Energy storage power station control system based on rainfall data of watershed meteorological station

Abstract

The invention, which relates to the technical field of energy storage power station control, discloses an energy storage power station control system based on rainfall data of a watershed meteorological station, comprising a meteorological data acquisition unit, a data preprocessing module, an intelligent analysis engine module, a power station operation decision module, an energy storage adjustment module and an emergency response module. Comprising a data cleaning sub-module which is used for removing noise, abnormal values and error data in collected data and ensuring the accuracy and reliability of the data; the data calibration sub-module is used for further improving the precision of the data through comparison and calibration with standard meteorological data; and the data conversion sub-module is used for converting the acquired data into a format which can be identified and processed by the system. The technical scheme provided by the invention has remarkable beneficial effects in multiple aspects of improving the operation efficiency and safety of the power station, improving the energy storage and adjustment capability, guaranteeing the power supply stability and the like, and also has positive significance in promoting sustainable development and reducing the operation cost.

Description

Technical Field

[0001] The present invention relates to the technical field of energy storage power station control, and particularly to an energy storage power station control system based on precipitation data of basin meteorological stations. Background Art

[0002] With the continuous growth of energy demand and the increasing emphasis on clean energy, energy storage power stations, as important energy storage and regulation facilities, play an increasingly crucial role in the power system. And the precipitation data of basin meteorological stations is crucial for the operation and management of energy storage power stations.

[0003] In practical applications, accurately obtaining and analyzing precipitation data faces many challenges. On the one hand, precipitation has a high degree of randomness and complexity, and its patterns and laws are difficult to accurately grasp. On the other hand, the safe and stable operation of the power station needs to comprehensively consider various factors, such as fluctuations in power demand, limitations of reservoir capacity, etc. Traditional power station control methods often have difficulty coping with complex and changing precipitation conditions, which may lead to problems such as low energy utilization efficiency and high operation risks.

[0004] To solve these problems, it is necessary to develop an advanced energy storage power station control system based on precipitation data of basin meteorological stations. This system uses advanced meteorological data acquisition technology to ensure the real-time and accuracy of data. Through the data preprocessing module, noise and abnormal data are effectively removed to improve data quality. Machine learning algorithms and pattern recognition technologies in the intelligent analysis engine module can deeply mine the patterns and laws in precipitation data, and the trend prediction module can relatively accurately predict future precipitation trends, providing strong support for power station operation decisions. The power station operation decision module can generate scientific and reasonable operation strategies by comprehensively considering various factors, the energy storage regulation module realizes precise control of reservoir energy storage, and the emergency response module ensures the safety of the power station in extreme weather conditions. Through such a comprehensive technical solution, the operation efficiency and safety of the energy storage power station can be improved, and it can better adapt to complex meteorological environments and changes in power demand. Summary of the Invention

[0005] An energy storage power station control system based on precipitation data of basin meteorological stations proposed by the present invention is to solve the problems mentioned in the above prior art.

[0006] To achieve the above object, the present invention adopts the following technical solutions: An energy storage power station control system based on precipitation data of basin meteorological stations includes the following modules:

[0007] Meteorological data acquisition unit: It includes a data acquisition terminal connected to the meteorological stations in the basin, which can obtain various parameters of precipitation in real time and accurately;

[0008] Data preprocessing module: It includes a data cleaning sub-module for removing noise, outliers, and error data from the collected data; a data calibration sub-module for improving the data accuracy by comparing and calibrating with standard meteorological data; a data conversion sub-module for converting the collected data into a format recognizable by the system;

[0009] Intelligent analysis engine module: A machine learning algorithm library containing algorithms applicable to precipitation data analysis; a trend prediction module for modeling and training using historical and real-time data to predict the precipitation trend in the future for a period of time;

[0010] Power station operation decision module: It includes a strategy generation sub-module for generating power station operation strategies based on the analysis results of the intelligent analysis engine and the various operation parameters of the power station; a risk assessment sub-module for assessing and warning the risks brought by various decisions and optimizing the formulated strategies;

[0011] Energy storage regulation module: It includes an electric valve control system for controlling the inlet and outlet valves of the reservoir to achieve the operations of water storage and release; a water level monitoring system for real-time monitoring of the water level changes in the reservoir;

[0012] Emergency response module: It includes a rainstorm monitoring and warning sub-module for immediately issuing an alarm when a rainstorm or abnormal precipitation is detected; an emergency avoidance strategy formulation sub-module for formulating measures to deal with emergencies such as rainstorms and real-time monitoring of the implementation effects.

[0013] Furthermore, the training steps of the time series analysis algorithm are as follows:

[0014] Set up a time axis with equally spaced time marks t at 1-hour intervals. Assume that there are i rain gauges in the basin controlled by the reservoir. Organize the historical precipitation data of each rain gauge obtained by the data preprocessing module into a two-dimensional data set P(i,t), and organize the historical incoming water data of the reservoir into a one-dimensional data set R(i,t). Here, i represents the number of the weather station, and t represents the timestamp when the data is recorded. The Unixtimestamp standard is selected as the timestamp standard;

[0015] If there is no match, fill in the data before and after the required time mark obtained by using the linear interpolation algorithm. Specifically, for the rain gauge, there is

[0016]

[0017] where Pi(t’) is the precipitation data of the rain gauge numbered i before the required time mark, Pi(t”) is the precipitation data of the rain gauge numbered i after the required time mark, t’ is the timestamp before the required time mark, t” is the timestamp after the required time mark, and t is the required time mark;

[0018] For a reservoir, there is

[0019]

[0020] where R(t’) is the inflow data of the reservoir before the required time stamp, R(t”) is the inflow data of the reservoir after the required time stamp, t’ is the time stamp before the required time stamp, t” is the time stamp after the required time stamp, and t is the required time stamp;

[0021] Each time stamp t and the corresponding data set P(i,t) are respectively combined with R(t) to form a data set I(t) = [t, P(1,t), P(2,t), …, P(i,t), R(t)], and it is used as the training set and test set of the time series data prediction model;

[0022] According to the data collected during the rainfall period in the basin, the unit hydrograph base width T of the basin is obtained. Let T i be a time window containing i t time periods. The sum of the base width T and the time window T i containing i t time periods is the single rainfall concentration time with i t as the rainfall duration, which is TT i，

[0023] Build a time series data prediction model. The SSM layer of the time series data prediction model includes the following parameters: the state vector x t represents the internal state of the system at time t, and the output dimension d is set to 128; the state transition matrix A defines how the state vector changes over time; the control input matrix B defines the influence of the control input on the state; the observation matrix C defines how the state vector is mapped to the observation vector, W and U are bias matrices, and b is the weight. The mixing layer includes the following parameters: the weight and bias matrix b, W;

[0024] Before the model starts training, initialize the relevant data. The specific operations are as follows:

[0025] For the state vector x t , set it to 0; the weight matrix W of the input mapping function g adopts uniform distribution sampling:

[0026]

[0027] where i is the number of rain gauges. The state transition matrix A, the control input matrix B, the observation matrix C, and the other bias matrices W and U adopt Gaussian sampling:

[0028]

[0029] According to the rainfall data of the basin, different rainfall durations T of the basin are considered. i The model is divided into individual SSMs. i For a certain SSM, i the dataset I(t) that meets the number of window lengths of T i is used to calculate the control input vector u through the following formula: t :

[0030] u t = σ(W g I(t) + b g )

[0031] where W g is the weight matrix, with size d×i; bg is the bias vector, with size d; σ is the activation function, and the Sigmoid function is selected here.

[0032] Update the state vector x according to the state update equation: t+1 :

[0033] x t+1 = Ax t + Bu t

[0034] Calculate the observation vector y according to the observation equation: t :

[0035] y t = Cx t

[0036] Adjust the state vector through the gating mechanism:

[0037] f t = σ(W f x t + U f I(t) + b f )

[0038] i t = σ(W i x t + U i I(t) + b i )

[0039] o t = σ(W o x t + U o I(t) + b o )

[0040] c t = f t ⊙ c t-1 + i t ⊙ tanh(Wc x t +U c I(t)+b c )

[0041] h t =o t ⊙tanh(c t )

[0042] Map the state vector after selective propagation back to the output vector:

[0043] S t =W out h t +b out

[0044] where σ is the activation function, and the Sigmoid function is selected here; ⊙ represents element-wise multiplication, and W, U, and b represent the corresponding weight and bias matrices.

[0045] Fuse and process the output of the SSM layer to extract higher-level features. Let d′ = 256, and project S t into a higher-dimensional space d′:

[0046] Z t =W t ·S t +b t

[0047] where W is the weight matrix of size d′×d, and b is the bias vector of size 256.

[0048] Apply the ReLU activation function to the result of the linear transformation:

[0049] A t =ReLU(Z t )=max(0,Z t )

[0050] Add a residual connection. To make a residual connection, S t needs to be mapped to the same dimension as A t :

[0051]

[0052] After transformation, Perform a residual connection with A t :

[0053] H t =A t +W r ·S t +b r

[0054] Perform layer normalization on the output H of the mixing layer: t Specifically:

[0055]

[0056] where W is a weight matrix of size d'×d; b is a bias vector of size d'; μt and are the mean and variance of H respectively, ∈ is a small constant to prevent division by zero; take ∈ = 10 t , γ and β are learnable scaling and offset parameters; -6 Output the result:

[0057] Specifically:

[0058]

[0059] where W out is a weight matrix of size 1×d'; b out is a scalar bias,

[0060] Repeat (adjusting the state vector) to (outputting the result) for T - 1 times; i The obtained result uses the mean squared error to construct a loss function to evaluate the difference between the predicted sequence

[0061] and the actual sequence I(t): Specifically:

[0062]

[0063] where T i is the duration of all data. Calculate the gradients of each parameter through the loss function and perform backpropagation parameter updates. The gradient calculation method is as follows:

[0064] For the output layer, f Output represents the linear transformation of the fully - connected layer, ψ are the parameters of the output layer. The gradient of the loss L with respect to the output layer parameters ψ:

[0065]

[0066] For the mixing layer, let f Mixer represent the non - linear transformation of the mixing layer, φ are the parameters of the mixing layer. The gradient of the loss L with respect to the mixing layer parameters φ:

[0067]

[0068] For the SSM layer, let f SSM represent the non - linear transformation of the SSM layer, θ are the parameters of the SSM layer. The gradient of the loss L with respect to the SSM layer parameters θ:

[0069]

[0070] Preferably, the AdamW optimizer is used to update the parameters:

[0071]

[0072] where, β 1 and β 2 are the exponential decay rates of the momentum terms, with a value of 0.9; m t and v t are the moving averages of the first and second moments of the gradients, respectively; and are the estimates after bias correction; ∈ is a small constant used to prevent division by zero, with a value of 10 -8 ; η t is the learning rate; τ is the parameter to be updated in the model; the arrow represents the update operation;

[0073] Repeat steps (calculating the SSM layer) to (updating the parameters) until the loss function converges. At this time, the training of the SSM time series data prediction model is completed, denoted as SSM().

[0074] Furthermore, the specific method of the power station operation decision module is as follows:

[0075] For a certain time point t, perform a summation operation on the data of the weather station network. When , it is determined that it is raining in the basin. Among them, a i is the control area of rain gauge station i, and F is the total area of the basin;

[0076] Set the mathematical mark of the current reservoir water storage as V 0, After waiting for the rain to end, call the SSM time series data prediction model to predict the future incoming water volume. The specific method is as follows:

[0077] Use the precipitation records of all rain gauge stations before and after the rainfall duration T i period to construct the model input:

[0078] I input =[1 P(1,1)...P(i,1) R(1),...,T i P(1,T i )...P(i,T i )R(T i )]

[0079] Among them, 1 P(1,1)...P(i,1) R(1),...,T i P(1,T i )...P(i,T i)R(T i ) is the time mark from time period 1 to time period T, the precipitation data of each rain gauge station, and the runoff data at the reservoir inlet, which are synthesized into a two-dimensional data set. This two-dimensional data set will be sliced into a one-dimensional data set i I(t) = [t P(1,t)...P(i,t) R(t)]

[0080] is continuously added to SSM(), and I

[0081] is input into the SSM time series data prediction model. The last item R of the result is summed to predict the total incoming water data within T time periods: input V(T) = ∑SSM(I

[0082] ) input ) last +V 0

[0083] where SSM(I input ) last is the last item of the model output vector, and T is the duration when the incoming water grows to return to the initial incoming water. This duration is determined according to the output of the model;

[0084] At this time, the decision-making estimated pumping volume is:

[0085]

[0086] where V a is the decision-making estimated pumping volume, V s is the maximum allowable reservoir capacity, and V u (t) is the water consumption for power generation of the pumped storage power station in time period t.

[0087] Furthermore, it includes the following steps:

[0088] S1. The meteorological data acquisition unit continuously collects the precipitation data of the basin meteorological stations through its various terminals at the set time intervals and transmits the data to the data preprocessing module in real time through a reliable communication link;

[0089] S2. After receiving the data, the data preprocessing module first performs data cleaning by the data cleaning sub-module to remove incorrect data. Then, the data calibration sub-module calibrates the data. Finally, the data conversion sub-module converts the processed data into a format that can be processed by the intelligent analysis engine and transmits it to the intelligent analysis engine;

[0090] S3. The machine learning algorithm library in the intelligent analysis engine is called to deeply analyze the input precipitation data. The pattern recognition module identifies different precipitation patterns from the data, and the trend prediction module combines historical data and real-time data for modeling and training to predict the precipitation trend in the next period of time.

[0091] S4. Based on the analysis results of the intelligent analysis engine and the current status of the power station reservoir water level and power generation power, the power station operation decision module generates the optimal power station operation strategy through the strategy generation sub-module, and the risk assessment sub-module conducts a risk assessment on the generated strategy and adjusts the strategy according to new data or situations.

[0092] S5. According to the strategy formulated by the power station operation decision module, the energy storage regulation module precisely controls the inlet and outlet valves of the reservoir through the electric valve control system to achieve the operations of water storage and water release. The water level monitoring system real-time feedbacks the changes in the reservoir water level, and the energy conversion and storage equipment ensures the efficient conversion and storage of electric energy.

[0093] S6. The rainstorm monitoring and early warning sub-module in the emergency response module continuously monitors the meteorological data and real-time monitors the implementation effect.

[0094] Compared with the existing technologies, the beneficial effects of the present invention are as follows:

[0095] Through accurate precipitation data analysis and trend prediction, the present invention can arrange the water storage and power generation of the power station more reasonably, optimize resource utilization. The emergency response module can timely respond to abnormal situations such as rainstorms, reduce potential risks, and ensure the safe and stable operation of the power station. The precise control of the energy storage regulation module helps to better play the energy storage role of the power station and adapt to different power demand situations. The power station operation decision module that comprehensively considers various factors can generate more scientific and reasonable strategies, reduce blindness. The data preprocessing module ensures the accuracy and reliability of the input data, providing a solid foundation for subsequent analysis and decision-making. The pattern recognition module can identify different precipitation patterns, enabling the power station to better adapt to complex and changeable meteorological conditions, efficiently utilize precipitation resources and clean energy, and contribute to the sustainable development of energy. Reasonable operation strategies and efficient energy storage regulation can reduce the operation cost of the power station to a certain extent. Through accurate prediction and flexible decision-making, it helps to maintain a stable power supply.

[0096] In summary, the technical solution of the present invention has significant beneficial effects in improving the operation efficiency and safety of the power station, enhancing the energy storage and regulation ability, improving the scientific nature of decision-making, ensuring the stability of power supply, etc. At the same time, it also has a positive significance for promoting sustainable development and reducing operation costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0097] Figure 1Block diagram of a control system for a pumped - storage power station based on precipitation data from basin meteorological stations proposed by the present invention;

[0098] Figure 2 Hardware composition diagram of a precipitation data acquisition module for a pumped - storage power station based on basin meteorological stations proposed by the present invention. Specific implementation manners

[0099] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts fall within the scope of protection of the present invention.

[0100] In the description of the present invention, it should be understood that the orientation or positional relationships indicated by the terms "center", "longitudinal", "transverse", "length", "width", "thickness", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", "clockwise", "counterclockwise", etc. are based on the orientation or positional relationships shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and thus should not be construed as a limitation of the present invention.

[0101] In addition, the terms "first" and "second" are only used for descriptive purposes and cannot be understood as indicating or implying relative importance or implicitly indicating the quantity of the indicated technical features. Thus, the features defined with "first" and "second" may explicitly or implicitly include one or more of the described features. In the description of the present invention, "a plurality" means two or more, unless otherwise specifically defined. In addition, the terms "installation", "connection", and "connection" should be understood in a broad sense. For example, it may be a fixed connection, a detachable connection, or an integral connection; it may be a mechanical connection or an electrical connection; it may be directly connected or indirectly connected through an intermediate medium, and it may be the communication inside two elements. For those of ordinary skill in the art, the specific meanings of the above terms in the present invention can be understood according to specific situations. The present invention will be further described in detail below with reference to the accompanying drawings.

[0102] Refer to Figure 1-2 : A control system for a pumped - storage power station based on precipitation data from basin meteorological stations, including a meteorological data acquisition unit, a data pre - processing module, an intelligent analysis engine module, a power station operation decision - making module, an energy storage regulation module, and an emergency response module:

[0103] Meteorological data acquisition unit: It includes multiple data acquisition terminals connected to meteorological stations within the basin. These terminals are by no means ordinary devices. They are equipped with high-precision sensors, which are like sensitive senses and can obtain various parameters of precipitation in real time and accurately. Take rainfall for example. It can be accurate to the millimeter level. Whether it is a gentle drizzle or a heavy downpour, it can be accurately recorded. For rainfall intensity, it can carefully capture the rainfall changes every minute or even every second. As for the rainfall duration, it can accurately time from the moment the rainfall starts until it ends. Whether it is a short shower or a rainfall process lasting for several days, it can be completely recorded, providing a solid and accurate data basis for subsequent data analysis;

[0104] Data preprocessing module: It includes multiple powerful sub-modules. First is the data cleaning sub-module, which removes various "impurities" in the collected data, including noise data caused by equipment failures or external interferences. These noises are like "background noises" in the data and will interfere with subsequent analysis; there are also outliers, such as extremely abnormal data that may occur due to momentary sensor malfunctions and obvious error data. It will identify and remove them one by one to ensure the accuracy and reliability of the data.

[0105] Next is the data calibration sub-module. By comprehensively and carefully comparing and calibrating with standard meteorological data, it is like using a standard ruler to measure each data, not letting go of any slight deviation, further improving the data accuracy, making the data more accurate and reliable, and being able to truly reflect the actual meteorological situation.

[0106] Finally is the data conversion sub-module, which can skillfully convert various formats of collected data, whether they are original, complex, or formats that cannot be directly recognized and processed by the system, into formats that the system can recognize and process, enabling the data to flow smoothly in the system and be analyzed;

[0107] Intelligent analysis engine module: It includes various algorithms applicable to precipitation data analysis. For example, time series analysis can deeply analyze precipitation data in chronological order, exploring the changing trends and patterns of precipitation in different time periods. Whether it is the precipitation fluctuations within a day, or the precipitation changes over a month, a year, or even a longer time span, it can be clearly presented; clustering analysis can classify similar precipitation data, just like gathering precipitation events with the same characteristics together, thus discovering different types of precipitation patterns, such as being able to distinguish different precipitation categories like heavy rain type, shower type, continuous rain type, etc.

[0108] The pattern recognition module can accurately identify different precipitation patterns and regularities from a vast amount of precipitation data. Even the subtle patterns hidden behind complex data can be keenly captured by it, providing an important basis for meteorological research and prediction.

[0109] The trend prediction module uses historical data and real-time data for modeling and training. Through complex and accurate mathematical models and algorithms, it can accurately predict the precipitation trend in the future for a certain period. Whether it is the precipitation situation in the short-term future for several hours or days, or the precipitation trend in the medium- and long-term future for several weeks or months, it can give relatively reliable predictions, providing valuable reference information for many fields such as agricultural production, urban planning, flood control and drought relief, helping people take preventive measures in advance and reducing the losses caused by meteorological disasters.

[0110] If there is no match, the adjacent data before and after the required time mark obtained by demand are filled in by the linear interpolation algorithm. Specifically, for the rain gauge station, there is

[0111]

[0112] where Pi(t’) is the precipitation data of the rain gauge station numbered i before the required time mark obtained, Pi(t”) is the precipitation data of the rain gauge station numbered i after the required time mark obtained, t’ is the time before the required time mark obtained, t” is the time after the required time mark obtained, and t is the required time mark.

[0113] For the reservoir, there is

[0114]

[0115] where R(t’) is the incoming water data of the reservoir before the required time mark obtained, R(t”) is the incoming water data of the reservoir after the required time mark obtained, t’ is the time before the required time mark obtained, t” is the time after the required time mark obtained, and t is the required time mark.

[0116] The power station operation decision module includes a strategy generation sub-module that generates a detailed power station operation strategy, including the power generation power, the timing and rate of water storage and discharge, by considering factors such as power demand and reservoir capacity based on the analysis results of the intelligent analysis engine and the operating parameters of the power station; a risk assessment sub-module that assesses and warns of the risks that various decisions may bring to ensure the safety and stability of the power station operation; a dynamic adjustment sub-module that can timely adjust and optimize the formulated strategy according to the real-time situation and new data input.

[0117] The energy storage regulation module includes an electric valve control system that precisely controls the inlet and outlet valves of the reservoir to achieve the operations of water storage and release; a water level monitoring system that real-time monitors the water level changes of the reservoir to provide accurate data for energy storage regulation; and energy conversion and storage equipment that ensures the efficient conversion and storage of electric energy and its rapid release when needed.

[0118] The emergency response module includes a rainstorm monitoring and warning sub-module that immediately issues an alarm once a rainstorm or abnormal precipitation situation is detected; an emergency avoidance strategy formulation sub-module that quickly formulates specific measures to deal with emergencies such as rainstorms, including increasing the water release rate, starting the standby power supply, etc.; and an emergency execution and monitoring sub-module that is responsible for executing the emergency strategy and real-time monitoring the execution effect to ensure the safety of the power station.

[0119] In the present invention, the composition method of the intelligent analysis engine module is as follows:

[0120] Set up a timeline with equidistant time marks t at 1-hour intervals. Assume that there are i rain gauges in the basin controlled by the reservoir. Organize the precipitation data of each rain gauge in previous years obtained by the data preprocessing module into a two-dimensional data set P(i, t), and organize the incoming water data of the reservoir over the years into a one-dimensional data set R(i, t). Here, i represents the number of the weather station, and t represents the timestamp when the data is recorded. It should be noted that the time requirement of the data matches the marks on the timeline.

[0121] Select the Unixtimestamp standard as the timestamp standard.

[0122] If they do not match, use the adjacent data before and after the required time mark to fill in the blanks through the linear interpolation algorithm. Specifically, for the rain gauge, there is

[0123]

[0124] where Pi(t’) is the precipitation data of the rain gauge numbered i before the required time mark, Pi(t”) is the precipitation data of the rain gauge numbered i after the required time mark, t’ is the timestamp before the required time mark, t” is the timestamp after the required time mark, and t is the required time mark.

[0125] For the reservoir, there is

[0126]

[0127] where R(t’) is the incoming water data of the reservoir before the required time mark, R(t”) is the incoming water data of the reservoir after the required time mark, t’ is the timestamp before the required time mark, t” is the timestamp after the required time mark, and t is the required time mark.

[0128] For each time stamp \(t\) and the corresponding data sets \(P(i,t)\) and \(R(t)\), they are combined into the data set \(I(t)=[t, P(1,t), P(2,t), \ldots, P(i,t), R(t)]\), and used as the training set and test set for the time series data prediction model.

[0129] According to the data collected during the rainfall period in the basin, several single-peak flood processes formed by single-peak rainfall processes with relatively short durations are selected. Using the flat-cut method, the sub-flood is segmented from the flood rising time point to obtain the base width \(T\) of the unit hydrograph of the basin. Let \(T\) i be a time window containing \(i\) \(t\) time periods. The sum of the base width \(T\) and the time window \(T\) containing \(i\) \(t\) time periods i is the single rainfall concentration duration with \(i\) \(t\) as the rainfall duration, which is \(TT\) i .

[0130] Build a time series data prediction model. The SSM layer of the time series data prediction model includes the following parameters: the state vector \(x\) t represents the internal state of the system at time \(t\), the output dimension \(d\) is set to 128; the state transition matrix \(A\) defines how the state vector changes over time; the control input matrix \(B\) defines the influence of the control input on the state; the observation matrix \(C\) defines how the state vector is mapped to the observation vector. The \(W\) and \(U\) bias matrices, and \(b\) are weights. The mixing layer includes the following parameters: the weights and the bias matrix \(b\), \(W\).

[0131] Before the model starts training, perform initialization operations on the relevant data. The specific operations are as follows:

[0132] For the state vector \(x\) t , it is set to 0; the weight matrix \(W\) of the input mapping function g adopts uniform distribution sampling:

[0133]

[0134] where \(i\) is the number of rain gauges.

[0135] The state transition matrix \(A\), the control input matrix \(B\), the observation matrix \(C\), and the other bias matrices \(W\) and \(U\) adopt Gaussian sampling:

[0136]

[0137] According to the rainfall data of the basin, the model is divided into each SSM according to different rainfall durations \(T\) of the basin i . For a certain SSM i , the data set \(I(t)\) that meets the number of window lengths of \(T\) i is used to calculate the control input vector \(u\) through the following formula i ​t :

[0138] u t = σ(W g I(t) + b g )

[0139] where W g is the weight matrix, with size d×i; bg is the bias vector, with size d; σ is the activation function, and here the Sigmoid function is selected.

[0140] Update the state vector x according to the state update equation t+1 :

[0141] x t+1 = Ax t + Bu t

[0142] Calculate the observation vector y according to the observation equation t :

[0143] y t = Cx t

[0144] Adjust the state vector through the gating mechanism:

[0145] f t = σ(W f x t + U f I(t) + b f )

[0146] i t = σ(W i x t + U i I(t) + b i )

[0147] o t = σ(W o x t + U o I(t) + b o )

[0148] c t = f t ⊙ c t-1 + i t ⊙ tanh(W c x t + U c I(t) + b c )

[0149] h t = o t ⊙ tanh(c t)

[0150] Map the state vector after selective propagation back to the output vector:

[0151] S t = W out h t + b out

[0152] where σ is the activation function, and the Sigmoid function is selected here; ⊙ represents element-wise multiplication, and W, U, and b represent the corresponding weight and bias matrices.

[0153] Fuse and process the output of the SSM layer to extract higher-level features. Let d' = 256, and project S t into a higher-dimensional space d':

[0154] Z t = W t · S t + b t

[0155] where W is the weight matrix of size d'×d, and b is the bias vector of size 256.

[0156] Apply the ReLU activation function to the result of the linear transformation:

[0157] A t = ReLU(Z t ) = max(0, Z t )

[0158] Add a residual connection. To make a residual connection, S t needs to be mapped to the same dimension as A t :

[0159]

[0160] After the transformation, is used to make a residual connection with A t :

[0161] H t = A t + W r · S t + b r

[0162] Perform layer normalization on the output H t of the mixing layer:

[0163]

[0164] Among them, W is a weight matrix of size d′×d; b is a bias vector of size d′; μt and are the mean and variance of H t respectively. ∈ is a small constant to prevent division by zero; take ∈ = 10 -6 . γ and β are learnable scaling and offset parameters.

[0165] Output the result:

[0166]

[0167] Among them, W out is a weight matrix of size 1×d′; b out is a scalar bias.

[0168] Repeat TT i -1 times (adjust the state vector) to (output the result).

[0169] The obtained result uses the mean square error to construct a loss function to evaluate the difference between the predicted sequence and the actual sequence I(t):

[0170]

[0171] Among them, TT i is the duration of all data.

[0172] Calculate the gradients of each parameter through the loss function and perform backpropagation parameter updates. The gradient calculation method is as follows:

[0173] For the output layer, f Output represents the linear transformation of the fully connected layer, and ψ is the parameter of the output layer. The gradient of the loss L with respect to the output layer parameter ψ:

[0174]

[0175] For the mixing layer, let, f Mixer represents the non-linear transformation of the mixing layer, and φ is the parameter of the mixing layer. The gradient of the loss L with respect to the mixing layer parameter φ:

[0176]

[0177] For the SSM layer, let f SSM represents the non-linear transformation of the SSM layer, and θ is the parameter of the SSM layer. The gradient of the loss L with respect to the SSM layer parameter θ:

[0178]

[0179] Preferably, use the AdamW optimizer to update the parameters:

[0180]

[0181] Among them, β 1 and β 2 are the exponential decay rates of the momentum term, with a value of 0.9; m t and v t are the moving averages of the first and second moments of the gradient respectively; and are the estimates after bias correction; ∈ is a small constant used to prevent division by zero, with a value of 10 -8 ; η t is the learning rate; τ is the parameter to be updated in the model; the arrow represents the update operation.

[0182] Repeat steps (calculating the SSM layer) to (updating the parameters) until the loss function converges. At this time, the training of the SSM time series data prediction model is completed, denoted as SSM().

[0183] In the present invention, a control method for a pumped storage power station based on precipitation data of a basin meteorological station is also disclosed, including the following steps:

[0184] S1: The meteorological data acquisition unit continuously collects the precipitation data of the basin meteorological station through its respective terminals at a set time interval, and transmits the data to the data preprocessing module in real time through a reliable communication link;

[0185] S2: After receiving the data, the data preprocessing module first performs data cleaning by the data cleaning sub-module to remove possible incorrect or abnormal data. Then, the data calibration sub-module calibrates the data to improve data accuracy. Finally, the data conversion sub-module converts the processed data into a format that can be processed by the intelligent analysis engine and transmits it to the intelligent analysis engine;

[0186] S3: The machine learning algorithm library in the intelligent analysis engine is called to deeply analyze the input precipitation data. The pattern recognition module identifies different precipitation patterns from the data, such as continuous rainfall, intermittent rainfall, etc. The trend prediction module combines historical data and real-time data for modeling and training, and predicts the precipitation trend in the future for a period of time, including the start time, duration, and intensity of precipitation;

[0187] S4: The power station operation decision-making module will fully refer to the detailed analysis results given by the intelligent analysis engine and work closely in combination with the current actual states of the reservoir water level and the power generation power of the power station. The specific method is as follows:

[0188] For a certain time point t, perform a summation operation on the data of the meteorological station network. When it is determined that rainfall is occurring in the basin. Among them, a i$F_i$ is the control area of rainfall station $i$, obtained using the Thiessen polygon method; $F$ is the total area of the basin.

[0189] Set the mathematical notation for the current reservoir water storage as $V$. 0 . After waiting for the rainfall to end, call the SSM time series data prediction model to predict the future water inflow. The specific method is as follows:

[0190] Use the rainfall duration $T$. i Construct the model input using the precipitation records of all rainfall stations before and after the time period:

[0191] $I$ input $=[P(1,1)\cdots P(i,1)R(1),\cdots,T$ i $P(1,T$ i )\cdots P(i,T$ i )R(T$ i )]$

[0192] where $P(1,1)\cdots P(i,1)R(1),\cdots,T$ i $P(1,T$ i )\cdots P(i,T$ i )R(T$ i ) is a two-dimensional data set synthesized from the time markers from time period 1 to time period $T$, the precipitation data of each rainfall station, and the runoff data at the reservoir inlet. This two-dimensional data set will be sliced into a one-dimensional data set i $I(t)=[t P(1,t)\cdots P(i,t)R(t)]$

[0193] and continuously added to $SSM()$. Input $I$

[0194] into the SSM time series data prediction model, take the sum of the last item $R$ of the result, and predict the total water inflow data within time period $T$: input $V(T)=\sum SSM(I$

[0195] )$ input $+V$ last where $SSM(I$ 0

[0196] )$ input is the last item of the model output vector. $T$ is the duration from the increase in water inflow to the regression to the initial water inflow, and this duration is determined according to the output of the model. last At this time, the decision-making estimated pumping volume is:

[0197] where $V$

[0198]

[0199] is the decision-making estimated pumping volume. $V$ a decision-making estimated pumping volume. $V$s is the maximum allowable reservoir capacity. V u (t) is the water consumption for power generation of the pumped-storage power station during the t period.

[0200] S5. According to the pumping decision formulated by the power station operation decision module, the energy storage regulation module pumps the downstream water to a height of V a into the reservoir to achieve energy storage operation.

[0201] S6. The emergency response module will conduct early warning and response by integrating the data of the operation decision module. When it is determined as an overload situation, the water volume of will be discharged in advance, and an alarm will be issued to remind the staff of the hydropower station, so as to ensure that the power station can always be in a safe state in the face of emergencies, and maximize the stable operation of the power station and the safety of the lives and property of relevant personnel.

[0202] The above is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention, according to the technical solution and inventive concept of the present invention, makes equivalent substitutions or changes, and all should be covered by the protection scope of the present invention.

Claims

1. A control system for an energy storage power station based on precipitation data from a river basin meteorological station, characterized in that: Includes the following modules: Meteorological data collection unit: It includes a data collection terminal connected to the meteorological station in the basin, which can obtain various parameters of precipitation in real time and accurately; Data preprocessing module: including data cleaning submodule, which is used to remove noise, outliers and erroneous data from the collected data; The data calibration submodule improves the accuracy of data by comparing and calibrating with standard meteorological data; The data conversion submodule converts the collected data into a format recognized by the system; Intelligent analysis engine module: machine learning algorithm library, including algorithms suitable for precipitation data analysis; The trend prediction module uses historical data and real-time data for modeling and training to predict precipitation trends in the future; Power plant operation decision module: including strategy generation submodule, which generates power plant operation strategy according to the analysis results of the intelligent analysis engine and various operation parameters of the power plant; The risk assessment submodule evaluates and warns of the risks brought about by various decisions and optimizes the formulated strategies; Energy storage regulation module: including electric valve control system, which controls the water inlet and outlet valves of the reservoir to realize water storage and water release operations; water level monitoring system, which monitors the water level changes of the reservoir in real time; Emergency response module: including rainstorm monitoring and early warning submodule, which will immediately issue an alarm when rainstorm or abnormal precipitation is detected; The emergency avoidance strategy formulation submodule formulates measures to deal with emergency situations such as heavy rains and monitors the implementation effect in real time.

2. The energy storage power station control system based on precipitation data from a river basin meteorological station according to claim 1 is characterized in that: The time series analysis algorithm training steps are: Set up a time axis with equidistant time marks t at intervals of 1 hour. Assuming that there are i rain gauges in the basin controlled by the reservoir, the precipitation data of each rain gauge obtained by the data preprocessing module in previous years are sorted into a two-dimensional data set P(i, t), and the reservoir's previous water flow data are sorted into a one-dimensional data set R(i, t), where i represents the number of the meteorological station and t represents the timestamp when the data is recorded. The Unix timestamp standard is selected as the timestamp standard. If there is a mismatch, the adjacent data before and after the required time stamp are used to fill the gap through a linear interpolation algorithm. Specifically, for the rainfall station, there is Where Pi(t') is the precipitation data of the rain gauge numbered i before the time mark to be obtained, Pi(t") is the precipitation data of the rain gauge numbered i after the time mark to be obtained, t' is the timestamp before the time mark to be obtained, t" is the timestamp after the time mark to be obtained, and t is the time mark to be obtained; For reservoirs, there are Wherein, R(t') is the water inflow data of the reservoir before the time mark required, R(t") is the water inflow data of the reservoir after the time mark required, t' is the timestamp before the time mark required, t" is the timestamp after the time mark required, and t is the time mark required; Each time tag t and the corresponding data sets P(i, t) and R(t) are combined into a data set I(t) = [t, P(1, t), P(2, t), …, P(i, t), R(t)], and used as the training set and test set of the time series data prediction model; According to the data collected during the rainfall period in the basin, the unit line bottom width T of the basin is obtained. Let T i is a time window containing i t periods, the bottom width T and the time window T containing i t periods i The sum of the rainfall duration is the duration of a single rainfall event with i t as the rainfall duration, which is TT i， Build a time series data prediction model. The SSM layer of the time series data prediction model includes the following parameters: state vector x t represents the internal state of the system at time t, and the output dimension d is set to 128; the state transfer matrix A defines how the state vector changes over time; the control input matrix B defines the impact of the control input on the state; the observation matrix C defines how the state vector is mapped to the observation vector, W and U bias matrices, b are weights, and the mixing layer includes the following parameters: weight and bias matrices b, W; Before the model starts training, the relevant data is initialized. The specific operations are as follows: For the state vector x t , set to 0; input the weight matrix W of the mapping function g Use uniform distribution sampling: Where i is the number of rainfall stations, and the state transfer matrix A, control input matrix B, observation matrix C, and other bias matrices W and U use Gaussian sampling: According to the rainfall data of the basin, according to the different rainfall duration T of the basin i Divide the model into individual SSMs i , for a certain SSM i , will meet T i The window length of the number of data sets I(t) is calculated by the following formula to control the input vector u t : u t =σ(W g I(t)+b g ) Among them, W g is the weight matrix, size is d×i; bg is the bias vector, size is d; σ is the activation function, here we use the Sigmoid function; Update the state vector x according to the state update equation t+1 : x t+1 =Ax t +Bu t Calculate the observation vector y according to the observation equation t : y t =Cx t Adjust the state vector through the gating mechanism: f t =σ(W f x t +U f I(t)+b f ) i t =σ(W i x t +U i I(t)+b i ) o t =σ(W o x t +U o I(t)+b o ) c t =f t ⊙c t-1 +i t ⊙tanh(W c x t +U c I(t)+b c ) h t =o t ⊙tanh(c t ) Map the state vector after selective propagation back to the output vector: S t =W out h t +b out Among them, σ is the activation function, here we use the Sigmoid function; ⊙ represents element-by-element multiplication, W, U and b represent the corresponding weight and bias matrices, The output of the SSM layer is fused and processed to extract higher-level features. Let d′ = 256 and S t Projection to a higher-dimensional space d′: Z t =W t ·S t +b t Where W is a weight matrix of size d′×d, b is a bias vector of size 256, Apply the ReLU activation function to the result of the linear transformation: AND t =ReLU(Z t )=max(0,Z t ) Add residual connection. In order to perform residual connection, S t Mapped to A t Same dimensions: After transformation, the With A t Make a residual connection: H t =A t +W r ·S t +b r The output H of the mixing layer t Perform layer normalization: Where W is a weight matrix of size d′×d; b is a bias vector of size d′; μt and They are H t The mean and variance of ∈ is a small constant to prevent division by zero; ∈=10 -6 , γ and β are learnable scaling and offset parameters; Output the results: Among them, W out is a weight matrix of size 1×d′; b out is the scalar bias, Repeat TT i -1 times (adjust the state vector) to (output the result); The obtained results use the mean square error to construct a loss function to evaluate the predicted sequence The difference between the actual sequence I(t) is: Among them, TT i is the duration of all data. The gradient of each parameter is calculated through the loss function, and the back propagation parameter is updated. The gradient calculation method is as follows: For the output layer, f Output represents the linear transformation of the fully connected layer, ψ is the parameter of the output layer, and the gradient of the loss L with respect to the output layer parameter ψ is: For the mixed layer, let f Mixer represents the nonlinear transformation of the mixing layer, φ is the parameter of the mixing layer, and the gradient of the loss L to the mixing layer parameter φ is: For the SSM layer, let f SSM represents the nonlinear transformation of the SSM layer, θ is the parameter of the SSM layer, and the gradient of the loss L to the SSM layer parameter θ is: Preferably, the parameters are updated using the AdamW optimizer: Among them, β1 and β2 are the exponential decay rates of the momentum term, which are set to 0.9; m t and v t are the moving averages of the first and second moments of the gradient, respectively; and is the bias-corrected estimate; ∈ is a small constant used to prevent division by zero, with a value of 10 -8 ; η t is the learning rate; τ is the parameter that needs to be updated in the model; the arrow represents the update operation; Repeat steps (calculate SSM layer) to (update parameters) until the loss function converges. At this point, the SSM time series data prediction model training is completed, recorded as SSM().

3. The energy storage power station control system based on precipitation data of a river basin meteorological station according to claim 1 is characterized in that: The specific method of the power station operation decision module is: At a certain time point t, the data of the weather station network are summed up. When , it is determined that the basin is experiencing rainfall, where a i is the control area of ​​rainfall station i, and F is the total area of ​​the basin; Set the current reservoir water storage capacity as V 0, After waiting for the rainfall to end, the SSM time series data prediction model is called to predict the future water inflow. The specific method is as follows: Using rainfall duration T i The precipitation records of all rain gauges before and after the period are used to construct the model input: I input =[1P(1,1)...P(i,1)R(1),...,T i P(1,T i )...P(i,T i )R(T i )] Where 1P(1,1)...P(i,1)R(1),...,T i P(1,T i )...P(i,T i )R(T i ) is from period 1 to period T i The time stamp, precipitation data of each rain gauge and runoff data at the reservoir inlet are synthesized into a two-dimensional data set. This two-dimensional data set will be sliced ​​into a one-dimensional data set I(t)=[t P(1,t)...P(i,t)R(t)] Continue to add to SSM(), I input Input it into the SSM time series data prediction model, take the last item R of the result and sum it, and predict the total amount of water inflow data in the T period: V(T)=∑SSM(I input ) last +V0 Among them, SSM(I input ) last is the last item of the model output vector, T is the time it takes for the water inflow to grow and return to the initial water inflow, which is determined by the output of the model; At this time, the estimated pumping volume is: Among them, V a Decision-making expected pumping volume, V s is the maximum allowable storage capacity of the reservoir, V u (t) is the water consumption of the pumped storage power station during the period t.

4. A method for applying a storage power station control system based on precipitation data of a river basin meteorological station as described in any one of claims 1 to 3, characterized in that: The following steps are involved: S1, the meteorological data acquisition unit continuously collects precipitation data from the basin meteorological station through its various terminals at set time intervals, and transmits the data to the data preprocessing module in real time through a reliable communication link; S2. After the data preprocessing module receives the data, the data cleaning submodule first cleans the data to remove erroneous data, then the data calibration submodule calibrates the data, and finally the data conversion submodule converts the processed data into a format that can be processed by the intelligent analysis engine and passes it to the intelligent analysis engine; S3: The machine learning algorithm library in the intelligent analysis engine is called to conduct in-depth analysis of the input precipitation data. The pattern recognition module identifies different precipitation patterns from the data. The trend prediction module combines historical data and real-time data for modeling and training to predict the precipitation trend in the future. S4, the power station operation decision module generates the optimal power station operation strategy through the strategy generation submodule based on the analysis results of the intelligent analysis engine and the current status of the power station reservoir water level and power generation power. The risk assessment submodule conducts risk assessment on the generated strategy and adjusts the strategy according to new data or situations; S5. The energy storage regulation module uses the strategy formulated by the power station operation decision module to accurately control the water inlet and outlet valves of the reservoir through the electric valve control system to achieve water storage and water release operations. The water level monitoring system provides real-time feedback on the changes in the reservoir water level. The energy conversion and storage equipment ensures the efficient conversion and storage of electrical energy. S6. The rainstorm monitoring and warning submodule in the emergency response module monitors meteorological data at all times and monitors the execution effect in real time.

Images