A power system source and load bilateral uncertainty characterization and tracing method and system

Abstract

The present application relates to the technical field of power system analysis, in particular to a power system source and load bilateral uncertainty characterization and tracing method and system, adopting the method provided by the present application, including dividing primary indicators, secondary indicators and tertiary indicators; obtaining data of the primary indicators at multiple time sections, setting a plurality of data judgment thresholds, classifying the current system operation state based on the judgment thresholds, performing step-by-step analysis and calculation through the feature importance score of each indicator about a certain indicator output by the machine learning model, through the above hierarchical and progressive analysis framework, combined with the machine learning method robust to collinearity, the source and load bilateral uncertainty in the high-proportion new energy power system can be quantified and traced systematically, providing a new, more accurate and scientific analysis tool for power grid risk warning, planning decision and operation optimization.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of power system analysis, in particular to a power system source-load bilateral uncertainty characterization and tracing method and system. BACKGROUND

[0002] With the deepening of the "double carbon" strategy, the new power system dominated by new energy is accelerating to be built. The strong volatility of new energy output and the increasingly complex electricity consumption behavior on the load side are intertwined, making both the source and load sides and the system regulation resources present high uncertainty. The traditional balance mechanism of "deterministic generation tracking uncertain load" in the traditional power system has evolved into a new mode of "bilateral uncertainty of source-load two-way random matching". This strong bilateral uncertainty of source and load brings great challenges to the reliable power supply (loss of load risk) and efficient consumption of new energy (curtailment risk) of the system.

[0003] The prior art has the following deficiencies in dealing with uncertainty:

[0004] (1) Single analysis dimension: Most studies only start from a few factors on the source side or the load side, which is difficult to cope with the increasing number of uncertainty factors and their complex coupling effects, and faces the problem of "combination explosion".

[0005] (2) Lack of systematic tracing framework: Existing researches are mostly focused on optimization scheduling under uncertain environment, while the quantitative characterization of uncertainty itself, especially the tracing of the transmission path from system-level operation risk (phenomenon) to basic technical indicators (root cause), lacks a systematic analysis framework.

[0006] (3) Limitations of analysis methods: Traditional correlation analysis cannot handle nonlinear relationships, and methods such as global sensitivity analysis (GSA) have strong assumptions about the independence of input variables, which are limited in practical scenarios where there is collinearity among indicators.

[0007] Therefore, there is an urgent need for a new method that can systematically quantify the strong uncertainty of source and load, accurately identify key influencing factors, and clearly reveal the transmission path of uncertainty, in order to improve the risk perception and decision support capability of the power grid. SUMMARY

[0008] The purpose of the present application is to provide a power system source-load bilateral uncertainty characterization and tracing method and system to solve the above problems in the prior art.

[0009] The present application is realized by the following technical solutions:

[0010] In a first aspect, the present application provides a power system source-load bilateral uncertainty characterization and tracing method, comprising:

[0011] Obtaining data of several power systems, grading the data of several power systems to several indexes, the several indexes including first-level indexes, second-level indexes and third-level indexes;

[0012] Obtaining data of the first-level indexes at multiple time sections, setting judgment thresholds of the data, classifying the current system operation state based on the judgment thresholds, if the classification is a certain state, then no processing is performed;

[0013] If the classification is an uncertain state, then several first machine learning models are established, each first machine learning model taking a certain first-level index as the only output and taking all second-level indexes as the input, and the feature importance score groups of each second-level index with respect to a certain first-level index in the output of each first machine learning model are obtained respectively, and the second-level index corresponding to the maximum feature importance score in the feature importance score group is selected as the key second-level index;

[0014] Then several second machine learning models are established, each second machine learning model taking a certain key second-level index as the only output and taking all third-level indexes as the input, and the feature importance score groups of each third-level index with respect to a certain key second-level index in the output of each second machine learning model are obtained respectively, and the third-level index corresponding to the maximum feature importance score in the feature importance score group is selected as the key third-level index;

[0015] The key third-level index is output as the source of uncertainty of the power system.

[0016] Preferably, the first-level indexes include load loss rate, load loss expectation, wind and light curtailment rate and new energy curtailment expectation;

[0017] The second-level indexes include net load daily maximum peak-valley difference rate, new energy peak regulation characteristic, source-load fluctuation matching degree, daily average new energy penetration rate and new energy penetration rate in load peak period;

[0018] The third-level indexes include new energy side indexes and load side indexes, the new energy side indexes include new energy installed capacity proportion, wind and photovoltaic daily output prediction accuracy, wind and photovoltaic daily utilization hours and wind and photovoltaic hourly output change rate, and the load side indexes include load daily total power prediction deviation rate, load hourly change rate, load daily maximum peak-valley difference rate and load daily peak duration.

[0019] Preferably, the setting of the judgment thresholds and the classification of the current system operation state based on the judgment thresholds include:

[0020] When the load loss rate and the wind and light curtailment rate are both greater than the respective preset thresholds, then the classification is a strong uncertainty state;

[0021] When only the load loss rate is greater than the preset threshold, then the classification is a load side uncertainty state.

[0022] When only the wind and light curtailment rate is greater than its preset threshold, then classified as source side uncertainty state;

[0023] When both are not greater than the respective preset threshold, then classified as certainty state.

[0024] Preferably, the machine learning model is a random forest model, based on random forest, by applying disturbance to out-of-bag data that does not participate in decision tree training, and then calculating the change of classification accuracy to obtain a score reflecting the importance of features.

[0025] Preferably, the loss of load rate includes:

[0026]

[0027]

[0028]

[0029]

[0030] In the formula, is the loss of load at the th time point, is the actual net load at the th time point, is the starting capacity of the thermal power unit on the day, is the starting capacity of other units on the day, is the maximum charge and discharge power of the energy storage, is the actual load at the th time point, is the actual output power of wind power at the th time point, is the actual output power of photovoltaic at the th time point, is the loss of load rate, is the 0-1 auxiliary variable of discrimination, is the corresponding maximum moment.

[0031] Preferably, the new energy curtailment amount expectation includes:

[0032]

[0033]

[0034]

[0035] In the formula, is the wind and light curtailment at the th time point, the minimum output power of the thermal power unit, the minimum output power of other units, the expected amount of abandoned new energy power,

[0036] Preferably, the daily average new energy penetration rate comprises:

[0037]

[0038] The new energy penetration rate during the load peak period comprises:

[0039]

[0040]

[0041] In the formula, is the daily average new energy penetration rate, is the new energy penetration rate during the load peak period, is the corresponding time, is the corresponding year.

[0042] Preferably, the net load daily maximum peak-valley difference rate comprises:

[0043]

[0044] In the formula, is the net load daily maximum peak-valley difference rate, is the corresponding day.

[0045] Preferably, the wind and photovoltaic daily utilization hours comprise:

[0046]

[0047]

[0048] In the formula, is the wind power utilization hours, is the predicted value of the wind turbine output at the th time point, is the photovoltaic power utilization hours, is the predicted value of the photovoltaic output at the th time point. In the second aspect, the present application also provides a power system source and load bilateral uncertainty characterization and tracing system for executing the power system source and load bilateral uncertainty characterization and tracing method described above, comprising:

[0049] In the second aspect, the present application also provides a power system source and load bilateral uncertainty characterization and tracing system for executing the power system source and load bilateral uncertainty characterization and tracing method described above, comprising:

[0050] ​The state judgment module is configured to obtain data of several power systems, grade the several power system data to several indexes, the several indexes including first-level indexes, second-level indexes and third-level indexes, obtain data of the first-level indexes at multiple time sections, set judgment thresholds of the several data, classify the current system operation state based on the judgment thresholds, and if the classification is a certain state, no processing is performed;

[0051] The traceability module is configured to if the classification is an uncertain state, establish several first machine learning models, each first machine learning model taking a certain first-level index as the only output and taking all second-level indexes as the input, respectively obtaining a feature importance score set of each second-level index about a certain first-level index output in each first machine learning model, selecting a second-level index corresponding to a feature importance score with the largest score in the feature importance score set as a key second-level index, then establishing several second machine learning models, each second machine learning model taking a certain key second-level index as the only output and taking all third-level indexes as the input, respectively obtaining a feature importance score set of each third-level index about a certain key second-level index output in each second machine learning model, selecting a third-level index corresponding to a feature importance score with the largest score in the feature importance score set as a key third-level index, and outputting the key third-level index as the source of uncertainty of the power system.

[0052] The technical scheme of the present application has at least the following advantages and beneficial effects:

[0053] By adopting the method provided by the present application, the first-level indexes, the second-level indexes and the third-level indexes are included, data of the first-level indexes at multiple time sections are obtained, judgment thresholds of the several data are set, the current system operation state is classified based on the judgment thresholds, the feature importance score of each index about a certain index output in the machine learning model is analyzed and calculated level by level, and through the above hierarchical and progressive analysis framework, combined with the machine learning method robust to multicollinearity, the source-load bilateral uncertainty in the high-proportion new energy power system can be quantitatively and systematically traced, thereby providing a new, more accurate and scientific analysis tool for risk early warning, planning decision and operation optimization of the power grid. BRIEF DESCRIPTION OF DRAWINGS

[0054] In order to more clearly illustrate the technical scheme of the embodiments of the present application, the following will briefly introduce the drawings needed to be used in the embodiments. It should be understood that the following drawings only show some embodiments of the present application, and therefore should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can also be obtained without creative labor on the basis of these drawings.

[0055] Figure 1 The flowchart of the present application;

[0056] Figure 2 The source-load bilateral strong uncertainty index system of the present application is shown in the figure;

[0057] Figure 3 The correlation matrix heat map between the secondary indexes of the present application is shown in the figure;

[0058] Figure 4 The importance of the secondary indexes of the present application to the loss load rate using random forest is shown in the figure;

[0059] Figure 5 The importance of the secondary indexes of the present application to the wind and light rejection rate using random forest is shown in the figure;

[0060] Figure 6 The global sensitivity analysis of the present application to the loss load rate is shown in the figure;

[0061] Figure 7 The global sensitivity analysis of the present application to the wind and light rejection rate is shown in the figure;

[0062] Figure 8 The importance of the tertiary indexes of the present application to the peak modulation characteristics of new energy is shown in the figure;

[0063] Figure 9 The importance of the tertiary indexes of the present application to the net load peak valley rate is shown in the figure. DETAILED DESCRIPTION

[0064] In order to make the objects, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are some but not all of the embodiments of the present application. The components of the embodiments of the present application described and shown in the drawings herein can be arranged and designed in various different configurations.

[0065] The terms "first", "second", and the like in the specification and claims of the present application and the above drawings are used to distinguish similar objects, and do not necessarily indicate a specific order or a chronological sequence. The naming or numbering of the steps appearing in the present application does not mean that the steps in the method flow must be executed in the order indicated by the naming or numbering, and the execution order of the named or numbered flow steps can be changed according to the technical purpose to be achieved, as long as the same or similar technical effect can be achieved.

[0066] The independent described module or submodule can be physically separated or not physically separated, can be software implemented or hardware implemented, and part of the module or submodule can be implemented by software, the function of the part of the module or submodule is called by the processor, and the other part of the module or submodule is implemented by hardware, for example, by hardware circuit.

[0067] Please refer to Figures 1-9 The application provides a power system source and load bilateral uncertainty characterization and tracing method, comprising:

[0068] S101: acquiring data of several power systems, classifying the data of the several power systems into several indexes, the several indexes comprising first-level indexes, second-level indexes and third-level indexes;

[0069] S102: acquiring data of the first-level indexes at multiple time sections, setting judgment thresholds of the data, classifying the current system operation state based on the judgment thresholds, and if the classification is a certain state, not processing;

[0070] This step aims to qualitatively classify the uncertainty level of each time section according to the actual consequences of system operation, and provide clear scene labels for subsequent tracing analysis. The classification basis is that the loss load rate and the wind and light curtailment rate in the first-level indexes are used as two-dimensional evaluation coordinates.

[0071] S103: if the classification is an uncertain state, then a plurality of first machine learning models are established, each first machine learning model takes a certain first-level index as the only output and takes all second-level indexes as the input, respectively obtains a feature importance score set of each second-level index about a certain first-level index output in each first machine learning model, and selects a second-level index corresponding to the maximum feature importance score in the feature importance score set as a key second-level index;

[0072] S104: then a plurality of second machine learning models are established, each second machine learning model takes a certain key second-level index as the only output and takes all third-level indexes as the input, respectively obtains a feature importance score set of each third-level index about a certain key second-level index output in each second machine learning model, and selects a third-level index corresponding to the maximum feature importance score in the feature importance score set as a key third-level index;

[0073] S104: outputting the key third-level index as the source of the uncertainty of the power system.

[0074] The method provided by the application comprises a primary index, a secondary index and a tertiary index; data of the primary index at multiple time sections is obtained, a plurality of data judgment thresholds are set, the current system operation state is classified based on the judgment thresholds, and each index about a certain index in the output of the machine learning model is analyzed and calculated in stages through the feature importance score, so that the above-mentioned layered and progressive analysis framework is combined with the machine learning method robust to collinearity, the source-load bilateral uncertainty in the high-proportion new energy power system can be quantified and traced in a systematic manner, a new, more accurate and scientific analysis tool is provided for risk early warning, planning decision and operation optimization of the power grid.

[0075] In an example embodiment of the application, the primary index comprises a loss of load rate, a loss of load expectation, a wind and solar curtailment rate and a new energy curtailment expectation; the primary index represents the final adverse result of the system under the impact of source-load bilateral uncertainty, which fails to completely suppress the risk. The loss of load rate and the loss of load expectation are core indexes for measuring the overall power supply capacity of the system, which directly quantify the final negative impact on social and economic activities when the system responds to uncertainty; the wind and solar curtailment rate and the new energy curtailment expectation are core indexes for measuring the new energy consumption capacity, which directly quantify the negative impact on the new energy development goal and the energy transformation strategy when the system responds to uncertainty.

[0076] Loss of Load Rate (LOLR):

[0077]

[0078] Net Load

[0079]

[0080]

[0081]

[0082] Expected Energy Not Supplied (EENS):

[0083]

[0084]

[0085] Wind Solar Curtailment Rate (WSCR):

[0086]

[0087]

[0088]

[0089] Expected Curtailed Renewable Energy (ECRE):

[0090]

[0091]

[0092] wherein, represents the curtailed load at the t-th time point, and is positive if curtailment occurs; represents the actual net load at the t-th time point; represents the on-line capacity of thermal power units on the day; represents the on-line capacity of other units (hydroelectric power, nuclear power, etc.) on the day; represents the maximum charge-discharge power of the energy storage; represents the actual load at the t-th time point; represents the actual output power of wind power at the t-th time point; represents the actual output power of photovoltaic at the t-th time point; represents the curtailed wind and light at the t-th time point, and is positive if curtailment occurs; represents the minimum output power of thermal power units; represents the minimum output power of other units (hydroelectric power, nuclear power, etc.); is a 0-1 auxiliary variable for discrimination, is the maximum time point.

[0093] The secondary indexes include a net load daily maximum peak-valley difference rate, a new energy peak regulation characteristic, a source-load fluctuation matching degree, a daily average new energy penetration rate, and a new energy penetration rate during a load peak period. The secondary indexes quantitatively reflect the system uncertainty characteristics after the interaction and mutual influence of the source and the load, and do not belong to the source side or the grid side alone. The daily average new energy penetration rate measures the proportion of new energy in daily power balance, and the higher the penetration rate, the greater the influence of the inherent uncertainty of new energy on the entire system. The new energy instantaneous penetration rate during the load peak period is representative and reflects the support capability of new energy for power supply during the load peak period. The net load is the final object that needs to be balanced by conventional power sources, and the net load daily maximum peak-valley difference rate comprehensively considers the peak-valley characteristics of the load and the output characteristics of new energy, and is the most core and direct comprehensive index for evaluating the demand for flexible resources of the system. The new energy peak regulation characteristic measures the change degree of the net load peak-valley difference after the access of new energy, and reflects the matching relationship between the output of new energy and the load peak. The source-load fluctuation matching degree quantitatively reflects the time synchronization of the fluctuation trend of the output of new energy and the load.

[0094] Specifically, the daily average new energy penetration rate (DAREP) is:

[0095]

[0096] The new energy penetration rate during the load peak period (REP-PL) is:

[0097]

[0098]

[0099] The net load daily maximum peak-valley difference rate (NLPVR) is:

[0100]

[0101] In the formula, is the load at a certain time, is the load in a certain year, is the load in a certain day.

[0102] The new energy peak regulation characteristic (RE-PRC) is:

[0103] The original load peak-valley difference is:

[0104]

[0105] Net load peak-valley difference:

[0106]

[0107] Peak-valley difference change value after connecting new energy:

[0108]

[0109] Greater than 0 indicates positive peak shaving effect, and the peak-valley difference decreases.

[0110] Source-Load Fluctuation Matching Degree (SL-FMD) is the correlation coefficient between new energy output and load.

[0111]

[0112] New energy output:

[0113]

[0114] In the formula, represents the peak-valley difference of the load on the day; represents the peak-valley difference of the net load on the day; represents the actual output of new energy (wind power and photovoltaic) at the tth time point; represents the average value of the actual output of new energy on the day; represents the average value of the actual load on the day.

[0115] The three-level index includes new energy side index and load side index, the new energy side index includes new energy installed capacity proportion, wind power and photovoltaic daily output prediction accuracy, wind power and photovoltaic daily utilization hours and wind power and photovoltaic 1-hour output change rate, and the load side index includes load daily total power prediction deviation rate, load 1-hour change rate, load daily maximum peak-valley difference rate and load daily peak duration.

[0116] The tertiary indicators respectively quantify the inherent, independent statistical characteristics and prediction accuracy of the new energy load from both sides, which are the basic indicators for measuring the uncertainty of the power system. The new energy installed capacity ratio, similar to the penetration rate, reflects the influence of the uncertainty of the new energy side on the entire system; the wind / solar power daily output prediction accuracy quantifies the prediction deviation caused by insufficient understanding (fuzziness) or meteorological randomness, and improving the prediction accuracy is one of the keys to reducing uncertainty; the wind / solar power daily utilization hours reflect the endowment of daily new energy resources and are a direct embodiment of "intermittency", and low utilization hours mean long periods of low output, increasing the demand for system reserve capacity; the wind / solar power 1h output change rate directly quantifies the "volatility" of new energy, reflecting the rate of rise or fall of new energy output in a short period of time, and determines the demand for system flexibility; the load daily total power prediction deviation rate quantifies the uncertainty of load-side prediction; the load 1h change rate reflects the volatility characteristics of the load itself and is also a major indicator for the demand for system flexibility; the load daily maximum peak-valley difference rate reflects the basic demand for system peak regulation capacity; the load daily peak duration reflects the increasingly significant load "peaking" characteristics. It should be noted that the longer the load peak duration, the smaller the overall load change rate, and the less significant the "peaking" characteristics.

[0117] New energy side indicators:

[0118] New energy installed capacity ratio (RE-ICR):

[0119]

[0120] Wind / solar power daily output prediction accuracy (WPFA / SPFA):

[0121]

[0122]

[0123] Wind / solar power daily generation utilization hours (WPUH / SPUH):

[0124]

[0125]

[0126] Hourly Wind / Solar Power Change Rate (WPCR / SPCR):

[0127]

[0128]

[0129] Load-side indicators:

[0130] Daily Load Forecast Accuracy (LFA):

[0131]

[0132] Hourly Load Change Rate (LCR):

[0133]

[0134] Daily Load Peak-to-Valley Rate (DLPVR):

[0135]

[0136] Daily Peak Load Duration (DPLD):

[0137] The time when the load is above 98% of the daily maximum load is defined as the load peak time:

[0138]

[0139]

[0140] wherein, represents the on-line capacity of the wind turbine on the day; represents the on-line capacity of the photovoltaic on the day; represents the predicted value of the wind turbine output at the tth time point; represents the predicted value of the photovoltaic output at the tth time point; represents the wind power output at the t+4th time point, and the reason why it is is that the time interval is 15 min in the embodiment, and 4 time intervals are 1 h, Similarly; represents the predicted value of the load at the tth time point.

[0141] Data acquisition: based on the actual operation data or production simulation data of the power system, the values of each index in the three-level index system at continuous multiple time sections (such as daily) are calculated to form a multi-dimensional time series data set.

[0142] In an example embodiment of the present application, the setting of the judgment threshold value includes classifying the current system operation state based on the judgment threshold value.

[0143] When the load loss rate and the wind and light curtailment rate are both greater than the respective preset threshold values, the system is classified as a strong uncertainty state;

[0144] When only the load loss rate is greater than the preset threshold value, the system is classified as a load-side uncertainty state;

[0145] When only the wind and light curtailment rate is greater than the preset threshold value, the system is classified as a source-side uncertainty state;

[0146] When neither of them is greater than the respective preset threshold value, the system is classified as a deterministic state.

[0147] It should be noted that the "load-side uncertainty system" and "source-side uncertainty system" here are not uncertainties originating from the load side or the new energy side, but the system-level uncertainty is ultimately reflected in the load-side power supply or the new energy-side consumption.

[0148] In an example embodiment of the present application, the machine learning model is a random forest model, and based on the random forest, the change in classification accuracy is calculated by applying disturbance to out-of-bag data that does not participate in decision tree training to obtain a score reflecting feature importance.

[0149] Specifically, the hierarchical progressive key factor identification and tracing based on the random forest. This step is the core of the present application, which aims to overcome the limitations of traditional analysis methods in dealing with multi-factor, non-linear, and collinearity problems, and to identify the key driving factors of uncertainty layer by layer and build the transmission path.

[0150] (1) Method selection: the random forest (RandomForest) algorithm which is robust to collinearity problems and does not require strict linear assumptions is used as the main analysis tool, and its built-in feature importance evaluation function based on permutation importance is used to obtain feature importance by applying disturbance to out-of-bag (OOB) data that does not participate in decision tree training and then calculating the change in classification accuracy.

[0151] Suppose there are M features in the sample, and the random forest randomly selects K data sets, corresponding to K OOB data sets, and the steps for sorting feature importance are as follows:

[0152] 1) Initialization .

[0153] 2) Train decision tree using the first data set , calculate the classification accuracy of the second OOB data set . .

[0154] 3) Apply perturbation to the features in the OOB data set, recalculate the accuracy . .

[0155] 4) Repeat steps 2) and 3) for .

[0156] 5) Calculate the importance of the features using the following formula, i.e.:

[0157]

[0158] 6) Sort in descending order to obtain the feature importance ranking, with higher ranking indicating higher importance. In the formula,

[0159] is the importance score of the ith feature. (2) Key coupling indicator identification. Train two independent random forest regression proxy models with LOLR and RECR as the target outputs and all secondary indicators as the inputs. Utilize the built-in feature importance evaluation function based on the prediction error of replacement to quantitatively calculate and rank the importance scores of each secondary indicator, identifying 1-2 key secondary indicators with the highest contribution to LOLR and RECR.

[0160] (3) Uncertainty depth tracing. Take the key secondary indicators identified in the previous step as the target outputs and all tertiary indicators (including source side and load side) as the inputs to train a new random forest proxy model. Similarly, utilize feature importance evaluation to calculate and rank the importance scores of each tertiary indicator, identifying the key tertiary indicators that have the greatest impact on the key secondary indicators.

[0161] (4) Conduction path construction: Based on the analysis results of the second and third layers, construct a data-driven uncertainty conduction path from the "key tertiary indicators" to the "key secondary indicators" and then to the "primary risk indicators".

[0162] 4) Uncertainty source judgment and cross-validation. This step aims to comprehensively judge and verify the reliability of the analysis results.

[0163]

[0164] ​​​(1) Source judgment: In all the identified key three-level indicators, according to the pre-defined properties (source side or load side), the importance weight proportion is analyzed, the greater the importance weight proportion, the higher the three-level indicator ranking, the greater the impact, so as to judge the main source of uncertainty leading to the overall risk of the system.

[0165] (2) Cross-validation: In order to ensure the reliability of the random forest analysis conclusion, global sensitivity analysis (GSA-Sobol' method) based on variance is used for cross-validation. In particular, in order to solve the problem of unstable proxy model of GSA on small sample data, the application adopts integrated neural network technology to build a proxy model. This technology fixes the random seed to ensure reproducibility, and integrates the prediction results of multiple (such as 10) independently trained neural networks, so as to obtain a stable and high-precision proxy model. By comparing the sensitivity index (especially the total sensitivity index TSI) calculated by GSA with the importance ranking of random forest, the consistency of the key factor identification result is verified.

[0166] The application also provides an electric power system source and load bilateral uncertainty characterization and tracing system for executing the electric power system source and load bilateral uncertainty characterization and tracing method.

[0167] The state judgment module is configured to obtain data of several electric power systems, classify the data of the several electric power systems into several indexes, the several indexes including first-level indexes, second-level indexes and third-level indexes, obtain data of the first-level indexes at multiple time sections, set judgment thresholds for the data, classify the current system running state based on the judgment thresholds, and if the classification is a certain state, then no processing is performed.

[0168] The tracing module is configured to, if the classification is an uncertain state, establish several first machine learning models, each first machine learning model taking a certain first-level index as the only output and taking all second-level indexes as the input, respectively obtaining a feature importance score set of each second-level index with respect to a certain first-level index in the output of each first machine learning model, and selecting a second-level index corresponding to a maximum feature importance score in the feature importance score set as a key second-level index; then establish several second machine learning models, each second machine learning model taking a certain key second-level index as the only output and taking all third-level indexes as the input, respectively obtaining a feature importance score set of each third-level index with respect to a certain key second-level index in the output of each second machine learning model, selecting a third-level index corresponding to a maximum feature importance score in the feature importance score set as a key third-level index; and outputting the key third-level index as the source of uncertainty of the electric power system.

[0169] This embodiment provides a specific example for further explaining the above content.

[0170] Index system construction, data acquisition and system state classification.

[0171] Data preparation: based on the operation data of Shandong power grid in 2024. The data includes: daily meteorological data (wind speed, light intensity), daily load peak and valley value, installed capacity of various types of units (thermal power, wind power, photovoltaic, energy storage) and so on. Based on the above basic data, 365 days of load curve, wind power and photovoltaic output curve with time resolution of 15 minutes are generated through production simulation.

[0172] Index calculation: according to the index system defined in section 1 of the "invention content" part, the numerical values of 20 first, second and third level indexes of each day in 365 days are calculated, forming a 365x20 data matrix.

[0173] System state classification: set the risk threshold of LOLR and RECR as 1%. Classify 365 days, and the example results show that: the number of "deterministic" state days is the most, a total of 288 days; the number of "strong uncertainty" state days is the least, a total of 2 days; the number of "source side uncertainty" and "load side uncertainty" days is 33 and 42 days respectively. The classification results provide different risk level sample sets for subsequent analysis.

[0174] 2. Key coupling index identification.

[0175] Collinearity diagnosis: the correlation matrix heat map between all indexes shows that there is strong correlation between some indexes, and the clustered indexes point to the same scene, such as the condition of large new energy generation or heavy load on the same day. It can be said that it is redundant, so it is necessary to identify the key uncertainty index. The correlation matrix heat map and variance inflation factor (VIF) of 5 second level indexes are calculated, and the results show that SL-FMD, REP-PL and NLPVR have strong correlation, and the VIF value is greater than 20, indicating that there is serious collinearity. This diagnostic result confirms the limitation of directly using GSA for all analysis, highlighting the necessity of using random forest in this invention.

[0176] Random forest analysis:

[0177] ① Model configuration: use the fitrensemble function in MATLAB, set 'Method' as 'Bag' and 'NumLearningCycles' as 100.

[0178] ② Analysis of LOLR: take 5 second level indexes as input and LOLR as output to train the model. Use predictorImportance function to calculate feature importance. The example results show that the importance score of RE-PRC (new energy peak regulation characteristic) is the highest.

[0179] ③ Analysis of RECR: Using five secondary indicators as input and RECR as output, the same analysis was performed. The results show that NLPVR (Net Load Daily Maximum Peak-to-Valley Ratio) has the highest importance score, far exceeding other indicators.

[0180] ④ Identification Results: Therefore, this step identifies RE-PRC and NLPVR as key secondary coupling indicators that lead to supply risk and absorption risk, respectively.

[0181] GSA cross-validation:

[0182] ① Input selection: To conduct effective GSA, three secondary indicators with weak collinearity were selected as input variables: DAREP (daily average new energy penetration rate), RE-PRC, and NLPVR.

[0183] ② Proxy Model Construction: Construct an ensemble neural network proxy model. Set the ensemble size to 10, the random seed rng to 123, and the network structure to [10, 5].

[0184] ③GSA execution: Perform Sobol' analysis with a basic sample size N of 1024. Calculate the first-order (FSI) and total-order (TSI) sensitivity indices of each input to LOLR and RECR.

[0185] ④ Verification conclusion: FSI reflects the input variables The effect of TSI on system output reflects the input variables. as well as The combined effect of the interaction with all other input random variables on the system output can be used to reflect the non-superposition characteristic of the input's influence on the output. GSA results show that the peak-shaving characteristics of new energy sources have the largest FSI for LOLR, and the daily maximum peak-valley difference rate of net load has the largest FSI for RECR, indicating that they have the largest individual impact on the primary indicators, verifying that they are the key indicators in the source-load coupling index. However, the TSI of other indicators is not 0 and is even greater than that of the key indicators, indicating that other indicators do have an impact on the primary indicators; their impact is reflected in the primary indicators through their interaction with the key indicators.

[0186] 3. In-depth tracing and source determination of uncertainty.

[0187] (1) In-depth source tracing:

[0188] ① Model configuration: The fitrensemble function in MATLAB is still used, with 'Method' set to 'Bag' and 'NumLearningCycles' set to 100.

[0189] ②Traceability analysis for RE-PRC: Take all three-level indicators as input and RE-PRC as output to train the random forest model. Evaluate the importance of each three-level indicator.

[0190] ③Traceability analysis for NLPVR: Take all three-level indicators as input and NLPVR as output to perform the same analysis.

[0191] (2) Result analysis and conduction path construction:

[0192] Based on the importance ranking results of the above two models, the example identifies that RE-ICR (new energy installed capacity ratio), SPUH (photovoltaic daily power generation utilization hours) and DPLD (daily peak load duration) are the top three key three-level indicators in overall contribution degree. This result shows that:

[0193] ① The new energy installed capacity ratio is the most important indicator among the three-level indicators. The large-scale access of new energy is the primary factor for the increase of uncertainty.

[0194] ② The photovoltaic power generation utilization hours is the second important indicator among the three-level indicators, indicating that the uncertainty introduced by photovoltaic power generation is large. This is consistent with the current situation of large-scale access of distributed photovoltaic in Shandong Province.

[0195] Based on this, a clear uncertainty conduction path can be constructed, for example: "the increase of RE-ICR (three-level indicator) is the primary factor leading to the increase of NLPVR (two-level indicator), and the increase of NLPVR is the key driving force leading to the deterioration of RECR (one-level indicator)".

[0196] (3) Uncertainty source judgment: Among the three key three-level indicators identified, RE-ICR and SPUH belong to new energy side indicators, and DPLD belongs to load side indicators. From the number (2:1) and cumulative importance weight, it can be concluded that in this example, the strong uncertainty faced by the system is mainly from the source side.

[0197] The embodiments of the present application realize comprehensive characterization of source and load uncertainty of the power system, accurate identification of key factors and clear traceability of conduction path through the above steps, and provide strong technical support for the power grid to cope with uncertainty challenges.

[0198] In addition, each functional unit in each embodiment of the present application can be integrated in one processing unit, or each unit can exist physically, or two or more units can be integrated in one unit. The integrated unit can be realized in the form of hardware or in the form of software functional unit.

[0199] The integrated unit, if implemented in the form of a software function unit and sold or used as an independent product, can be stored in a computer readable storage medium. The computer software product stored in a storage medium includes a plurality of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the embodiments of the method of the present application. The aforementioned storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a magnetic disk or an optical disk, and various media that can store program codes.

[0200] The above merely describes the preferred embodiments of the present application and is not intended to limit the present application. The present application can have various modifications and changes for those skilled in the art. Any modification, equivalent replacement, improvement, etc. within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A power system source-load bilateral uncertainty characterization and tracing method, characterized in that, Comprising; obtaining data of several power systems, grading the data of several power systems to several indexes, the several indexes including first-level indexes, second-level indexes and third-level indexes; obtaining data of the first-level indexes at multiple time sections, setting judgment thresholds of the data, classifying the current system operation state based on the judgment thresholds, and if the classification is a certain state, not processing; if the classification is an uncertain state, establishing several first machine learning models, each first machine learning model taking a certain first-level index as the only output and taking all second-level indexes as the input, respectively obtaining a feature importance score set of each second-level index about a certain first-level index output in each first machine learning model, and selecting a second-level index corresponding to the maximum feature importance score in the feature importance score set as a key second-level index; then establishing several second machine learning models, each second machine learning model taking a certain key second-level index as the only output and taking all third-level indexes as the input, respectively obtaining a feature importance score set of each third-level index about a certain key second-level index output in each second machine learning model, and selecting a third-level index corresponding to the maximum feature importance score in the feature importance score set as a key third-level index; outputting the key third-level index as the source of uncertainty of the power system; the first-level indexes include loss of load rate, loss of load expectation, wind and light curtailment rate and new energy curtailment expectation; the second-level indexes include net load daily maximum peak-valley difference rate, new energy peak regulation characteristics, source-load fluctuation matching degree, daily average new energy penetration rate and new energy penetration rate in load peak period; the third-level indexes include new energy side indexes and load side indexes, the new energy side indexes include new energy installed capacity proportion, wind and photovoltaic daily output prediction accuracy, wind and photovoltaic daily utilization hours and wind and photovoltaic hourly output change rate, and the load side indexes include load daily total power prediction deviation rate, load hourly change rate, load daily maximum peak-valley difference rate and load daily peak duration; the setting of the judgment thresholds and the classification of the current system operation state based on the judgment thresholds include: when the loss of load rate and the wind and light curtailment rate are both greater than their respective preset thresholds, the classification is a strong uncertainty state; when only the loss of load rate is greater than its preset threshold, the classification is a load side uncertainty state; when only the wind and light curtailment rate is greater than its preset threshold, the classification is a source side uncertainty state; when both are not greater than their respective preset thresholds, the classification is a certain state; the machine learning model is a random forest model, and the score reflecting the feature importance is obtained by applying disturbance to the out-of-bag data not participating in the decision tree training and then calculating the change of the classification accuracy based on the random forest.

2. The power system source and load bilateral uncertainty characterization and tracing method of claim 1, wherein, the loss of load rate includes: In the formula, For the first The amount of load loss at each point in time. For the first The actual net load at each point in time This represents the operating capacity of the thermal power units on that day. This represents the operating capacity of other generating units for the day. This represents the maximum charge and discharge power of the energy storage. For the first The actual load at each point in time For the first The actual output power of wind power at a given time point For the first The actual output power of photovoltaics at each point in time. The load shedding rate, 0-1 auxiliary variables for judgment, This corresponds to the maximum time.

3. The power system source and load bilateral uncertainty characterization and provenance method of claim 2, wherein, the new energy curtailment expectation includes: In the formula, is the minimum output power of the thermal power unit, is the amount of wind and light abandoned at the i th time point, is the minimum output power of the thermal power unit, is the minimum output power of the other units, is the expected amount of new energy abandoned electricity.

4. The power system source and load bilateral uncertainty characterization and tracing method of claim 3, wherein, the daily average new energy penetration rate includes: the new energy penetration rate in the load peak period includes: In the formula, is the daily average new energy penetration rate, is the renewable energy penetration, is the corresponding time, is the corresponding year.

5. The power system source and load bilateral uncertainty characterization and provenance method of claim 4, wherein, the net load daily maximum peak-valley difference rate includes: In the formula, is the maximum peak-valley difference rate of the net load on the day, is the corresponding day.

6. The power system source and load bilateral uncertainty characterization and provenance method of claim 5, wherein, the wind and photovoltaic daily utilization hours include: wherein is the wind power utilization hours, is the predicted value of the wind farm output at the time point, is the photovoltaic power utilization hours, is the predicted value of the photovoltaic output at the time point.

7. A power system source-load bilateral uncertainty characterization and provenance system configured to perform the power system source-load bilateral uncertainty characterization and provenance method of any one of claims 1-6, wherein, Comprising: The state judgment module is configured to obtain data of several power systems, grade the data of the several power systems into several indexes, the several indexes including first-level indexes, second-level indexes, and third-level indexes, obtain data of the first-level indexes at multiple time sections, set judgment thresholds for the data, classify the current system operation state based on the judgment thresholds, and if the classification is a certain state, then no processing is performed; The traceability module is configured to if the classification is an uncertain state, then establish several first machine learning models, each first machine learning model taking a certain first-level index as the only output and taking all second-level indexes as the input, respectively obtaining a feature importance score set of each second-level index about a certain first-level index output in each first machine learning model, selecting a second-level index corresponding to a feature importance score with the largest score in the feature importance score set as a key second-level index, then establishing several second machine learning models, each second machine learning model taking a certain key second-level index as the only output and taking all third-level indexes as the input, respectively obtaining a feature importance score set of each third-level index about a certain key second-level index output in each second machine learning model, selecting a third-level index corresponding to a feature importance score with the largest score in the feature importance score set as a key third-level index, and outputting the key third-level index as a source of uncertainty of the power system.

Images