A method and system for regulating a power system by fusing data-driven risk perception
By constructing a time series prediction and over-limit probability prediction model and combining it with active dispatch control, we can achieve probability prediction and active regulation of future risk scenarios of the power system, solve the shortcomings of passive dispatch in existing technologies, and improve the risk response capabilities and decision-making accuracy of the power grid.
Patent Information
- Application Number
- CN202410433062.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-04-11
- Publication Date
- 2025-10-14
- Estimated Expiration
- 2044-04-11
AI Technical Summary
Existing power system dispatching and control methods are mostly passive modes, lacking forward-looking warning and proactive response capabilities for future risk scenarios. Traditional physical model-driven risk assessment methods are insufficient when faced with complex power grids, and existing data-driven methods lack in-depth consideration of future scenarios.
Construct a time series prediction model, an over-limit probability prediction model and an active dispatch control model, use long-short term memory network and support vector machine classification model, combine the characteristics of key variables in the power system, and conduct probability prediction and dispatch strategy formulation for future risk scenarios, including data-driven safety risk constraints.
It has achieved forward-looking early warning and proactive regulation of power system risks, provided specific risk information, enhanced the accuracy and efficiency of decision-making, and improved the operational safety and risk response capabilities of the power grid.
Smart Images
Figure CN118378743B_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application belongs to the technical field of power regulation, and more particularly relates to a regulation method and system for a power system integrating data-driven risk perception. BACKGROUND
[0002] With the rapid development of UHV AC / DC hybrid power grids and renewable energy, the power grid structure and energy supply mode are undergoing unprecedented changes. This change mainly reflects two aspects: first, UHV technology allows the power system to transmit power over a longer distance, which means that energy can be optimally configured in a wider area; second, with the large-scale integration of renewable energy such as wind and solar energy, the power supply structure of the power grid is increasingly diversified. The output characteristics of these new energy sources, especially their intermittency and uncertainty, are having a profound impact on the power grid. This impact not only leads to profound changes in the power system itself, but also makes global monitoring, network-wide prevention and control, and centralized decision-making particularly critical.
[0003] In response to the dual uncertainty of sources and loads, the current dispatching and control methods are mostly still in a passive mode. The fundamental problem of this passive control strategy is that it is based on the current known system state to make decisions, rather than on the future possible state. For example, when abnormal situations such as overload or voltage out-of-limit occur in the power grid, dispatchers often start to act only after the event has occurred. This passive reaction mode has obvious time delay and effect limitation. First, it often means that the best control time window is missed, resulting in the inability to fully utilize resources with longer response times, such as demand response resources. Second, passive control strategies often result in greater losses for the power grid when safety incidents occur. Considering the urgency and complexity of dispatching decisions, the safety and stability of the power grid may thus be in a passive and uncertain state.
[0004] Considering the passivity of the current dispatching and control mode, it is particularly important to predict possible regulation scenarios in advance and develop strategies accordingly. Currently, research in this area has begun to focus on power system situation awareness and transient stability assessment. Although these studies provide valuable information about the future state of the power system, relying solely on this state category information (label information) may not be sufficient. Staff still need to develop actual strategies based on actual situations and dispatching experience. Therefore, many scholars have begun to focus on risk warning research, hoping to provide more accurate decision-making support for dispatchers.
[0005] For the risk assessment and early warning of power systems, traditional methods are mostly based on physical model-driven and focus on the deterministic characteristics of events, such as line overload, voltage out-of-limit scenarios, etc. However, with the large-scale integration of new energy and power electronic devices, the dynamic characteristics of the system become more and more complex. This makes the evaluation method based on fixed assumptions or models gradually show limitations in estimating the actual risk of the system. In order to more accurately describe and quantify the risk, data-driven evaluation methods have received more and more attention in recent years. The core idea of these methods is to learn the data behavior characteristics of the system directly from the measured data, rather than based on predetermined physical model assumptions. However, although the data-driven method enhances the accuracy and speed of the evaluation to some extent, existing researches still mainly focus on the improvement of indicators and label-type risk warning, lacking of quantitative consideration of the possibility. Recent researches have begun to explore methods that combine scenario probability and risk assessment to provide a more comprehensive risk perspective. For example, some researches use advanced machine learning techniques such as generative adversarial networks and iterative random forests, combined with real-time data for dynamic risk probability assessment. These methods not only have good robustness, but also can provide more comprehensive risk information. However, the existing research methods still focus on evaluating the current risk state of the system, and lack of in-depth consideration of the possible scenarios in the future. How to combine these evaluation results with the formulation of dispatching control strategies is still a problem to be studied. SUMMARY
[0006] The present application provides a kind of regulation and control method and system of power system fusion data-driven risk perception, solve the problems that prior art cannot realize the forward warning of the security risk of power system, the active response capability of the risk of power system needs to be improved.
[0007] The present application provides a kind of regulation and control method of power system fusion data-driven risk perception, comprising:
[0008] A time series prediction model, an out-of-limit probability prediction model and an active dispatching control model are constructed, and the active dispatching control model includes a data-driven security risk constraint;
[0009] Based on the key variable feature subset of the power system, the multi-dimensional measurement time series data of the power system is collected; the multi-dimensional measurement time series data is input into the time series prediction model, and the time series prediction model outputs a prediction data vector corresponding to the key variable feature;
[0010] The prediction data vector is input into the out-of-limit probability prediction model, and the out-of-limit probability prediction model outputs out-of-limit probability prediction information containing the probability evaluation result of each key object appearing out-of-limit in the future;
[0011] Combine the severity index of the power system and the out-of-limit probability prediction information to obtain a comprehensive risk index of the power system;
[0012] According to the comprehensive risk index and a preset threshold, it is judged whether to trigger a regulation program of the power system; if the regulation program is triggered, a dispatch control strategy is generated by using the active dispatch control model to realize active regulation of the power system.
[0013] Preferably, the time series prediction model realizes time series prediction by using a long short-term memory network, which is represented as:
[0014] x t+α =M s (X t )
[0015] In the formula, M s represents the long short-term memory network, X t represents the multi-dimensional measurement time series data, x t+α represents the prediction data vector, and a represents the prediction time in advance.
[0016] Preferably, the out-of-limit probability prediction model is constructed by using a support vector machine classification model and a confidence correction module.
[0017] Preferably, for each key object, one out-of-limit probability prediction model is trained; the probability that the zth key object appears out of limit in the future is represented as:
[0018]
[0019] In the formula, represents the zth out-of-limit probability prediction model, x t+α represents the prediction data vector, a represents the prediction time in advance, and p z represents the probability that the zth key object appears out of limit at the prediction time.
[0020] Preferably, the severity index is constructed by using a severity function of the power system.
[0021] The comprehensive risk index is represented as:
[0022]
[0023] In the formula, a represents the prediction time in advance, Risk t+α represents the comprehensive risk index at the prediction time, S z represents the severity index corresponding to the zth key object, and p z represents the probability that the zth key object appears out of limit at the prediction time.
[0024] Preferably, the calculation method of the severity index corresponding to the voltage limit overrun, line overload and transformer limit overrun is as follows:
[0025]
[0026]
[0027]
[0028] wherein, respectively represent the severity index corresponding to the voltage limit overrun, line overload and transformer limit overrun at the prediction time, S z According to the above calculation method, one of the above calculation methods is selected according to the category to calculate; respectively represent the node voltage, line power flow active power and transformer current at the prediction time; respectively represent the upper limit value and the lower limit value of the voltage; is the voltage limit value, is the power flow limit value, is the transformer current limit value.
[0029] Preferably, the active dispatching control model takes the minimum total scheduling cost as the target, which is represented as:
[0030]
[0031] wherein, F represents the total scheduling cost, h represents the scheduling period index, t represents the current time, H is the scheduling period length, G is the set of traditional generators, is the output of the traditional generator group at h time, is the traditional unit scheduling cost, ε is the unit load adjustment cost factor, U is the set of system nodes, is the active load adjustment amount, B is the set of battery energy storage, is the battery energy storage power, is the battery energy storage scheduling cost.
[0032] Preferably, the safety risk constraint based on data driving is represented as:
[0033]
[0034] wherein, Risk h represents the comprehensive risk index at any time within the scheduling period, R lim represents a preset threshold, h represents the scheduling period index, t represents the current time, and H is the scheduling period length.
[0035] Preferably, the active dispatch control model further comprises the following constraints: power balance constraint, generator output constraint, generator ramping output constraint, line transmission active power constraint, battery energy storage charge and discharge state limit, battery energy storage charge and discharge power limit, and battery energy storage state of charge constraint.
[0036] In another aspect, the application provides a power system regulation system integrating data-driven risk perception, comprising a memory, a processor, and a computer program stored in the memory and capable of running on the processor, wherein the processor implements the steps in the above-mentioned power system regulation method integrating data-driven risk perception when executing the computer program.
[0037] The one or more technical solutions provided in the application have at least the following technical effects or advantages:
[0038] The application firstly collects multi-dimensional measurement time series data of the power system based on a key variable feature subset of the power system, inputs the multi-dimensional measurement time series data into a constructed time series prediction model, and uses the time series prediction model to output a prediction data vector corresponding to the key variable features (i.e. to obtain future scenario prediction data about the key objects); then the prediction data vector is input into a constructed out-of-limit probability prediction model, and the out-of-limit probability prediction model outputs out-of-limit probability prediction information containing probability evaluation results of each key object appearing out of limit in the future (i.e. the probability of the key objects in the system possibly appearing out of limit in the future); then the severity index of the power system and the out-of-limit probability prediction information are combined to obtain a comprehensive risk index of the power system; finally, according to the comprehensive risk index and a preset threshold, it is judged whether to trigger the regulation program of the power system; if the regulation program is triggered, a dispatch control strategy is generated by using a constructed active dispatch control model to realize active regulation of the power system (i.e. source-load-storage joint dispatch control). The active dispatch control model contains a data-driven safety risk constraint. The application can predict the risk scenarios that may occur in the future by using historical data and develop active regulation strategies in advance to prevent or mitigate these risks; the out-of-limit probability prediction model used in the application can output the out-of-limit probability, compared with the model that only provides a classification label, the application can provide more specific and valuable information for power grid dispatch and operation personnel, and enhance the accuracy and efficiency of decision-making; the application adds a data-driven safety risk constraint in the active dispatch control model, so that the generated dispatch control strategy strictly meets the system safety requirements; the data-driven method is adopted, which eliminates the limitations of traditional parameter models and ensures the safety risk monitoring robustness under various system operating states and physical parameter changes; the application can provide forward-looking warning information for dispatch and operation personnel to help them adjust the system in advance, and enhance the operation safety and risk active response ability of the power grid. In summary, the application realizes the probability perception evaluation of the power system risk, and effectively combines with the source-load-storage dispatch control of the power system, which can greatly improve the active resistance ability of the safety risk of the power system. BRIEF DESCRIPTION OF DRAWINGS
[0039] Figure 1 is a flowchart of a power system regulation method provided by an embodiment of the application;
[0040] Figure 2 is a grid structure diagram of a provincial example system in an example of the application;
[0041] Figure 3 is a key variable subset used in the example of the application;
[0042] Figure 4 is a training error iteration curve corresponding to the best time series prediction model in the example of the application;
[0043] Figure 5 is the power prediction result of line 13-14 in the example of the present application;
[0044] Figure 6 is the power prediction result of line 17-18 in the example of the present application;
[0045] Figure 7 is the power prediction result of line 14-17 in the example of the present application;
[0046] Figure 8 is the probability prediction of the southeast region heavy load scenario of the provincial grid system in the example of the present application;
[0047] Figure 9 is the comparison chart before and after the comprehensive risk control of the system heavy load in the example of the present application;
[0048] Figure 10 is the action situation of the direct regulation power plant of the system in the example of the present application;
[0049] Figure 11 is the action situation of the pumped storage power plant at node 31 in the example of the present application;
[0050] Figure 12 is the system power flow heat map without active regulation in the example of the present application;
[0051] Figure 13 is the system power flow heat map with active regulation in the example of the present application. DETAILED DESCRIPTION
[0052] In order to better understand the above technical solutions, the above technical solutions will be described in detail below in combination with the drawings of the specification and specific embodiments.
[0053] Embodiment 1:
[0054] Embodiment 1 provides a method for regulating a power system fused with data-driven risk perception, comprising:
[0055] A time series prediction model, an out-of-limit probability prediction model and an active dispatching control model are constructed, and the active dispatching control model contains a data-driven safety risk constraint;
[0056] Based on a key variable feature subset of the power system, multi-dimensional measurement time series data of the power system is collected; the multi-dimensional measurement time series data is input into the time series prediction model, and the time series prediction model outputs a prediction data vector corresponding to the key variable features;
[0057] The prediction data vector is input into the out-of-limit probability prediction model, and the out-of-limit probability prediction model outputs out-of-limit probability prediction information containing probability evaluation results of each key object appearing out of limit in the future;
[0058] obtain a comprehensive risk index of the power system in combination with the severity index of the power system and the out-of-limit probability prediction information;
[0059] determine whether to trigger a regulation program of the power system according to the comprehensive risk index and a preset threshold value; if the regulation program is triggered, generate a dispatch control strategy by using the active dispatch control model to realize active regulation of the power system.
[0060] The time series prediction model realizes time series prediction by using a long short-term memory network, and is represented as:
[0061] x t+α =M s (X t )
[0062] In the formula, M s represents the long short-term memory network, X t represents the multi-dimensional measurement time series data, x t+α represents the predicted data vector, and a represents the prediction time in advance.
[0063] The out-of-limit probability prediction model is constructed by using a support vector machine classification model and a confidence correction module. Specifically, for each key object, one out-of-limit probability prediction model is trained; the probability that the zth key object appears out of limit in the future is represented as: In the formula, P (z, a) represents the zth out-of-limit probability prediction model, x t+α represents the predicted data vector, a represents the prediction time in advance, and P t+α (z, a) represents the probability that the zth key object appears out of limit in the future a time.
[0064] The severity index is constructed by using a severity function of the power system; and the comprehensive risk index is represented as: In the formula, a represents the prediction time in advance, Risk t+α represents the comprehensive risk index at the prediction moment, S z (z) represents the severity index corresponding to the zth key object, and P t+α (z, a) represents the probability that the zth key object appears out of limit at the prediction moment.
[0065] The active dispatch control model takes the minimum total dispatch cost as the target, and is represented as:
[0066]
[0067] In the formula, F represents the total dispatch cost, h represents the dispatch period index, t represents the current moment, H is the length of the dispatch period, G is the set of traditional generators, is the output of the traditional generator set at time h, is the traditional unit dispatching cost, ε is the unit load adjustment cost factor, U is the system node set, is the load active adjustment amount, B is the battery energy storage set, is the battery storage power, Dispatch costs for battery energy storage.
[0068] The data-driven security risk constraint is expressed as: In the formula, Risk h Represents the comprehensive risk index at any moment in the scheduling cycle, R lim represents the preset threshold, h represents the scheduling period index, t represents the current time, and H represents the scheduling period length.
[0069] In addition, the active scheduling control model also includes the following constraints: power balance constraint, generator output constraint, generator climbing output constraint, line transmission active power constraint, battery energy storage charge and discharge state constraint, battery energy storage charge and discharge power constraint, and battery energy storage charge state constraint.
[0070] The present invention will be further described below.
[0071] The present invention provides a data-driven risk-aware power system control method, see Figure 1 , including the following steps:
[0072] Step 1: Based on a subset of key variable features of the power system, multidimensional measurement time series data of the power system is collected; the multidimensional measurement time series data is input into a constructed time series prediction model, and the time series prediction model outputs a prediction data vector corresponding to the key variable features.
[0073] Specifically, based on the key variable feature subset of typical power system control scenarios (where F i represents the i-th key variable feature, D is the total number of key variable features), collects the latest power system multidimensional measurement time series data X t , t represents the current time, L represents the length of the time window for time series data collection, Represents the measured value of the i-th key variable feature at time t.
[0074] The key variable feature subset for typical power system control scenarios includes the key variable features that can be measured in the system and are most relevant to the target warning scenario. The key variable feature subset can be obtained based on correlation evaluation methods such as the maximum information coefficient method, mutual information method, variance analysis, etc., or based on special feature selection methods such as the wrapper feature selection method and the embedded feature selection method.
[0075] The obtained measurement data (i.e. multi-dimensional measurement time series data) is input into the constructed time series prediction model to obtain a prediction data vector of the characteristics of the key variables in the future, i.e. x t+α , where α represents the time of prediction in advance, and the time series prediction is completed by the long short-term memory network M s , i.e. x t+α =M s (X t ).
[0076]
[0077] In the above formula, X t is the multi-dimensional measurement time series data in the form of a matrix with a size of D×L; and x t+α is the prediction data vector in the form of a vector with a size of D×1.
[0078] Step 2: inputting the prediction data vector into the constructed out-of-limit probability prediction model, and the out-of-limit probability prediction model outputs out-of-limit probability prediction information containing probability evaluation results of the future out-of-limit of each key object.
[0079] That is, the prediction data vector x t+α is input into the out-of-limit probability prediction model for the system key object to obtain the probability evaluation results of the future out-of-limit of each key object.
[0080] The system key objects mainly include key lines, key voltage nodes and key transformers (the selection of the key objects can be determined according to whether the out-of-limit has occurred in the history), and the out-of-limit probability prediction of each key object is completed by a support vector machine classification model superimposed with a confidence correction module, and the out-of-limit probability prediction model is represented as follows: , where z represents the index of the key object.
[0081] In a specific application, for each key object, an out-of-limit probability prediction model (a support vector machine classification model superimposed with a confidence correction module) needs to be trained to obtain more comprehensive system early warning information.
[0082] Therefore, the probability of the zth key object appearing out-of-limit in the future can be represented as follows:
[0083] Specifically,
[0084] In the formula, σ(.) is an activation function, and in the present application, a sigmoid function is taken, i.e.
[0085]
[0086] In the formula, logits, also known as decision scores; a z and b z is a correction parameter of the confidence correction module, denotes a support vector machine classification model.
[0087] The support vector machine classification model outputs a decision score about whether the key object is out-of-limit based on a prediction data vector, when it indicates that the model predicts that the key object is about to be out-of-limit. After passing through the confidence correction module, the output out-of-limit probability is between 0 and 1, i.e. The larger the value is, the greater the out-of-limit probability of the key object is.
[0088] The confidence correction parameter a z and b z is obtained by minimizing the negative log-likelihood on a reserved correction data set. For example, there is a correction data set This data set is independent of the training data set, x k denotes the kth data sample, is the out-of-limit label of the corresponding zth key object, denotes out-of-limit, denotes not out-of-limit, N c denotes the number of correction samples. The sample data is input into the trained support vector machine classification model The corresponding decision score set, i.e. is obtained based on the decision score and the label set According to minimizing the negative log-likelihood, the parameters a z and b z are obtained.
[0089] The above optimization model belongs to an unconstrained optimization model, which can be solved by using the Scipy extension package of Python, and the solving algorithm can select the L-BFGS (Limited-memory Broyden-Fletcher-Goldfarb-Shanno) method.
[0090] Step 3, combining the severity index of the power system and the out-of-limit probability prediction information, to obtain a comprehensive risk index of the power system.
[0091] That is, based on the probability that the key object will be out-of-limit in the future (i.e. out-of-limit probability prediction information) The system comprehensive risk index (i.e. the severity index of the power system) is constructed by using the power system severity function, to form the comprehensive risk index of the power system.
[0092] The severity function is often used to quantify the security or stability of the power system under a certain state or fault, which provides a numerical measure of the potential problems or risks of the power grid, thus helping the operation and control decisions.
[0093] Specifically, the calculation methods of the severity indices of voltage out-of-limit, line overload and transformer out-of-limit are as follows:
[0094]
[0095]
[0096]
[0097] wherein, respectively represent the severity indices corresponding to the voltage out-of-limit, line overload and transformer out-of-limit at the prediction time, and the zth key object belongs to which category takes the corresponding calculation method; respectively are the node voltage, line power flow active power and transformer current at the prediction time; is the corresponding voltage limit value, which can take the corresponding upper limit value or lower limit value according to the specific situation, respectively are the upper limit value and the lower limit value of the voltage; is the power flow limit value of the corresponding object; is the transformer current limit value.
[0098] Based on the out-of-limit probability and the severity index of each key object, the overall risk of the system can be calculated according to the following formula, i.e. wherein S z is the severity index of the zth key object, which is taken as or is the corresponding out-of-limit probability.
[0099] Through the quantitative system security operation risk comprehensive index, the operator and the decision maker can better understand the current state of the system, so as to make more intelligent and timely decisions. At the same time, the quantitative risk index based on the data-driven method also lays a foundation for the active dispatching control of the power system.
[0100] Step 4, according to the comprehensive risk index and a preset threshold value, judging whether to trigger the regulation program of the power system; if the regulation program is triggered, generating a dispatching control strategy by using the active dispatching control model to realize the active regulation of the power system.
[0101] that is, the calculated system online security risk early warning index Risk t+α is compared with a preset threshold value R limIf the comparison exceeds the allowed range, the available capacity information of each adjustable resource in the system is immediately collected, and the constructed dispatching control model is used to trigger the source-load-storage joint dispatching control.
[0102] The preset threshold value can be set according to historical dispatching records, and the out-of-limit condition of the corresponding key object is found according to the historical out-of-limit control records, and the severity at this time is calculated. In the jth historical out-of-limit event (the out-of-limit event triggering the dispatching control is not considered), the system comprehensive severity is S j , which is obtained by adding the severity of all out-of-limit objects in the system. The threshold value can be set as follows: R
[0103] When the active dispatching control model (i.e., the source-load-storage joint active dispatching control model) is constructed, a data-driven safety risk constraint is added to the model, so that the generated dispatching control strategy strictly meets the system safety requirements.
[0104] The active dispatching control model aims to minimize the total dispatching cost to obtain the optimal dispatching control strategy in multiple time sections, mainly considering the system power balance constraint, the upper and lower limits of unit output, the unit output ramping constraint, the line power flow constraint, and the overall system safety risk constraint.
[0105] The constructed active dispatching control model is specifically shown as follows:
[0106]
[0107] In the formula, F represents the total dispatching cost, h represents the dispatching period index, t represents the current time, H represents the dispatching period length, and G represents the set of traditional generators. is the output of the traditional generator at time h, is the dispatching cost of the traditional unit, which is calculated according to a quadratic cost function; ε is a unit load adjustment cost factor, U is the set of system nodes, is the active load adjustment amount, B is the set of battery energy storage, is the battery energy storage power, greater than 0 indicates discharging, and less than 0 indicates charging, is the dispatching cost of the battery energy storage.
[0108] wherein, is calculated using the following formula:
[0109]
[0110] In the formula, a i , b i , and c i are the corresponding fuel cost coefficients.
[0111] wherein, The calculation is obtained by using the following formula:
[0112]
[0113] In the formula, I si.ch (h) and I si.dis (h) are the state of charge variable and the state of discharge variable, respectively, which control the state of charge and discharge of the battery at time h; is the modulus of the charging and discharging power, C ch.i (h) is the state of charge cost function at time h, C dis.i (h) is the state of discharge cost function at time h.
[0114] Specifically, C ch.i (h) and C dis.i (h) are calculated as follows:
[0115]
[0116]
[0117]
[0118]
[0119] In the formula, C init is the investment cost of the battery, k ch , k dis are the charging and discharging impact factors, respectively, SOC i (h) is the state of charge at time h, SOC imax is the maximum state of charge, N SB (x i (h)) represents the maximum number of cycles corresponding to the battery, x i (h) is the charging and discharging depth at time h.
[0120] As can be seen from the above formula, the battery energy storage operation loss cost is jointly affected by the initial and final states of charge SOC i (h-1) and SOC i (h), the charging and discharging power and the charging and discharging impact factors.
[0121] The constraints of the active dispatching control model constructed mainly include the following:
[0122] 1) Power balance.
[0123]
[0124] In the formula, P (h) is the output of traditional generator at h, P (h) is the battery storage power, P (h) is the predicted value of node load at h, P (h) is the line active loss, P (h) is the active load adjustment, M is the set of transmission lines.
[0125] 2) Generator output constraint.
[0126]
[0127] In the formula, P (h) is the upper limit of generator active output, P (h) is the lower limit of generator active output.
[0128] 3) Generator climbing output constraint.
[0129]
[0130] In the formula, ΔP i,max,down P (h) is the maximum down-climbing rate of generator active output, ΔP i,max,up P (h) is the maximum up-climbing rate of generator active output.
[0131] 4) Line transmission active constraint.
[0132]
[0133] In the formula, P (h) is the active transmission limit of transmission line, P (h) is the active power transmitted by transmission line at h.
[0134] The constraint conditions of battery storage include charge and discharge state limit, discharge power limit, charge power limit, state of charge constraint, etc., which are described below.
[0135] 5) Battery storage charge and discharge state limit.
[0136]
[0137] In the formula, I si.ch (h) and I si.dis (h) can only be equal to 1 at most, indicating that the battery can only be in one of the three states of charging, discharging, and neither charging nor discharging. I si.ch (h) ∈ {0, 1}, I si.dis (h) ∈ {0, 1}.
[0138] 6) Battery storage charge and discharge power limit.
[0139]
[0140]
[0141] wherein, and are the lower and upper limits of discharge power, respectively, and are the lower and upper limits of charge power, respectively.
[0142] 7) Battery energy storage state of charge constraint.
[0143]
[0144]
[0145] wherein, δ is the charge-discharge efficiency, S iAh is the ampere capacity of the battery, SOC iR is the state of charge safety margin.
[0146] 8) Data-driven system safety risk constraint.
[0147]
[0148] wherein, Risk h < R lim represents that at any time in the scheduling period, the data-driven safety risk constraint needs to meet the requirements.
[0149] Solving the above active scheduling control model can obtain the active scheduling control strategy for potential safety risks, and complete the active regulation of the power system that fuses data-driven risk perception.
[0150] Embodiment 2:
[0151] Embodiment 2 provides a regulation system of a power system that fuses data-driven risk perception, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps in the regulation method of the power system that fuses data-driven risk perception as described in Embodiment 1 when executing the computer program.
[0152] Since the functions of the system provided in Embodiment 2 correspond to the steps in the method provided in Embodiment 1, they can be understood by referring to the description of Embodiment 1, which will not be repeated here.
[0153] The following gives an example to verify and illustrate the present application.
[0154] This example is verified based on the actual network frame of a provincial grid system, and the system network frame is as follows: Figure 2This example mainly takes the overload scenario in the southeast region of the provincial network as the risk event. The key variable characteristics based on risk perception are as follows: Figure 3 As shown. Based on the key variable feature subset and sample database, the long short-term memory network is used to perform time series rolling forecasts on the key variable features. The grid search and cross-validation method is used to find the best model and network parameters, and the complete model is rebuilt based on the found optimal parameters to perform rolling forecasts on the key variable features included in the system. The training error iteration curve corresponding to the best model found in this example is shown in Figure 4 shown.
[0155] from Figure 4 It can be seen that the long short-term memory network time series rolling prediction model constructed in this example can reduce the error to a very small range as the number of iterations increases, and the training error and validation set error are within a reasonable distribution range. The model has good generalization ability, thus laying a good foundation for accurate probability prediction. The time series prediction effect of relevant key variable features is as follows Figure 5 、 Figure 6 、 Figure 7 shown.
[0156] It can be seen from the rolling prediction graph that the model constructed by the present invention effectively realizes the accurate prediction of key variable characteristics. The same data samples are input into the off-limit probability prediction model of the key objects of the system, and the off-limit label prediction and confidence correction of the support vector machine classification model are performed and then converted into off-limit probability output. Figure 8 The effectiveness of the scenario probability prediction model during an overload risk event in the southeastern part of the provincial grid system is demonstrated. The simulation results show that the model constructed by this invention effectively predicts cross-sectional overload events in the system. When an overload event actually occurs in the system, the predicted probability also shows an upward trend and remains at a high value. Furthermore, when the 0-1 label prediction fails, the probability curve can still effectively reflect the cross-sectional overload trend of the system and accurately reflect the risks faced by the system. This shows that probabilistic prediction can provide a better reference value for system regulation.
[0157] Based on the probabilistic early warning model and safety risk assessment model mentioned above, when the system comprehensive risk exceeds the set threshold, early joint regulation is initiated to integrate the controllable resources in the network and achieve optimal allocation. The historical overload risk events in the southeast region of the provincial network system are selected as the regulation object, and the system overload comprehensive risk assessment results are used as the input trigger instruction to guide the source-load-storage joint regulation model to perform early regulation. Figure 9 As shown in , when the system overload comprehensive risk exceeds a certain threshold, the scheduling control is started. Figure 9It can be seen that if the source load storage active regulation is not performed, the system overload comprehensive risk will continue to rise. Considering the role of the power system active regulation model integrating data-driven risk perception, joint regulation is started when the system overload comprehensive risk exceeds the threshold.
[0158] After starting the joint regulation program, considering the safety risk constraints of the system and the characteristics of various resources, a standardized cost rating system is constructed to regulate and control with the minimum regulation cost as the goal. In this risk prevention and control event, after the triggering instruction of the risk perception program, joint scheduling control is started to adjust and allocate resources for various resources in the system. The adjustment results are shown in Figure 10 、 Figure 11 .
[0159] As can be seen from Figure 10 , after the system issues the regulation instruction, the direct regulation power plant starts to act. Due to the existence of demand side response adjustment, the output of the direct regulation power plant that originally tracks the system power rise slows down (power plant 1), and the output of other direct regulation power plants decreases to a certain extent to track the decrease of load power, so that the transmission power is effectively reduced. Similarly, as shown in Figure 11 , the pumped storage power plant at the provincial grid node 31 also tracks the change of system load, and the risk perception regulation of the advance type avoids the worse impact of the continuous rise of system load.
[0160] As shown in Figure 9 , after the system adjustment, the southeast region of the provincial grid system is significantly reduced, which further verifies the effectiveness of the active regulation method of the power system integrating data-driven risk perception proposed in the present application. In order to further compare the control effect, the power flow heat map of the system without active scheduling control is drawn, as shown in Figure 12 、 Figure 13 , wherein Figure 12 is the power flow heat map of the system without active regulation, Figure 13 is the power flow heat map of the system with active regulation.
[0161] Comparing Figure 12 and Figure 13As can be seen (the darker the line color, the higher the transmission power percentage), when no early preventive control is performed, the southeast area is heavily loaded, and the system line is in a state of obvious heavy load. At this time, if no control is performed, further more serious faults may occur. When the power system active regulation strategy fusing data-driven risk perception is used, the heavy load in the southeast area is effectively eliminated in advance, and the color of the system line is obviously lighter. It can be seen that the power system active regulation strategy fusing data-driven risk perception can effectively perceive and regulate the system risk in advance, effectively saves the regulation time and regulation resources, and effectively eliminates the system heavy load risk.
[0162] Finally, it should be explained that the above specific embodiments are only used to illustrate the technical solutions of the present application and are not limiting. Although the present application is described in detail with reference to examples, those skilled in the art should understand that the technical solutions of the present application can be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present application, and they should be covered in the scope of the claims of the present application.
Claims
1. A method for controlling a power system by integrating data-driven risk perception, characterized in that: include: Constructing a time series prediction model, an over-limit probability prediction model, and an active scheduling control model. The active scheduling control model includes data-driven safety risk constraints and battery energy storage constraints. The battery energy storage constraints include battery energy storage charge and discharge state restrictions, battery energy storage charge and discharge power restrictions, and battery energy storage state of charge constraints. Based on a subset of key variable characteristics of the power system, multidimensional measurement time series data of the power system is collected; the multidimensional measurement time series data is input into the time series prediction model, and the time series prediction model outputs a prediction data vector corresponding to the key variable characteristics; Inputting the predicted data vector into the limit crossing probability prediction model, the limit crossing probability prediction model outputting limit crossing probability prediction information including a probability assessment result of each key object exceeding the limit in the future; Combining the severity index of the power system and the limit-crossing probability prediction information to obtain a comprehensive risk index of the power system; Determining whether to trigger a power system control procedure based on the comprehensive risk index and a preset threshold; If the control program is triggered, the active dispatch control model is used to generate a dispatch control strategy to achieve active control of the power system; The active scheduling control model aims to minimize the total scheduling cost, which can be expressed as: Where, F represents the total scheduling cost, h Indicates the scheduling period index, t Indicates the current moment, H is the scheduling cycle length, G For traditional generator sets, for h The output of traditional generator sets at all times, is the traditional unit dispatching cost, is the unit load adjustment cost factor, U is the set of system nodes, is the load active adjustment amount, B For battery energy storage collection, is the battery storage power, Dispatch costs for battery energy storage.
2. The method for controlling a power system based on fusion data driven risk perception according to claim 1, characterized in that: The time series prediction model uses the long short-term memory network to achieve time series prediction, which is expressed as: Where, represents the long short-term memory network, Represents multidimensional measurement time series data, represents the predicted data vector, Indicates the time ahead of the forecast.
3. The method for controlling a power system based on fusion data driven risk perception according to claim 1, characterized in that: The limit crossing probability prediction model is constructed by superimposing a support vector machine classification model on a confidence correction module.
4. The method for controlling a power system based on fusion data driven risk perception according to claim 3, characterized in that: For each key object, train a limit-crossing probability prediction model; z The probability of a key object exceeding the limit in the future is expressed as: Where, Indicates the z A limit crossing probability prediction model, represents the predicted data vector, Indicates the time in advance of the forecast, Indicates the z Key objects in the future The probability of exceeding the limit occurs.
5. The method for controlling a power system based on fusion data driven risk perception according to claim 1, characterized in that: constructing the severity index using a severity function of the power system; The comprehensive risk index is expressed as: Where, Indicates the time in advance of the forecast, represents the comprehensive risk index at the time of prediction, Indicates the z The severity index corresponding to each key object, Indicates the z The probability of a key object exceeding the limit at the prediction moment.
6. The method for controlling a power system based on fusion data driven risk perception according to claim 5, characterized in that: The calculation method for the severity index corresponding to voltage over-limit, line overload, and transformer over-limit is as follows: Where, 、 、 They represent the severity indexes corresponding to voltage over-limit, line overload, and transformer over-limit at the prediction moment, Select one of the above calculation methods according to the category; 、 、 They are the node voltage, line power flow active power, and transformer current at the prediction moment respectively; 、 are the upper and lower limits of voltage respectively; is the voltage limit, is the tidal limit, is the transformer current limit.
7. The method for controlling a power system based on fusion data driven risk perception according to claim 1, characterized in that: The data-driven security risk constraint is expressed as: Where, Represents the comprehensive risk index at any moment in the scheduling cycle, Indicates the preset threshold, h Indicates the scheduling period index, t Indicates the current moment, H is the scheduling cycle length.
8. The method for controlling a power system based on fusion data driven risk perception according to claim 1, characterized in that: The active dispatch control model also includes the following constraints: power balance constraint, generator output constraint, generator ramp output constraint and line transmission active power constraint.
9. A power system control system integrating data-driven risk perception, characterized in that: include: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the steps in the method for controlling a power system driven by risk perception using fused data are implemented as described in any one of claims 1 to 8.