Real-time warning method and device for gate opening and closing status based on TCN neural network
Through the method based on TCN neural network, the gate opening and closing status is monitored in real time, which solves the problem that the existing technology cannot judge the gate position in real time, and accurately monitors and early warnings of the gate opening and closing status, improving safety and reliability.
Patent Information
- Application Number
- CN202310350066.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2023-03-31
- Publication Date
- 2024-12-20
- Estimated Expiration
- 2043-03-31
AI Technical Summary
The prior art cannot monitor and judge the gate position status in real time during the gate opening and closing process, resulting in problems such as lax door drop, rework and delay in maintenance.
Using a method based on TCN neural network, by constructing a mathematical model, the gravity, buoyancy, main hook weight and other stress parameters of the gate are obtained, the mathematical model of the input sequence and the output sequence is established, the TCN time series model is constructed, and the model is trained in real time. Predict the weight of the main hook, and by comparing the difference between the measured value and the predicted value, real-time monitoring and early warning of the gate opening and closing state.
Real-time monitoring and early warning of the gate opening and closing status is realized, safety and reliability are improved, and foreign object blocking and inadequate closing of the gate are detected in a timely manner to ensure the correctness of gate opening and closing operations.
Smart Images
Figure CN116579230B_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of gate opening and closing state monitoring and early warning, and in particular relates to a gate opening and closing state real-time early warning method and device based on a TCN neural network. Background Art
[0002] The gate is a common water retaining metal structure of a hydropower station. It is generally composed of a flat panel, a frame, supporting walking parts, a sling and a water stop. It moves up and down in the gate slot. According to the installation location or function, it is divided into unit water inlet working door, unit water inlet maintenance door, unit tailwater working door, unit tailwater maintenance door, etc.
[0003] When overhauling the units of a hydropower station, it is necessary to close the gates and keep the flow channel dry. However, there may be debris such as tree stumps in the gate slot, which may cause the gates to fail to close properly. If this problem is not discovered in time, the gates will not be closed properly until the subsequent work such as drainage of the flow channel is not going smoothly, which will lead to rework and delay the maintenance period. Therefore, how to monitor the gate position status in real time during the gate opening and closing process, detect gate position abnormalities at the first time, and determine whether the gate is closed properly after the gate is closed is a difficult point in the quality control of gate opening and closing operations that needs to be improved urgently.
[0004] In the field of hydropower, the current methods for judging the gate status include the following:
[0005] 1. The staff will make some marks on the cable, and when the gate is lowered, they can judge the gate in place to a certain extent based on their experience, but the reliability is limited, and once the operator is replaced, it cannot be controlled;
[0006] 2. Install a high-definition underwater camera on the hoist grab beam, which can clearly see part of the underwater impact. However, due to factors such as underwater clarity and light, the observation distance is limited;
[0007] 3. Install an encoder on the gate hoist to obtain the height of the hook according to the length of the wire rope, which can reflect the gate position to a certain extent. However, when the gate is blocked and cannot continue to fall, the hook will continue to fall, resulting in the hook position value not being able to reflect the true position of the gate;
[0008] Since the flow channel of a hydropower station unit is usually tens of meters underwater, the underwater visibility is poor and it is impossible to deploy sensors. Therefore, there is currently no mature technical means to determine whether the gate position is correct in real time during the gate opening and closing process. Summary of the invention
[0009] In view of the technical problems existing in the background technology, the real-time early warning method and device for gate opening and closing status based on TCN neural network provided by the present invention, when the gate opening and closing machine performs gate opening and closing operations, by building a model, collecting relevant parameters, and predicting the weight change of the main hook, the gate status can be judged in real time, and abnormal situations can be warned, thereby improving safety and reliability.
[0010] In order to solve the above technical problems, the present invention adopts the following technical solutions to achieve:
[0011] A real-time early warning method for gate opening and closing status based on TCN neural network, the steps are:
[0012] Step 1, obtain data: obtain the gate's gravity G and buoyancy f 浮力 , the main hook weight T, and obtain the friction force f when the gate is opened and closed normally 摩擦 , other forces on the gate are expressed in N;
[0013] Step 2, perform a force analysis on the gate;
[0014] Step 3: Construct a mathematical model based on force analysis:
[0015] Input sequence: x=(Process code, gate weight, gate height, upstream waterlevel, main hook height);
[0016] Output sequence: Mathematical model of y= (Weight of main hook); real-time acquisition of process code, gate weight, gate height, upstream water level and main hook height values;
[0017] Among them, Process code is the process code;
[0018] gate weight is the gate weight parameter;
[0019] gate height is the gate height parameter;
[0020] upstream water level is the upstream water level parameter;
[0021] Main hook height is the main hook height parameter;
[0022] Weight of main hook;
[0023] Step 4, build a TCN time series model;
[0024] Step 5: Perform model training;
[0025] Step 6, conduct prediction and early warning: collect data to predict the main hook weight while the gate is opening and closing, compare the measured value and the predicted value of the main hook weight in real time, and monitor the changes in the difference between the two to achieve the purpose of real-time judgment of the gate opening and closing status; according to the different changing trends of the measured value and the predicted value, conduct real-time monitoring and early warning in various situations.
[0026] Preferably, in step 2, gravity G, buoyancy f 浮力 , main hook weight T, friction force f 摩擦 The force analysis formula for other forces N is:
[0027] T+f 浮力 +f 摩擦 =G+N
[0028] T = G + Nf 浮力 -f 摩擦
[0029] The formula between the main hook height and buoyancy is:
[0030] f 浮力 =ρgh,
[0031] h=h 主钩高度 +h 闸门高度 -(h 主梁高程 -h 水位高程 )
[0032] f 摩擦 =μ× Fn,
[0033] In the formula, f 摩擦 Relatively fixed, other forces N are the main influencing factors causing the change of main hook weight during the opening and closing process of the gate. Therefore, the opening and closing status of the gate can be obtained by studying the change of main hook weight.
[0034] Preferably, in step 4, the TCN time series model includes a 4-layer network, and the key parameters of the 4-layer TCN time series convolutional network are: convolution kernel k=2, void exaggeration factor d=[1,2,4,8], the activation function used is relu, the training loss is MAE, and the optimizer is adam.
[0035] Preferably, in step 5, the model training process is as follows:
[0036] Input sequence: x=(Process code, gate weight, gate height, upstream waterlevel, main hook height);
[0037] Output sequence: y = (Weight of main hook);
[0038] Among them, Process code is the process code;
[0039] gate weight is the gate weight parameter;
[0040] gate height is the gate height parameter;
[0041] upstream water level is the upstream water level parameter;
[0042] Main hook height is the main hook height parameter;
[0043] Weight of main hook;
[0044] The data of normal operation of the gate hoist is obtained, and the process code, gate weight, gate height, upstream water level and main hook height parameters are used to form an input vector, and the main hook weight is used to form an output vector, which are respectively input into the TCN neural network to predict the time series change value of the main hook weight.
[0045] Preferably, after the network training is completed, data is collected to predict the main hook weight while the gate is opened and closed, and the measured value and the predicted value of the main hook weight are compared in real time. By monitoring the change of the difference between the two, the purpose of real-time judgment of the gate opening and closing state is achieved; according to the different change trends of the measured value and the predicted value, including at least the following three working conditions, the system can realize real-time monitoring and early warning:
[0046] 1) If the gate enters the gate slot incorrectly and collides with the gate slot, the main hook weight value T may suddenly increase;
[0047] 2) If there are foreign objects such as wooden stakes at the bottom of the door slot, the main hook height may not reach the maximum value and the main hook weight value T may decrease in advance;
[0048] 3) If an operation process is completed, and the difference between the predicted value and the measured value of the main hook weight is within the threshold range, the system determines that the gate opening and closing operation process is correct and the gate opening and closing status is correct.
[0049] A real-time early warning device for gate opening and closing status based on TCN neural network adopts the real-time early warning method for gate opening and closing status based on TCN neural network.
[0050] This patent can achieve the following beneficial effects:
[0051] 1. First, obtain the full process parameters affected by the state of the normal operation process of opening and closing the gate, and use the process code, gate weight, gate height, upstream water level and main hook height value to form an input sequence, and the main hook weight value to form an output sequence; then use the input and output sequences to train the main hook weight value prediction model based on the TCN neural network; finally, by comparing the difference between the predicted value and the measured value of the main hook weight in real time, the purpose of real-time warning of the gate opening and closing status can be achieved.
[0052] 2. TCN stands for Temporal Convolutional Network, which is a relatively advanced time series prediction method. TCN neural network is more accurate, simpler and clearer than typical recursive networks (such as LSTM and GRU). Based on the TCN neural network, it can capture parameter information such as hook height, gate parameters and upstream water level during the gate opening and closing process, and realize real-time prediction of the main hook weight. On the one hand, it can timely detect abnormal conditions such as foreign objects blocking and gate not closing in place during the gate opening and closing process. On the other hand, it can independently judge whether the gate is opened or closed after the gate opening and closing operation is completed. BRIEF DESCRIPTION OF THE DRAWINGS
[0053] The present invention will be further described below in conjunction with the accompanying drawings and embodiments:
[0054] Figure 1 It is a schematic diagram of the gate opening and closing state of the present invention;
[0055] Figure 2 Schematic diagram of the gate opening and closing state of the present invention;
[0056] Figure 3 It is a prediction flow chart of the present invention;
[0057] Figure 4 This is the TCN neural network structure diagram of the present invention. DETAILED DESCRIPTION
[0058] Embodiment 1:
[0059] The preferred solution is Figures 1 to 4 As shown, a real-time warning method and device for gate opening and closing status based on TCN neural network, the steps are:
[0060] Step 1, obtain data: obtain the gate's gravity G and buoyancy f 浮力 , the main hook weight T, and obtain the friction force f when the gate is opened and closed normally 摩擦 , other forces on the gate are expressed in N;
[0061] Step 2, perform a force analysis on the gate;
[0062] Gravity G, buoyancy f 浮力, main hook weight T, friction force f 摩擦 The force analysis formula for other forces N is:
[0063] T+f 浮力 +f 摩擦 =G+N
[0064] T = G + Nf 浮力 -f 摩擦
[0065] Gate gravity is the weight of the gate to be opened and closed. The value varies depending on the gate to be opened and closed, but the value is determined within an operation process;
[0066] The buoyancy will change in real time with the gate opening and closing height. During the gate opening and closing process, the gate height, the main beam elevation of the gantry crane and the upstream water level elevation are all fixed values. Only the main hook height changes with time, so the buoyancy changes linearly with time; the formula between the main hook height and buoyancy is:
[0067] f 浮力 =ρgh,
[0068] h=h 主钩高度 +h 闸门高度 -(h 主梁高程 -h 水位高程 )
[0069] Friction force f 摩擦 =μ× Fn,
[0070] f 摩擦 Under normal circumstances, it is relatively fixed.
[0071] Other forces N are the main factors affecting the weight change of the main hook during the opening and closing process of the gate.
[0072] If only dirt and other debris are attached to the gate, the weight of the main hook will increase, but in the same opening and closing operation process, obvious periodic characteristics can be seen, such as Figure 2 The following are the main hook weight diagrams for three consecutive gate closing periods. If the gate enters the gate slot incorrectly and collides with the gate slot, the main hook weight value T may change suddenly; if there are foreign objects such as wooden stakes at the bottom of the gate slot, the main hook height may not reach the maximum value and the main hook weight value T may decrease in advance.
[0073] Step 3: Construct a mathematical model based on force analysis:
[0074] Input sequence: x=(Process code, gate weight, gate height, upstream waterlevel, main hook height);
[0075] Output sequence: Mathematical model of y= (Weight of main hook); real-time acquisition of process code, gate weight, gate height, upstream water level and main hook height values;
[0076] Among them, Process code is the process code; a complete gate closing process includes: trolley movement, main hook grab beam descending to the gate head, grab beam pinning, unlocking, gate descending, gate closing in place, grab beam pin withdrawal, grab beam rising, which are process codes (1, 2, 3, 4, 5, 6, 7, 8) respectively, and can be obtained through the process code acquisition module;
[0077] Gate weight is the gate weight parameter, which is obtained through the gate weight acquisition module;
[0078] Gate height is the gate height parameter, which is obtained through the gate height acquisition module;
[0079] Upstream water level is the upstream water level parameter, which is obtained through the upstream water level acquisition module;
[0080] main hook height is the main hook height parameter; it is obtained through the main hook height acquisition module.
[0081] Weight of main hook: The real-time true value y can be obtained through the main hook weight acquisition module, and the predicted value y' can be obtained through the prediction module, and the difference between the two can be compared.
[0082] Step 4: Build a TCN time series model. The key parameters of the constructed 4-layer TCN time series convolutional network are: convolution kernel k=2, void exaggeration factor d=[1, 2, 4,8]. The activation function used is relu, the training loss is MAE, and the optimizer is adam.
[0083] The key parameters for constructing a 4-layer TCN temporal convolutional network are: convolution kernel k=2, void exaggeration factor d=[1, 2, 4,8], activation function used is relu, training loss is MAE, and optimizer is adam.
[0084] Step 5: Perform model training;
[0085] The model training process is as follows:
[0086] Input sequence: x=(Process code, gate weight, gate height, upstream waterlevel, main hook height);
[0087] Output sequence: y=(Weight of main hook).
[0088] Process code is the process code. A complete gate closing process includes: trolley movement, main hook grab beam descending to the gate head, grab beam pinning, unlocking, gate descending, gate closing in place, grab beam pin withdrawal, grab beam rising, which are process codes (1, 2, 3, 4, 5, 6, 7, 8) respectively, and can be obtained through the process code acquisition module;
[0089] Gate weight is the gate weight parameter, which is obtained through the gate weight acquisition module;
[0090] Gate height is the gate height parameter, which is obtained through the gate height acquisition module;
[0091] Upstream water level is the upstream water level parameter, which is obtained through the upstream water level acquisition module;
[0092] Main hook height is the main hook height parameter, which is obtained through the main hook height acquisition module.
[0093] Weight of main hook is the weight of the main hook. The real-time true value y can be obtained through the main hook weight acquisition module, and the predicted value y' can be obtained through the prediction module, and the difference between the two can be compared.
[0094] The data of normal operation of the gate hoist is obtained, and the process code, gate weight, gate height, upstream water level and main hook height parameters are used to form an input vector, and the main hook weight is used to form an output vector, which are respectively input into the TCN neural network to predict the time series change value of the main hook weight.
[0095] Step 6, conduct prediction and early warning: collect data to predict the main hook weight while the gate is opening and closing, compare the measured value and the predicted value of the main hook weight in real time, and monitor the changes in the difference between the two to achieve the purpose of real-time judgment of the gate opening and closing status; according to the different changing trends of the measured value and the predicted value, conduct real-time monitoring and early warning in various situations.
[0096] After the network training is completed, data is collected to predict the main hook weight while the gate is opening and closing. The measured value and predicted value of the main hook weight are compared in real time. By monitoring the change of the difference between the two, the purpose of real-time judgment of the gate opening and closing status is achieved. According to the different change trends of the measured value and the predicted value, including at least the following three working conditions, the system can monitor and warn in real time:
[0097] 1) If the gate enters the gate slot incorrectly and collides with the gate slot, the main hook weight value T may suddenly increase;
[0098] 2) If there are foreign objects such as wooden stakes at the bottom of the door slot, the main hook height may not reach the maximum value and the main hook weight value T may decrease in advance;
[0099] 3) If an operation process is completed, and the difference between the predicted value and the measured value of the main hook weight is within the threshold range, the system determines that the gate opening and closing operation process is correct and the gate opening and closing status is correct.
[0100] A real-time warning device for gate opening and closing status based on TCN neural network adopts the real-time warning method for gate opening and closing status based on TCN neural network, which includes gate weight acquisition module, gate height acquisition module, upstream water level acquisition module, main hook height acquisition module and main hook weight acquisition module. On the one hand, the device can timely detect abnormal conditions such as foreign matter blocking and gate not closing in place during gate opening and closing, and on the other hand, it can autonomously judge whether the gate is opened or closed in place after the gate opening and closing operation is completed.
[0101] The above embodiments are only preferred technical solutions of the present invention and should not be regarded as limiting the present invention. The protection scope of the present invention shall be the technical solutions recorded in the claims, including equivalent replacement solutions of the technical features in the technical solutions recorded in the claims. That is, equivalent replacement improvements within this scope are also within the protection scope of the present invention.
Claims
1. A real-time early warning method for gate opening and closing status based on TCN neural network, characterized by: Step 1, obtain data: obtain the gate's gravity G and buoyancy f 浮力 , the main hook weight T, and obtain the friction force f when the gate is opened and closed normally 摩擦 , other forces on the gate are expressed in N; Step 2, perform a force analysis on the gate; Step 3: Construct a mathematical model based on force analysis: Input sequence: x=(Process code, gate weight, gate height, upstream water level, main hook height); Output sequence: Mathematical model of y= (Weight of main hook); real-time acquisition of process code, gate weight, gate height, upstream water level and main hook height values; Among them, Process code is the process code; gate weight is the gate weight parameter; gate height is the gate height parameter; upstream water level is the upstream water level parameter; Main hook height is the main hook height parameter; Weight of main hook; Step 4, build a TCN time series model; Step 5: Perform model training; Step 6, prediction and early warning: collect data to predict the main hook weight while the gate is opening and closing, compare the measured value and the predicted value of the main hook weight in real time, and monitor the change of the difference between the two to achieve the purpose of real-time judgment of the gate opening and closing status; according to the different change trends of the measured value and the predicted value, real-time monitoring and early warning are carried out in various situations; In step 2, gravity G and buoyancy f 浮力 , main hook weight T, friction force f 摩擦 The force analysis formula for other forces N is: T+f 浮力 +f 摩擦 =G+N T= G+N-f 浮力 -f 摩擦 The formula between the main hook height and buoyancy is: f 浮力 =ρgh, h=h 主钩高度 +h 闸门高度 -(h 主梁高程 -h 水位高程 ) f 摩擦 =μ× Fn, In the formula, f 摩擦 Relatively fixed, other forces N are the main factors affecting the weight change of the main hook during the gate opening and closing process. Therefore, the gate opening and closing status is obtained by studying the weight change of the main hook. In step 4, the TCN time series model includes a 4-layer network. The key parameters of the 4-layer TCN time series convolutional network are: convolution kernel k=2, void exaggeration factor d=[1,2,4,8], activation function used is relu, training loss is MAE, and optimizer is adam; In step 5, the model training process is as follows: Input sequence: x=(Process code, gate weight, gate height, upstream water level, main hook height); Output sequence: y = (Weight of main hook); Among them, Process code is the process code; gate weight is the gate weight parameter; gate height is the gate height parameter; upstream water level is the upstream water level parameter; Main hook height is the main hook height parameter; Weight of main hook; Obtain the data of normal operation of the gate hoist, and use the process code, gate weight, gate height, upstream water level and main hook height parameters to form an input vector, and the main hook weight to form an output vector, which are respectively input into the TCN neural network to predict the time series change value of the main hook weight; After the network training is completed, data is collected to predict the main hook weight while the gate is opening and closing. The measured value and the predicted value of the main hook weight are compared in real time. By monitoring the change of the difference between the two, the purpose of real-time judgment of the gate opening and closing status is achieved; according to the different change trends of the measured value and the predicted value, including at least the following three working conditions, the system can realize real-time monitoring and early warning: 1) If the gate enters the gate slot incorrectly and collides with the gate slot, the main hook weight value T may suddenly increase; 2) If there are foreign objects such as wooden stakes at the bottom of the door slot, the main hook height may not reach the maximum value and the main hook weight value T may decrease in advance; 3) If an operation process is completed, and the difference between the predicted value and the measured value of the main hook weight is within the threshold range, the system determines that the gate opening and closing operation process is correct and the gate opening and closing status is correct.