power load acquisition unit is used to acquire the power load of each metering node according to a preset frequency timing;
[0049] A spectrum identification unit, used to sort the acquired power loads based on time, and perform discrete Fourier transform on the sorted power loads to determine frequency distribution characteristics;
[0050] A reference function determination unit, used for selecting a preset number of frequencies in the frequency distribution characteristics, and creating a frequency-based trigonometric function and as a reference function for the fitting process;
[0051] The power load fitting unit is used to fit the sorted power load with the reference function to obtain the fitting function of each metering node;
[0052] A fitting function derivation unit, used for derivation of the fitting function to obtain a load derivative function, and determining the instantaneous load at each moment based on the load derivative function;
[0053] The benchmark function is: Where N is the preset number; a0, a n and b n are all randomly determined constants, w n It is the nth frequency value after sorting the selected N frequencies.
[0054] As a further solution of the present invention: the node clustering module includes:
[0055] A derivative function interception unit, used for receiving a time period input by a management party, and intercepting a load derivative function within the time period;
[0056] The node distance calculation unit is used to calculate the difference function of the load derivative function for any two metering nodes, and calculate the integral of the difference function over the time period as the node distance;
[0057] A clustering execution unit is used to cluster the metering nodes based on the node distance.
[0058] As a further solution of the present invention: the node selection application module includes:
[0059] A category query unit, used for randomly selecting a metering node and obtaining the category to which the metering node belongs when receiving a prediction instruction;
[0060] A quantity determination unit, used to calculate the quantity of metering nodes of each category, and when the quantity does not exceed a preset quantity threshold, continue to select, and when the quantity exceeds the preset quantity threshold, stop selecting;
[0061] A ratio calculation unit, used to obtain the instantaneous load of the selected metering node and calculate the ratio of the instantaneous load to the instantaneous load at the previous moment;
[0062] The comparison prediction unit is used to calculate the product of the instantaneous load at the previous moment and the ratio for other metering nodes of the same category as the predicted instantaneous load.
[0063] Compared with the prior art, the beneficial effect of the present invention is that when acquiring data of metering nodes, the present invention clusters the metering nodes according to the data of the metering nodes. In the actual use stage, only the collected data of a limited number of metering nodes are acquired, and the data of other metering nodes are predicted by the collected data, and then the prediction model obtained by deep learning is applied. This greatly reduces the resource pressure of acquiring data from the metering nodes while only partially reducing the accuracy. BRIEF DESCRIPTION OF THE DRAWINGS
[0064] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present invention.
[0065] Figure 1 This is a flowchart of the power load forecasting method based on big data.
[0066] Figure 2 This is the first sub-process flowchart of the power load forecasting method based on big data.
[0067] Figure 3 This is the second sub-process flowchart of the power load forecasting method based on big data.
[0068] Figure 4 This is the third sub-process flowchart of the power load forecasting method based on big data.
[0069] Figure 5 This is the fourth sub-process flowchart of the power load forecasting method based on big data.
[0070] Figure 6 This is a structural block diagram of the power load forecasting system based on big data. DETAILED DESCRIPTION
[0071] In order to make the technical problems, technical solutions and beneficial effects to be solved by the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.
[0072] Figure 1 The flowchart of the power load forecasting method based on big data is as follows. In an embodiment of the present invention, a power load forecasting method based on big data includes:
[0073] Step S100: Acquire a power grid model, configure metering nodes based on the power grid model and number them;
[0074] The power grid model is a model composed of power supply lines and power supply equipment. In the power supply site, the power grid model belongs to existing data, that is, for the technical solution of the present invention, the power grid model is known data. The power grid model is obtained, and metering nodes are configured and numbered in the power grid model. The metering nodes are points where monitoring equipment is installed in the power grid. The monitoring equipment can use conventional power meters. In actual applications, each metering node is marked in the power grid model for each installation. The function of the power grid model containing metering nodes is mainly to connect with the administrator, so that the administrator can control the monitoring process of the entire power system.
[0075] Step S200: acquiring power load based on the metering node and synchronously calculating instantaneous load;
[0076] An electric power monitoring meter is installed at the metering node. The electric power monitoring meter is used to obtain the voltage, current or other electric power data in the power grid, collectively referred to as the electric power load, and process the electric power load to calculate the instantaneous load. It should be noted that in actual scenarios, the data obtained by the electric power monitoring meter is instantaneous data itself, such as instantaneous current, etc., that is, the electric power load in the above content is instantaneous data. On this basis, the instantaneous load further calculated is actually the rate of change of the instantaneous data.
[0077] Step S300: Count the instantaneous load within a preset time period, compare the counted instantaneous load, and cluster the metering nodes;
[0078] The management party of the execution body of this method sets a time period, such as ten minutes, and obtains the instantaneous load of all metering nodes within the time period. By comparing the instantaneous load, similar metering nodes can be classified into one category. In this process, it should be noted that the instantaneous load is the rate of change of instantaneous data, which is a unitless value. The metering nodes are clustered based on the instantaneous load, and the metering nodes can be clustered without limitation on type.
[0079] Step S400: when a prediction instruction is received, a metering node is randomly selected until the selected metering node meets the preset node condition, the instantaneous load of the selected metering node is obtained, and the instantaneous load of all metering nodes is predicted based on the clustering result;
[0080] When the execution subject of the method receives a prediction instruction, it randomly selects a metering node, determines the category of the metering node according to the clustering result, and determines in real time how many metering nodes of each category there are in the selected metering nodes. When the number of each category of metering nodes is greater than a preset number threshold, it is considered that the node condition is met, and then the instantaneous load of the selected metering node is used as the instantaneous load of other metering nodes of the same category. In this process, it is equivalent to simplifying the global load acquisition process of multiple metering nodes into a load sampling process.
[0081] Step S500: synchronously verifying the error rate of the previous prediction process, and updating the node condition based on the error rate;
[0082] Each time a prediction is made, data from a limited number of metering nodes (randomly selected) are obtained. The data from these metering nodes are real, and the others are predicted. In the previous prediction process, the data from the same metering node was most likely predicted. By comparing the real data and predicted data of the same metering node, the error rate can be determined. The larger the error rate, the more metering nodes are needed, and the smaller the error rate, the fewer metering nodes are needed.
[0083] Specifically, the node condition is a quantity threshold. When the number of any type of metering nodes among the selected metering nodes is greater than the quantity threshold, the node condition is considered to be met. When the error rate is large, the quantity threshold is large, and when the error rate is small, the quantity threshold is small.
[0084] Figure 2 The first sub-flow chart of the power load prediction method based on big data, wherein the steps of obtaining the power load based on the metering node and synchronously calculating the instantaneous load include:
[0085] Step S201: For each metering node, obtain the power load according to a preset frequency timing;
[0086] Step S202: sorting the acquired power loads based on time, and performing discrete Fourier transform on the sorted power loads to determine frequency distribution characteristics;
[0087] Step S203: selecting a preset number of frequencies from the frequency distribution characteristics, and creating a frequency-based trigonometric function sum as a reference function for the fitting process;
[0088] Step S204: fitting the sorted power loads with the reference function to obtain a fitting function for each metering node;
[0089] Step S205: deriving the fitting function to obtain a load derivative function, and determining the instantaneous load at each moment based on the load derivative function;
[0090] The benchmark function is: Where N is the preset number; a0, a n and b n are all randomly determined constants, w n It is the nth frequency value after sorting the selected N frequencies.
[0091] For each metering node, the power load is obtained regularly according to the preset frequency. After the power load is obtained, it is sorted according to the time sequence to obtain a discrete data group. In practical applications, the power load has a certain periodicity. For example, the power load in industrial areas is higher during working hours on weekdays, and lower during rest time, but this periodicity is not strict.
[0092] By performing discrete Fourier transform on discrete data sets, we can get a spectrum diagram, obtain the frequency distribution in the discrete data set, select the most important frequencies, and create a trigonometric function and form from the frequencies as the reference function of the fitting process. Then, the discrete data set is fit by the reference function. Since the frequency is known, only the amplitude and phase need to be determined. The fitting process can use the minimum distance as the goal, that is, the sum of the distances between each data in the discrete data set and the corresponding function value in the fitting function is minimized.
[0093] The fitting function is derived to obtain a load derivative function, and the instantaneous load at each moment, that is, the rate of change of the instantaneous data, is determined based on the load derivative function.
[0094] Figure 3 The second sub-flow chart of the power load prediction method based on big data, wherein the steps of counting instantaneous loads within a preset time period, comparing the counted instantaneous loads, and clustering the metering nodes include:
[0095] Step S301: receiving a time period input by a management party, and extracting a load derivative function within the time period;
[0096] Step S302: for any two metering nodes, calculate the difference function of the load derivative function, and calculate the integral of the difference function over the time period as the node distance;
[0097] Step S303: clustering the metering nodes based on the node distances.
[0098] The above content is a clustering process, which receives the time period input by the management party and intercepts the load derivative function within the time period. The time period can be five minutes, ten minutes or other shorter time. After the time period is determined, the time period part is intercepted in the determined load derivative function.
[0099] For any metering node, the load derivative function within the time period can be obtained. For any two metering nodes, the difference function between the load derivative functions is calculated, and then the integral of the difference function over the time period is calculated as the node distance, which is used to characterize the degree of difference between the two nodes. The metering nodes can be clustered based on the node distance.
[0100] Specifically, the step of clustering the metering nodes based on the node distance includes:
[0101] The number of categories entered by the receiving management party;
[0102] Randomly select the number of metering nodes of the classification as the category center among the metering nodes;
[0103] For any metering node, it is classified according to its distance from the center of each category;
[0104] After all metering nodes are classified once, the category center is updated;
[0105] The loop is executed until a preset exit condition is reached; the exit condition includes that the number of iterations reaches a threshold and the update amount of the category center is less than a preset update amount threshold.
[0106] The above content provides a specific metering node clustering scheme, which is essentially an application of the K-means clustering method. First, the number of categories input by the management party of the execution subject of this method is received, and the number of metering nodes of the category is randomly selected as the category center among all the metering nodes. Then, for any metering node, the nearest category center is selected to classify it, and the number of categories can be obtained. After all metering nodes are classified once, the center of each type of metering node is calculated as the new category center; finally, the above process is executed in a loop. If the category center no longer changes or the change is small enough, the classification is considered to be completed. At this time, the last classification result can be retained.
[0107] For any metering node, the rule for classifying it according to its distance from the center of each category is: label i =argmin 1≤j≤k ||x i -μ j ||; label i is the classification result of the i-th metering node, x i represents the i-th metering node, μ j represents the jth category center, and k represents the total number of category centers;
[0108] The updated category center is the center of the same type of metering nodes in space.
[0109] Figure 4 The third sub-flow chart of the power load prediction method based on big data, when receiving the prediction instruction, randomly selecting a metering node until the selected metering node meets the preset node condition, obtaining the instantaneous load of the selected metering node, and predicting the instantaneous load of all metering nodes based on the clustering result includes:
[0110] Step S401: when a prediction instruction is received, a metering node is randomly selected to obtain the category to which it belongs;
[0111] Step S402: Calculate the number of metering nodes of each category, and when the number does not exceed the preset number threshold, continue to select; when the number exceeds the preset number threshold, stop selecting;
[0112] Step S403: Obtain the instantaneous load of the selected metering node, and calculate its ratio to the instantaneous load at the previous moment;
[0113] Step S404: For other metering nodes of the same category, the product of the instantaneous load at the previous moment and the ratio is calculated as the predicted instantaneous load.
[0114] When a prediction instruction is received, it means that the user needs to obtain the metering data of each metering node. When the number of metering nodes is large, the process of obtaining data actually requires a lot of resources. Under the architecture of the technical solution of the present invention, only a limited number of metering nodes need to be randomly selected. When the selected metering node meets the conditions, the metering data of the selected metering node is obtained, and the metering data of other metering nodes of the same type are predicted. The prediction method is:
[0115] For the selected metering node, the instantaneous load of the selected metering node is obtained, and its ratio to the instantaneous load at the previous moment is calculated. For the same type of metering nodes, the instantaneous load at the previous moment is obtained, and its product with the ratio is calculated to obtain the predicted instantaneous load of the metering node.
[0116] Figure 5 The fourth sub-process flowchart of the power load forecasting method based on big data, wherein the step of synchronously verifying the error rate of the previous forecasting process and updating the node condition based on the error rate includes:
[0117] Step S501: Read the calculated ratio, and calculate the absolute value of the difference between the ratio and 1 as the error rate;
[0118] Step S502: determining a scaling ratio according to the direct proportion of the error rate, multiplying the scaling ratio by the quantity threshold and rounding the result to an integer, and using the result as the updated node condition.
[0119] In an example of the technical solution of the present invention, the calculated ratio is read, the absolute value of the difference between the ratio and one is calculated, and the difference is converted into a percentage as the error rate; the scaling ratio is determined by the error rate, and the larger the error rate, the larger the scaling ratio. The updated quantity threshold can be obtained by multiplying the scaling ratio by the quantity threshold and rounding the integer. The larger the error rate, the larger the quantity threshold, and the more metering nodes are selected.
[0120] It should be noted that the range of the scaling ratio can be pre-set by the administrator, for example, 0 to 3. At this time, a constant is determined based on this range, and the error rate is converted to the range of 0 to 3 by the constant, wherein the rounding process is generally rounded up.
[0121] Figure 6 : is a structural block diagram of a power load forecasting system based on big data. In an embodiment of the present invention, a power load forecasting system based on big data, the system 10 includes:
[0122] The node configuration module 11 is used to obtain a power grid model, configure metering nodes based on the power grid model and number them;
[0123] The instantaneous load acquisition module 12 is used to acquire the power load based on the metering node and synchronously calculate the instantaneous load;
[0124] The node clustering module 13 is used to count the instantaneous load within a preset time period, compare the counted instantaneous load, and cluster the metering nodes;
[0125] The node selection application module 14 is used to randomly select a metering node until the selected metering node meets the preset node condition when receiving the prediction instruction, obtain the instantaneous load of the selected metering node, and predict the instantaneous load of all metering nodes based on the clustering result;
[0126] A node condition updating module 15, used for synchronously verifying the error rate of the previous prediction process and updating the node condition based on the error rate;
[0127] The node condition is a quantity threshold. When the number of any type of metering nodes among the selected metering nodes is greater than the quantity threshold, the node condition is considered to be met.
[0128] Furthermore, the instantaneous load acquisition module 12 includes:
[0129] The power load acquisition unit is used to acquire the power load of each metering node according to a preset frequency timing;
[0130] A spectrum identification unit, used to sort the acquired power loads based on time, and perform discrete Fourier transform on the sorted power loads to determine frequency distribution characteristics;
[0131] A reference function determination unit, used for selecting a preset number of frequencies in the frequency distribution characteristics, and creating a frequency-based trigonometric function and as a reference function for the fitting process;
[0132] The power load fitting unit is used to fit the sorted power load with the reference function to obtain the fitting function of each metering node;
[0133] A fitting function derivation unit, used for derivation of the fitting function to obtain a load derivative function, and determining the instantaneous load at each moment based on the load derivative function;
[0134] The benchmark function is: Where n is the preset number; a0, a n and b n are all randomly determined constants, w nIt is the nth frequency value after sorting the selected N frequencies.
[0135] Specifically, the node clustering module 13 includes:
[0136] A derivative function interception unit, used for receiving a time period input by a management party, and intercepting a load derivative function within the time period;
[0137] The node distance calculation unit is used to calculate the difference function of the load derivative function for any two metering nodes, and calculate the integral of the difference function over the time period as the node distance;
[0138] A clustering execution unit is used to cluster the metering nodes based on the node distance.
[0139] Furthermore, the node selection application module 14 includes:
[0140] A category query unit, used for randomly selecting a metering node and obtaining the category to which the metering node belongs when receiving a prediction instruction;
[0141] A quantity determination unit, used to calculate the quantity of metering nodes of each category, and when the quantity does not exceed a preset quantity threshold, continue to select, and when the quantity exceeds the preset quantity threshold, stop selecting;
[0142] A ratio calculation unit, used to obtain the instantaneous load of the selected metering node and calculate the ratio of the instantaneous load to the instantaneous load at the previous moment;
[0143] The comparison prediction unit is used to calculate the product of the instantaneous load at the previous moment and the ratio for other metering nodes of the same category as the predicted instantaneous load.
[0144] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.
Claims
1. A method for power load forecasting based on big data, characterized in that: The method comprises: Obtain the power grid model, configure the metering nodes based on the power grid model and number them; Obtain power load based on metering nodes and calculate instantaneous load synchronously; Count the instantaneous load within a preset time period, compare the counted instantaneous load, and cluster the metering nodes; When a prediction instruction is received, a metering node is randomly selected until the selected metering node meets the preset node condition, the instantaneous load of the selected metering node is obtained, and the instantaneous load of all metering nodes is predicted based on the clustering result; Synchronously verifying the error rate of the previous prediction process, and updating the node condition based on the error rate; The node condition is a quantity threshold. When the number of any type of metering nodes among the selected metering nodes is greater than the quantity threshold, the node condition is considered to be met.
2. The power load forecasting method based on big data according to claim 1 is characterized in that: The step of acquiring the power load based on the metering node and synchronously calculating the instantaneous load comprises: For each metering node, the power load is obtained regularly according to the preset frequency; The acquired power load is sorted based on time, and a discrete Fourier transform is performed on the sorted power load to determine the frequency distribution characteristics; A preset number of frequencies are selected from the frequency distribution characteristics, and a frequency-based trigonometric function and a frequency-based trigonometric function are created as a reference function for the fitting process; The sorted power load is fitted by the benchmark function to obtain the fitting function of each metering node; Deriving the fitting function to obtain a load derivative function, and determining the instantaneous load at each moment based on the load derivative function; The benchmark function is: Where N is the preset number; a0, a n and b n are all randomly determined constants, w n It is the nth frequency value after sorting the selected N frequencies.
3. The power load forecasting method based on big data according to claim 2 is characterized in that: The step of counting instantaneous loads within a preset time period, comparing the counted instantaneous loads, and clustering the metering nodes comprises: Receive the time period input by the management party, and intercept the load derivative function within the time period; For any two metering nodes, calculate the difference function of the load derivative function, and calculate the integral of the difference function over the time period as the node distance; The metering nodes are clustered based on the node distances.
4. The power load forecasting method based on big data according to claim 3 is characterized in that: The step of clustering the metering nodes based on the node distance comprises: The number of categories entered by the receiving management party; Randomly select the number of metering nodes of the classification as the category center among the metering nodes; For any metering node, it is classified according to its distance from the center of each category; After all metering nodes are classified once, the category center is updated; The loop is executed until a preset exit condition is reached; the exit condition includes that the number of iterations reaches a threshold number and the update amount of the category center is less than a preset update amount threshold; For any metering node, the rule for classifying it according to its distance from the center of each category is: label i =argmin1 ≤j≤k ||x i -μ j ||; label i is the classification result of the i-th metering node, x i represents the i-th metering node, μ j represents the jth category center, and k represents the total number of category centers; The updated category center is the center of the same type of metering nodes in space.
5. The method for predicting power load based on big data according to claim 1, characterized in that: When a prediction instruction is received, a metering node is randomly selected until the selected metering node meets a preset node condition, the instantaneous load of the selected metering node is obtained, and the step of predicting the instantaneous load of all metering nodes based on the clustering result includes: When a prediction instruction is received, a metering node is randomly selected to obtain the category to which it belongs; Calculate the number of metering nodes of each category, and when the number does not exceed a preset number threshold, continue to select, and when the number exceeds the preset number threshold, stop selecting; Obtain the instantaneous load of the selected metering node and calculate its ratio to the instantaneous load at the previous moment; For other metering nodes of the same category, the product of the instantaneous load at the previous moment and the ratio is calculated as the predicted instantaneous load.
6. The method for predicting power load based on big data according to claim 5, characterized in that: The step of synchronously verifying the error rate of the previous prediction process and updating the node condition based on the error rate comprises: Read the calculated ratio, and calculate the absolute value of the difference between the ratio and 1 as the error rate; The scaling ratio is determined according to the direct proportion of the error rate, and the scaling ratio is multiplied by the quantity threshold and rounded to an integer to be used as the updated node condition.
7. A power load forecasting system based on big data, characterized in that: The system comprises: A node configuration module is used to obtain a power grid model, configure metering nodes based on the power grid model, and number them; The instantaneous load acquisition module is used to obtain the power load based on the metering node and synchronously calculate the instantaneous load; The node clustering module is used to count the instantaneous load within a preset time period, compare the counted instantaneous load, and cluster the metering nodes; The node selection application module is used to randomly select a metering node when receiving a prediction instruction until the selected metering node meets the preset node condition, obtain the instantaneous load of the selected metering node, and predict the instantaneous load of all metering nodes based on the clustering result; A node condition updating module, used for synchronously verifying the error rate of the previous prediction process and updating the node condition based on the error rate; The node condition is a quantity threshold. When the number of any type of metering nodes among the selected metering nodes is greater than the quantity threshold, the node condition is considered to be met.
8. The power load forecasting system based on big data according to claim 7 is characterized in that: The instantaneous load acquisition module comprises: The power load acquisition unit is used to acquire the power load of each metering node according to a preset frequency timing; A spectrum identification unit, used to sort the acquired power loads based on time, and perform discrete Fourier transform on the sorted power loads to determine frequency distribution characteristics; A reference function determination unit, used for selecting a preset number of frequencies in the frequency distribution characteristics, and creating a frequency-based trigonometric function and as a reference function for the fitting process; The power load fitting unit is used to fit the sorted power load with the reference function to obtain the fitting function of each metering node; A fitting function derivation unit, used for derivation of the fitting function to obtain a load derivative function, and determining the instantaneous load at each moment based on the load derivative function; The benchmark function is: Where N is the preset number; a0, a n and b n are all randomly determined constants, w n It is the nth frequency value after sorting the selected N frequencies.
9. The power load forecasting system based on big data according to claim 8 is characterized in that: The node clustering module includes: A derivative function interception unit, used for receiving a time period input by a management party, and intercepting a load derivative function within the time period; The node distance calculation unit is used to calculate the difference function of the load derivative function for any two metering nodes, and calculate the integral of the difference function over the time period as the node distance; A clustering execution unit is used to cluster the metering nodes based on the node distance.
10. The power load forecasting system based on big data according to claim 9, characterized in that: The node selection application module includes: A category query unit, used for randomly selecting a metering node and obtaining the category to which the metering node belongs when receiving a prediction instruction; A quantity determination unit, used to calculate the quantity of metering nodes of each category, and when the quantity does not exceed a preset quantity threshold, continue to select, and when the quantity exceeds the preset quantity threshold, stop selecting; A ratio calculation unit, used to obtain the instantaneous load of the selected metering node and calculate the ratio of the instantaneous load to the instantaneous load at the previous moment; The comparison prediction unit is used to calculate the product of the instantaneous load at the previous moment and the ratio for other metering nodes of the same category as the predicted instantaneous load.
Citation Information
Patent Citations
Short-term power energy load parallel-prediction method applied to power quality comprehensive-governance scene and system
CN108734355A
Medium and short term electricity demand prediction analysis method based on demand side dynamic response
CN116630086A
Industrial user load feasible region prediction method and system based on quick response
CN116865246A
Metering data lightweight traceability method based on artificial intelligence clustering
CN117093753A
Load prediction method and load prediction system employing general distribution
WO2021213192A1
