Intelligent sludge discharge control method, system, equipment and medium for sedimentation tank in water plant

Abstract

The present invention relates to an intelligent sludge discharge control method, system, equipment and medium for a sedimentation tank in a water plant, and the method comprises: dividing the sludge discharge data into multiple segments or samples, and obtaining a data set after processing; constructing and coupling an empirical model and a deep learning model to construct a sedimentation tank sludge production prediction model; constructing and training a sediment distribution prediction model using a neural network model and a data set; obtaining the sludge accumulation in each area of ​​the sedimentation tank according to the average sludge production output by the sludge production prediction model, the sludge distribution ratio of each area output by the sediment distribution prediction model, and the total water inflow corresponding to the sludge discharge cycle, and obtaining the mud level height of each area of ​​the sedimentation tank in combination with the mud level data; outputting a sludge discharge plan based on the mud level height and mud level threshold of each area of ​​the sedimentation tank. Therefore, this series of innovative methods and technical means of the present invention significantly improves the intelligent level of sedimentation tank sludge discharge, and brings substantial progress to the field of water treatment.

Description

Technical Field

[0001] The present invention relates to the technical field of water treatment, and in particular to an intelligent sludge discharge control method, system, equipment and medium for a sedimentation tank in a waterworks. Background Art

[0002] In current sludge control practices, the periodic sludge discharge strategy is widely adopted. However, this method is not flexible enough when facing fluctuations in influent conditions and reagent dosage, and often leads to two problems: one is that the sludge discharge interval is too long, resulting in excessive sludge accumulation; the other is that the sludge discharge is too frequent, resulting in unnecessary loss of water resources. The key constraint behind this situation is that there is currently a lack of accurate estimation of sludge generation and detailed simulation of the distribution of bottom sludge in sedimentation tanks.

[0003] In the existing model spectrum for predicting sludge production, empirical models and deep learning models are two important categories. Although empirical models have certain practicality, their effectiveness depends largely on accurate input parameters, which makes it difficult to dynamically adapt to real-time changes in influent conditions. On the other hand, deep learning models can mine complex correlations from a large amount of historical data. However, due to the relatively long sludge discharge cycle, it becomes very difficult to obtain high-quality training samples, which to some extent affects the maturity and accuracy of the model.

[0004] In terms of simulating sediment distribution, traditional methods mainly rely on manually measuring the mud level at each point in the sedimentation tank to depict the sedimentation trend. This method is not only time-consuming and laborious, but also has a low monitoring frequency and cannot meet the real-time monitoring needs of sedimentation tank sediment distribution. In addition, the distribution of sediment is also significantly affected by seasonal changes and water inlet characteristics. For example, in summer, due to the high turbidity of raw water and the increase in heavy pollutants, the amount of coagulant needs to be increased, which will lead to an increase in the proportion of sludge at the front end of the sedimentation tank, further increasing the difficulty of simulating sediment distribution. Summary of the invention

[0005] 1. Technical issues to be resolved

[0006] In view of the above-mentioned shortcomings and deficiencies of the prior art, the present invention provides an intelligent sludge discharge control method, system, equipment and medium for the sedimentation tank of a water plant, which solves the technical problem that the existing sludge discharge control strategies, sludge production prediction models and bottom mud distribution simulation methods all have obvious deficiencies and are difficult to meet the flexibility and accuracy requirements in practical applications.

[0007] (II) Technical solution

[0008] In order to achieve the above object, the main technical solutions adopted by the present invention include:

[0009] In a first aspect, an embodiment of the present invention provides an intelligent sludge discharge control method for a sedimentation tank in a water plant, comprising:

[0010] The collected mud discharge data is divided into multiple segments or samples and processed to obtain data sets;

[0011] Construct an empirical model that maps the relationship between key influencing factors and sludge production and a deep learning model for learning the relationship between multiple influencing factors, and couple the empirical model and deep learning model trained based on the data set to construct a sedimentation tank sludge production prediction model;

[0012] Use neural network models and data sets to build and train sediment distribution prediction models;

[0013] According to the average sludge production output by the sludge production prediction model, the sludge distribution ratio of each area output by the bottom sediment distribution prediction model, and the total water inflow of the corresponding sludge discharge period, the sludge accumulation in each area of ​​the sedimentation tank is obtained, and the mud level height of each area of ​​the sedimentation tank is obtained in combination with the mud level data;

[0014] Based on the mud level heights in each area of ​​the sedimentation tank and the dynamically generated mud level threshold, at least one mud discharge plan is output.

[0015] Optionally, the collected mud discharge data is divided into multiple segments or samples, and processed to obtain a data set including:

[0016] Based on the collected sludge discharge information, the average sludge production in a single sludge discharge cycle and the proportion distribution of sludge in each area during each sludge discharge period are calculated;

[0017] Determine the monitoring index data of interest, including the hourly average values ​​of inlet turbidity, outlet turbidity, inlet flow, flocculant dosage, coagulant aid dosage, water temperature, sedimentation tank liquid level, and sedimentation tank mud level;

[0018] The collected mud discharge data are traversed, and according to the mud discharge status, the mud discharge data are divided into multiple segments or samples of different lengths, including at least one set of monitoring indicator data concerned in the mud discharge cycle, the average mud production in a single mud discharge cycle, and the distribution ratio of the sedimentation tank bottom mud;

[0019] Down-sample the monitoring indicator data of interest to obtain a new input sequence;

[0020] A data set is formed based on the new input sequence, the average sludge production in a single sludge discharge cycle, and the distribution ratio of sedimentation tank sludge, and the data set is divided into a training set and a test set according to the set ratio.

[0021] Optionally, based on the collected sludge discharge information, the average sludge production in a single sludge discharge cycle and the proportion distribution of sludge in each area in each sludge discharge period are calculated, including:

[0022] At the beginning and end of each sludge discharge cycle, the amount of dry sludge discharged during the sludge discharge period is calculated based on the collected liquid level height of the sludge discharge pool before and after sludge discharge, the average sludge concentration in the sludge discharge pool before and after sludge discharge, the bottom area of ​​the sludge discharge pool, the flow rate discharged from the sludge discharge pool, and the duration of sludge discharge after the sludge discharge pool reaches the upper limit of capacity;

[0023] According to the amount of dry sludge discharged during the sludge discharge period and the cumulative water intake of the corresponding sludge discharge cycle, the average sludge production of a single sludge discharge cycle is calculated:

[0024] Determine the length of the sedimentation tank, the speed of the sludge suction vehicle and the sludge discharge period;

[0025] When the sludge suction vehicle starts to discharge sludge at the travel speed, the sludge flow rate and sludge concentration data corresponding to each position during the vehicle's travel, as well as the time it takes for the vehicle to pass through each position, are recorded;

[0026] According to the sludge flow rate and sludge concentration data at each position during the driving process and the time the vehicle passes through each position, the sludge discharge volume of the sludge suction vehicle at each position is calculated;

[0027] Based on the sludge discharge volume of the sludge suction vehicle at any position, the sludge distribution of the sludge volume at any position as a percentage of the total sludge production in the sedimentation tank is calculated;

[0028] The sedimentation tank is equally divided into a number of unit areas, and the sludge discharge period is equally divided into a number of time periods corresponding to the number of unit areas, and the proportion of sludge production in each unit area or each time period is calculated respectively;

[0029] in,

[0030] The amount of dry sludge discharged during the sludge discharge period is:

[0031] ;

[0032] In the formula, S total is the amount of dry sludge discharged during the sludge discharge period, H 1 and H 0 are the liquid level heights of the mud discharge pool before and after mud discharge, SS 1 and SS 0 are the average sludge concentrations in the sludge pool before and after sludge discharge, A is the bottom area of ​​the sludge pool, Q 1 is the flow rate discharged from the sludge pool, T 1 is the duration of sludge discharge from the sludge pool after reaching the upper capacity limit;

[0033] The average mud production in a single mud discharge cycle is:

[0034] ;

[0035] In the formula, S is the average production of a single sludge discharge cycle, The cumulative water intake for this sludge discharge cycle is: Q t for t The water flow rate of the sedimentation tank at any time, T total The duration of this sludge discharge cycle;

[0036] The amount of mud discharged by the mud suction vehicle at each position is:

[0037] S x = SS t × Q t × dt ;

[0038] In the formula, S x for t Driving to location at all times x The amount of sludge discharged SS t and Q t They are the sludge concentration and sludge flow rate measured by the sensor on the sludge discharge pipe at this position, dt For driving through x The time spent at the location;

[0039] The distribution of sludge at any position as a percentage of the total sludge produced in the sedimentation tank is:

[0040] ;

[0041] In the formula, For location x The proportion of sludge produced in the sedimentation tank to the total sludge produced in the sedimentation tank. T RS It is the mud discharge period;

[0042] The proportion of mud production in each unit area or each time period is:

[0043] ;

[0044] In the formula, and For driving in i Average sludge concentration and average flow rate per unit area / time period, M is the number of divided units.

[0045] Optionally, an empirical model for mapping the relationship between key influencing factors and sludge production and a deep learning model for learning the relationship between multiple influencing factors are constructed, and the empirical model and the deep learning model trained based on the data set are coupled to construct a sedimentation tank sludge production prediction model including:

[0046] Determine the key factors affecting sludge production, including influent turbidity and reagent dosage;

[0047] Based on the law of conservation of mass, an empirical model is constructed to map the relationship between key influencing factors and sludge production;

[0048] Based on the data set, taking the mean square error as the optimization target, the parameters in the empirical model are solved by at least one round of iteration through the gradient descent method to complete the training of the empirical model;

[0049] A long short-term memory network is used to construct a deep learning model including an input layer, a mask layer, two LSTM layers, and two fully connected layers;

[0050] In the data set, hourly inlet turbidity, outlet turbidity, inlet flow, dosage, water temperature, sedimentation tank level and sedimentation tank mud level are selected as input features of the deep learning model, and the sedimentation tank mud level is selected from the input features for first-order difference operation to obtain the difference value of the mud level as a new feature to be added to the input feature of the deep learning model;

[0051] The longest mud discharge cycle in history is used as the sequence length, and zero-filling is performed on input features whose sequence length is less than the set length, thereby constructing input features in the form of multidimensional time series.

[0052] Input features in the form of multidimensional time series and the average sludge production as the corresponding output are input into the deep learning model in batches, and the mean square error is used as the optimization target. The parameters in the empirical model are solved by the Adam optimizer for at least one round of iteration to complete the training of the deep learning model.

[0053] The sludge production is predicted using the trained empirical model and deep learning model respectively, generating two sets of prediction results and combining them into a new feature combination set;

[0054] Pair the new feature combination set with the actual sludge production to form a new data set;

[0055] Select the linear regression model as the meta-model and use the new dataset to train the meta-model;

[0056] After the training is completed, the prediction results of the trained empirical model and the deep learning model are automatically weighted and combined through the meta-model to generate the final sludge production prediction result, so as to complete the coupling of the trained empirical model and the deep learning model and form a sedimentation tank sludge production prediction model.

[0057] Optionally, using a neural network model and a data set to construct and train a sediment distribution prediction model includes:

[0058] A neural network model is used to construct a sediment distribution prediction model;

[0059] The hourly mean of the water inlet conditions during the mud discharge period in the data set is used as input data, and the corresponding sediment distribution data is used as output data to obtain the data set of the sediment distribution prediction model;

[0060] Construct a loss function that considers the decreasing order penalty term;

[0061] The sediment distribution prediction model is trained based on the data set of the sediment distribution prediction model, and the parameters of the model are continuously adjusted during the training to minimize the loss function that takes into account the decreasing law;

[0062] After the training is completed, a sediment distribution prediction model is obtained that can predict sediment distribution according to real-time water inflow conditions;

[0063] Among them, the loss function considering the decreasing order penalty term is:

[0064] ;

[0065] In the formula, N train is the number of samples in the training set, and the first term is the standard mean square error MSE , y ij For the i The sample j The actual output value, For the i The sample j The predicted output value, λ is the weight coefficient used to balance the mean square error and the decreasing order penalty term, is the penalty term in descending order, k The order of the predicted values ​​output by the model.

[0066] Optionally, the sludge accumulation in each area of ​​the sedimentation tank is obtained according to the average sludge yield output by the sludge yield prediction model, the sludge distribution ratio in each area output by the bottom sediment distribution prediction model, and the total water inflow in the corresponding sludge discharge period, and the mud level height in each area of ​​the sedimentation tank is obtained in combination with the mud level data, including:

[0067] The sludge production in each area is obtained according to the average sludge production output by the sludge production prediction model, the sludge distribution ratio in each area output by the bottom sludge distribution prediction model, and the total water flow rate in the corresponding sludge discharge period;

[0068] For each area of ​​the sedimentation tank, the sludge accumulation in each area of ​​the sedimentation tank at any time is obtained according to the mud level height and sludge accumulation in each area of ​​the sedimentation tank at the initial time and the sludge production in each area;

[0069] According to the accumulated amount of sludge in each area of ​​the sedimentation tank at any time and the obtained mud level data, the mud level height of each area of ​​the sedimentation tank is obtained;

[0070] in,

[0071] The sludge production in each area is:

[0072] ;

[0073] In the formula, yes t 0~ t 0+△ t First i Sludge production in the area, is the average sludge production calculated by the sludge production prediction model within the △t time period, is the total water flow rate within the time △t, is the sludge proportion in each area calculated by the sediment distribution prediction model;

[0074] The accumulated amount of sludge in each area of ​​the sedimentation tank at any time is:

[0075] ;

[0076] In the formula, yes t 0+△ t Time sedimentation tank i The amount of sludge accumulated in the block area, is t 0 time sedimentation tank i The amount of sludge accumulated in a block area.

[0077] Optionally, based on the mud level heights of various areas of the sedimentation tank and the dynamically generated mud level threshold, outputting at least one mud discharge scheme includes:

[0078] Determine the lower and upper limits of the mud level in each area. The lower limit of the mud level ensures that the total volume of the sludge to be discharged is greater than the water discharge from the mud discharge pipe, and the upper limit of the mud level is based on experience or is set to twice the lower limit of the mud level;

[0079] Based on the mud level height of each area of ​​the sedimentation tank, the mud discharge demand of each area is judged. When the mud level height of a certain area exceeds its corresponding mud level lower limit, the area is added to the waiting mud discharge sequence;

[0080] Monitor the mud level changes in each area in the mud discharge sequence. When the mud level in any area in the sequence reaches its upper limit, the mud discharge action is triggered.

[0081] After the mud discharge action is triggered, the global mud discharge mode or the local mud discharge mode is determined according to the number of regions included in the mud discharge sequence and the size of the set number;

[0082] If the number of areas included in the mud discharge sequence is equal to the set number, the global mud discharge mode is adopted, and the mud discharge path for the mud suction vehicle to perform mud discharge in all areas of the sedimentation tank is planned;

[0083] If the number of areas included in the mud discharge sequence is less than or equal to the set number, a local mud discharge mode is adopted, and the mud suction vehicle is planned to discharge mud only in the areas included in the mud discharge sequence;

[0084] At least one mud discharge plan is generated based on the mud discharge triggering timing, mud discharge mode and mud discharge path.

[0085] In a second aspect, an embodiment of the present invention provides an intelligent sludge discharge control system for a sedimentation tank in a water plant, comprising:

[0086] A data set creation module is used to divide the collected mud discharge data into multiple segments or samples and obtain data sets after processing;

[0087] The sludge production prediction module is used to build an empirical model that maps the relationship between key influencing factors and sludge production and a deep learning model for learning the relationship between multiple influencing factors. The empirical model trained based on the data set and the deep learning model are coupled to build a sedimentation tank sludge production prediction model.

[0088] Sediment distribution simulation module, used to build and train sediment distribution prediction model using neural network model and data set;

[0089] The mud level height obtaining module is used to obtain the sludge accumulation in each area of ​​the sedimentation tank according to the average sludge production output by the sludge production prediction model, the sludge distribution ratio of each area output by the bottom sediment distribution prediction model, and the total water inflow of the corresponding sludge discharge period, and obtain the mud level height of each area of ​​the sedimentation tank in combination with the mud level data;

[0090] The mud discharge plan output module is used to output at least one mud discharge plan based on the mud level height of each area of ​​the sedimentation tank and the dynamically generated mud level threshold.

[0091] In a third aspect, an embodiment of the present invention provides an intelligent mud discharge control device for a sedimentation tank in a water plant, comprising:

[0092] at least one database;

[0093] and a memory in communication with the at least one database;

[0094] Wherein, the memory stores instructions that can be executed by the at least one database, and the instructions are executed by the at least one database so that the at least one database can execute the intelligent sludge discharge control method for the sedimentation tank of the water plant as described above.

[0095] In a fourth aspect, an embodiment of the present invention provides a computer-readable medium having computer executable instructions stored thereon, and when the executable instructions are executed by a processor, the intelligent sludge discharge control method for a sedimentation tank in a waterworks is implemented as described above.

[0096] (III) Beneficial effects

[0097] The beneficial effects of the present invention are:

[0098] First, the present invention divides the mud discharge data into multiple segments or samples, and obtains a data set after careful processing, which provides a solid foundation for building an efficient and accurate prediction model.

[0099] Next, the present invention is based on the mapping relationship from influent turbidity and chemical dosage to sludge production established by the empirical model, and couples the deep learning model to explore the nonlinear and long-term dependencies between multiple variables. The constructed sedimentation tank sludge production prediction model is not only based on traditional empirical knowledge, but also integrates advanced machine learning methods, thereby achieving accurate prediction of sludge production.

[0100] At the same time, the present invention establishes a refined mapping model from water inlet characteristics to bottom mud distribution, which can dynamically calculate the sludge generation and bottom mud distribution in the sedimentation tank according to the real-time water inlet conditions and the dosage of the reagent head, solving the problems of low sampling frequency, high labor intensity and inability to monitor in real time in traditional methods, and provides an important reference for subsequent mud discharge strategies.

[0101] Furthermore, the present invention can accurately calculate the sludge accumulation in each area of ​​the sedimentation tank by combining the average sludge production output by the sludge production prediction model, the sludge distribution ratio in each area output by the bottom sediment distribution prediction model, and the total water inflow corresponding to the sludge discharge cycle. Combined with the real-time mud level data, the mud level height in each area of ​​the sedimentation tank can be obtained, which provides a key reference indicator for mud discharge control.

[0102] Finally, based on the mud level height of each area of ​​the sedimentation tank and the dynamically generated mud level threshold, the output mud discharge plan of the present invention not only takes into account the sludge production and distribution, but also fully combines the real-time operating conditions, achieving a dual improvement in the flexibility and accuracy of the mud discharge strategy. This intelligent mud discharge control method not only improves the mud discharge efficiency, but also effectively reduces resource waste, making a positive contribution to the sustainable development of the water treatment industry.

[0103] Therefore, the application of this series of innovative methods and technical means of the present invention has significantly improved the intelligence level of sedimentation tank sludge discharge and brought substantial progress to the field of water treatment. BRIEF DESCRIPTION OF THE DRAWINGS

[0104] Figure 1 A schematic diagram of a flow chart of an intelligent sludge discharge control method for a water plant sedimentation tank provided by an embodiment of the present invention;

[0105] Figure 2 A specific flow chart of step S1 of a method for intelligent sludge discharge control of a sedimentation tank in a water plant provided by an embodiment of the present invention;

[0106] Figure 3 A specific flow chart of step S11 of a method for intelligent sludge discharge control of a sedimentation tank in a water plant provided by an embodiment of the present invention;

[0107] Figure 4 A specific flow chart of step S2 of a method for intelligent sludge discharge control of a sedimentation tank in a water plant provided by an embodiment of the present invention;

[0108] Figure 5 A specific flow chart of step S3 of a method for intelligent sludge discharge control of a sedimentation tank in a water plant provided by an embodiment of the present invention;

[0109] Figure 6 A specific flow chart of step S4 of a method for intelligent sludge discharge control of a sedimentation tank in a water plant provided by an embodiment of the present invention;

[0110] Figure 7 A specific flow chart of step S5 of a method for intelligent sludge discharge control of a sedimentation tank in a water plant provided by an embodiment of the present invention;

[0111] Figure 8 A schematic diagram of the overall process of an intelligent sludge discharge control method for a sedimentation tank in a water plant provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0112] In order to better explain the present invention and facilitate understanding, the present invention is described in detail below through specific implementation modes in conjunction with the accompanying drawings.

[0113] like Figure 1As shown, an intelligent sludge discharge control method for a water plant sedimentation tank proposed in an embodiment of the present invention includes: dividing the collected sludge discharge data into multiple segments or samples, and obtaining a data set after processing; constructing an empirical model that maps the relationship between key influencing factors and sludge production and a deep learning model for learning the relationship between multiple influencing factors, and coupling the empirical model and the deep learning model trained based on the data set to construct a sludge production prediction model for the sedimentation tank; using a neural network model and a data set to construct and train a sediment distribution prediction model; according to the average sludge production output by the sludge production prediction model, the sludge distribution ratio of each area output by the sediment distribution prediction model, and the total water inlet flow rate corresponding to the sludge discharge period, the sludge accumulation in each area of ​​the sedimentation tank is obtained, and the mud level height in each area of ​​the sedimentation tank is obtained in combination with the mud level data; based on the mud level height in each area of ​​the sedimentation tank and the dynamically generated mud level threshold, at least one sludge discharge plan is output.

[0114] First, the present invention divides the mud discharge data into multiple segments or samples, and obtains a data set after careful processing, which provides a solid foundation for building an efficient and accurate prediction model.

[0115] Next, the present invention is based on the mapping relationship from influent turbidity and chemical dosage to sludge production established by the empirical model, and couples the deep learning model to explore the nonlinear and long-term dependencies between multiple variables. The constructed sedimentation tank sludge production prediction model is not only based on traditional empirical knowledge, but also integrates advanced machine learning methods, thereby achieving accurate prediction of sludge production.

[0116] At the same time, the present invention establishes a refined mapping model from water inlet characteristics to bottom mud distribution, which can dynamically calculate the sludge generation and bottom mud distribution in the sedimentation tank according to the real-time water inlet conditions and the dosage of the reagent head, solving the problems of low sampling frequency, high labor intensity and inability to monitor in real time in traditional methods, and provides an important reference for subsequent mud discharge strategies.

[0117] Furthermore, the present invention can accurately calculate the sludge accumulation in each area of ​​the sedimentation tank by combining the average sludge production output by the sludge production prediction model, the sludge distribution ratio in each area output by the bottom sediment distribution prediction model, and the total water inflow corresponding to the sludge discharge cycle. Combined with the real-time mud level data, the mud level height in each area of ​​the sedimentation tank can be obtained, which provides a key reference indicator for mud discharge control.

[0118] Finally, based on the mud level height of each area of ​​the sedimentation tank and the dynamically generated mud level threshold, the output mud discharge plan of the present invention not only takes into account the sludge production and distribution, but also fully combines the real-time operating conditions, achieving a dual improvement in the flexibility and accuracy of the mud discharge strategy. This intelligent mud discharge control method not only improves the mud discharge efficiency, but also effectively reduces resource waste, making a positive contribution to the sustainable development of the water treatment industry.

[0119] Therefore, the application of this series of innovative methods and technical means of the present invention has significantly improved the intelligence level of sedimentation tank sludge discharge and brought substantial progress to the field of water treatment.

[0120] In order to better understand the above technical solution, exemplary embodiments of the present invention will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of the present invention are shown in the accompanying drawings, it should be understood that the present invention can be implemented in various forms and should not be limited by the embodiments described herein. On the contrary, these embodiments are provided to enable a clearer and more thorough understanding of the present invention and to fully convey the scope of the present invention to those skilled in the art.

[0121] Specifically, an embodiment of the present invention provides an intelligent sludge discharge control method for a sedimentation tank in a water plant, comprising:

[0122] S1. Divide the collected mud discharge data into multiple segments or samples, and process them to obtain a data set.

[0123] The present invention relates to the prediction of sedimentation tank sludge production and simulation of sediment distribution, and it is necessary to collect historical data of relevant water quality, water volume, sludge concentration and other indicators for model training. Specifically, historical data of water plant monitoring indicators such as inlet turbidity, outlet turbidity, inlet flow, flocculant dosage, coagulant aid dosage, water temperature, sedimentation tank liquid level, sedimentation tank mud level, sludge concentration in sludge pipe, sludge pipe flow, sludge tank liquid level, sludge concentration in sludge tank and so on are collected for at least one year, with a frequency of at least one data every ten minutes. The higher the frequency, the more detailed sludge discharge information can be provided.

[0124] According to the start-up status of the sludge suction vehicle, the sludge discharge cycle of the sedimentation tank is divided into sludge discharge periods ( T RS ) and non-sludge discharge period ( T NS ), the whole period from the end of the last mud discharge to the end of this mud discharge is recorded as a mud discharge cycle, then the duration of the mud discharge cycle is In normal production, the horizontal flow sedimentation tank operates continuously, and the sludge is produced throughout the entire sludge discharge cycle. During the sludge discharge period, the sludge is discharged through the sludge discharge pipe on the sludge suction vehicle by siphon or pump suction, and then transported to the sludge discharge pool through the sludge discharge channel (usually open), and then temporarily stored in the sludge discharge pool and transported to the subsequent sludge concentration and treatment section.

[0125] Among the above monitoring indicators, the inlet turbidity, outlet turbidity, inlet flow, flocculant dosage, coagulant aid dosage, water temperature, sedimentation tank level, and sedimentation tank mud level are continuously monitored throughout the sludge discharge cycle, and will be used as input features for the sludge production prediction model and the bottom sludge distribution prediction model. The sludge tank level and sludge tank sludge concentration are collected during the sludge discharge period and are used to calculate the total amount of sludge discharged during the sludge discharge period. Under ideal conditions (i.e., the sludge discharge is thorough and there is no sludge remaining in the tank after the sludge discharge), it should be equal to the total amount of sludge produced during the sludge discharge cycle. The sludge concentration and flow rate of the sludge discharge pipe are also collected during the sludge discharge period, and they will be used to calculate the sludge distribution ratio along the length of the sedimentation tank.

[0126] The following will introduce in detail the methods of calculating sludge production, sediment distribution, and data set construction.

[0127] Furthermore, if Figure 2 As shown, step S1 includes:

[0128] S11. Based on the collected sludge discharge information, the average sludge production in a single sludge discharge cycle and the proportional distribution of sludge in each area during each sludge discharge period are calculated.

[0129] Furthermore, if Figure 3 As shown, step S11 includes:

[0130] S111. At the beginning and end of each sludge discharge cycle, the amount of dry sludge discharged during the sludge discharge period is calculated based on the collected liquid level height of the sludge discharge pool before and after sludge discharge, the average sludge concentration in the sludge discharge pool before and after sludge discharge, the bottom area of ​​the sludge discharge pool, the flow rate discharged from the sludge discharge pool, and the duration of sludge discharge after the sludge discharge pool reaches the upper limit of capacity.

[0131] S112. Calculate the average sludge production in a single sludge discharge cycle based on the amount of dry sludge discharged during the sludge discharge period and the cumulative water intake during the corresponding sludge discharge cycle.

[0132] S113. Determine the length of the sedimentation tank, the speed of the sludge suction vehicle and the sludge discharge period.

[0133] S114. When the sludge suction vehicle starts to discharge sludge at the traveling speed, the sludge flow rate and sludge concentration data corresponding to each position during the driving process of the vehicle and the time for the vehicle to pass each position are recorded.

[0134] S115, calculating the sludge discharge volume of the sludge suction vehicle at each position according to the sludge flow rate and sludge concentration data corresponding to each position during the driving process of the vehicle and the time the vehicle passes through each position.

[0135] S116. Based on the sludge discharge volume of the sludge suction vehicle at any position, the distribution of the sludge at any position as a percentage of the total sludge production of the sedimentation tank is calculated.

[0136] S117, dividing the sedimentation tank into a plurality of unit areas, and dividing the sludge discharge period into a plurality of time periods corresponding to the number of unit areas, and calculating the proportion of sludge production in each unit area or each time period.

[0137] In a specific embodiment, there is currently no reliable online monitoring instrument for directly monitoring sludge production. Therefore, it is necessary to indirectly calculate the sludge production in each sludge discharge cycle based on the changes in the liquid level, flow rate and sludge concentration in the sludge discharge pool before and after sludge discharge. The amount of dry sludge discharged during the sludge discharge period is:

[0138] ;

[0139] In the formula, S total is the amount of dry sludge discharged during the sludge discharge period, that is, the total sludge production of this sludge discharge cycle. H 1 and H 0 are the liquid level heights of the mud discharge pool before and after mud discharge, SS 1 and SS 0 are the average sludge concentrations in the sludge pool before and after sludge discharge, A is the bottom area of ​​the sludge pool, Q 1 is the flow rate discharged from the sludge pool, T 1 is the duration of sludge discharge from the sludge pool after reaching its upper capacity limit.

[0140] The average mud production in a single mud discharge cycle is:

[0141] ;

[0142] In the formula, S is the average production of a single sludge discharge cycle, Indicates the cumulative water intake during this sludge discharge cycle. Q t yes t The water flow rate of the sedimentation tank at any time, T total The duration of this mud discharge cycle.

[0143] Next, the sedimentation tank sludge distribution is calculated: Assuming that the sludge was emptied in the previous sludge discharge cycle, the sludge is discharged from the sludge discharge pipe during this sludge discharge process. The flow meter and sludge concentration meter installed at the outlet of the sludge discharge pipe can be used to calculate the location of the sludge suction vehicle (denoted as x ) of sludge production ( S x ). As the vehicle travels through the entire sedimentation tank (length is recorded asL ), then according to the different positions S x The sludge distribution along the length of the entire sedimentation tank is obtained. The sludge suction vehicle usually runs at a constant speed. Assuming that the vehicle travels at a speed of v 0, then v 0= L / T RS Assume that the vehicle is driving t At time 0, from the water inlet of the sedimentation tank x 0 starts to move towards the end of the sedimentation tank, discharging mud while moving forward. After a period of mud discharge, T RS Reach the end x L During the return journey, no mud is discharged, which is also called one-way mud discharge.

[0144] Suction sludge crane at any position x The amount of sludge discharged at is:

[0145] S x = SS t × Q t × dt ;

[0146] In the formula, S x for t Driving to location at all times x The amount of sludge discharged is the sludge production. SS t Respectively Q t It is the sludge concentration and sludge flow rate measured by the sensor on the sludge discharge pipe at this position. dt For driving through x The time spent at the location, where , t Moment and x One to one correspondence.

[0147] It can be obtained that the sludge distribution of any position in the total sludge production of the sedimentation tank is:

[0148] ;

[0149] In the formula, Indicates location x The proportion of sludge produced in the sedimentation tank to the total sludge produced in the sedimentation tank.

[0150] In order to facilitate modeling and calculation, the above distribution is discretized, that is, the sedimentation tank is divided into equal areas along its length. MThe corresponding mud discharge period T RS Divide into equal length M The above formula can be converted into the proportion of mud production in each unit area or each time period:

[0151] ;

[0152] In the formula, and For driving in i Average sludge concentration and average flow rate per unit area / time period, M is the number of units divided. The time part in the numerator and denominator After reduction, M Take 10.

[0153] After calculation, the proportion of sludge distribution in each area during each sludge discharge period is finally obtained. =( y 1, y 2, ..., y M ),in i ∈(1, M ), y i Indicates the number of steps from the inlet to the outlet of the sedimentation tank. i The proportion of mud production in the corresponding section area to the total mud production, y i Satisfying the normalization requirement is , and usually with i The increase is decreasing.

[0154] In the present invention n Take 10, then y 1, y 2, ..., y 10 They are the proportion of sludge in the first 10%, 10%~20%,…, 90%~100% of the length of the sedimentation tank to the total sludge volume.

[0155] S12. Determine the monitoring index data of interest, including the hourly average values ​​of inlet turbidity, outlet turbidity, inlet flow, flocculant dosage, coagulant aid dosage, water temperature, sedimentation tank liquid level, and sedimentation tank mud level.

[0156] S13. Traverse the collected mud discharge data, and divide the mud discharge data into multiple segments or samples of different lengths according to the mud discharge status, including at least one set of monitoring indicator data concerned in the mud discharge cycle, the average mud production in a single mud discharge cycle, and the distribution ratio of the sedimentation tank bottom mud.

[0157] S14. Down-sample the monitoring indicator data of interest to obtain a new input sequence.

[0158] S15. A data set is formed according to the new input sequence, the average sludge production in a single sludge discharge cycle, and the distribution ratio of the sedimentation tank bottom sludge, and the data set is divided into a training set and a test set according to a set ratio.

[0159] Specifically, the historical data set is traversed and all data are divided into segments of varying lengths according to the mud discharge status. Each segment corresponds to a set of mud discharge cycles. Group fragments or samples, each sample can be represented as For the sample i , and the corresponding average sludge production in the sludge discharge cycle is S i Sediment distribution ratio , which will be used as the output of the sludge production prediction model and the sediment distribution prediction model respectively. X i It is the hourly average value of 8 monitoring indicators including inlet turbidity, outlet turbidity, inlet flow, flocculant dosage, coagulant aid dosage, water temperature, sedimentation tank liquid level and sedimentation tank mud level during this sludge discharge cycle.

[0160] Since the original data is one record every 10 minutes, we downsample the original data to one record per hour. D i The original input feature matrix For example, the original data can be expressed as ,in x i is a vector of length 8, N is the number of original data of the sample. Then the new input sequence obtained after downsampling can be expressed as:

[0161] ;

[0162] In the formula, X i It represents the new input sequence obtained after downsampling, in which each data is the average of the six original data in its hour interval. is the number of processed data. Indicates rounding down.

[0163] Based on the above processing, we finally get a complete data set , the input feature matrix for each sample X i Indicates monitoring indicators such as hourly water inlet conditions during the sludge discharge cycle. S iIt indicates the average mud production in this mud discharge cycle. Indicates the distribution of sediment.

[0164] Finally, the data set is divided into training set and training set in a ratio of 8:2. and test set , used for model training and performance verification respectively.

[0165] S2. Construct an empirical model that maps the relationship between key influencing factors and sludge production and a deep learning model for learning the relationship between multiple influencing factors, and couple the empirical model and deep learning model trained based on the data set to construct a sedimentation tank sludge production prediction model.

[0166] In step S2, a sedimentation tank sludge production prediction model is constructed based on historical data, and the sludge production is calculated in real time according to the influent conditions and drug addition conditions of the sedimentation tank. Existing methods include empirical models based on the law of conservation of mass, and deep learning models based on big data. The empirical model relies on accurate metering parameters and cannot be dynamically adjusted according to the influent conditions; the deep learning model discovers rules from a large amount of historical data, but due to the long sludge discharge cycle, the number of effective samples is small and the model cannot be fully trained. Therefore, the present invention is based on the mapping relationship from influent turbidity and drug dosage to sludge production established by the empirical model, and couples the deep learning model to explore the nonlinear and long-term dependencies between multiple variables to achieve accurate prediction of sludge production.

[0167] Furthermore, if Figure 4 As shown, step S2 includes:

[0168] S21. The key factors affecting sludge production include influent turbidity and chemical dosage.

[0169] S22. Based on the law of conservation of mass, an empirical model is constructed to map the relationship between key influencing factors and sludge production.

[0170] S23. Based on the data set, taking the mean square error as the optimization target, the various parameters in the empirical model are iteratively solved for at least one round by the gradient descent method to complete the training of the empirical model.

[0171] In a specific embodiment, according to the law of conservation of mass, the sedimentation tank sludge mainly comes from the particulate matter in the raw water and the added flocculant. In addition, it is also related to other agents such as soluble substances, coagulants, activated carbon, etc. in the raw water. When aluminum salt is used as a coagulant, the dry sludge output can be calculated according to the following formula:

[0172] ;

[0173] In the formula, is the dry sludge output in a single sludge discharge cycle, in units of t / d , The cumulative water flow rate of the sedimentation tank during this sludge discharge cycle is: Tu is the average raw water turbidity during the sludge discharge period, E 1 is turbidity and SS The conversion factor, E 2 is the conversion coefficient of aluminum salt coagulant and Al(OH)3, Al is the average dosage of aluminum salt coagulant (in Al 2 O 3), B is the influence of other factors in the water treatment process.

[0174] It should be noted that the model input features in the empirical model Tu , Al is the average value within the mud discharge cycle. , X i Influent turbidity Tu , Flocculant dosage Al The requirements are averaged and then used for empirical model training. X i The corresponding sludge discharge cycle is T Hours, then ,in , Respectively represent i The first j Average influent turbidity and flocculant dosage per hour.

[0175] The average sludge production during this sludge discharge cycle is:

[0176] ;

[0177] The coefficients in the formula are: E 1, E 2, B It needs to be calibrated by the historical data in the data set. The specific method is to use the mean square error (MSE) between the empirical model prediction value and the actual value as the optimization target, and calculate each coefficient through the gradient descent method in the entire training set. The calculation process of MSE is as follows:

[0178] ;

[0179] in, L is the mean square error, N train is the number of samples in the training set, For the i The actual sludge production of each sample, , It is i The average influent turbidity and average dosage of aluminum salt coagulant for each sample.

[0180] First determine E 1, E 2, B The initial value of can be set based on human experience to accelerate model convergence, for example E 1 out of 2.2, E 2 takes 1.53, B Take 0.2. Then calculate the loss function in each round L For each parameter E 1, E 2, B The partial derivative of is used to guide the direction and step size of gradient descent.

[0181]

[0182] In the formula, , , Respectively represent the loss function E 1, E 2, B The partial derivative of .

[0183] Then according to the calculated gradient, combined with the learning rate h Calculate the step size of the gradient descent for this round, for each parameter E 1, E 2, B to update.

[0184]

[0185] In the formula, h represents the learning rate, , , Represents the new coefficients obtained after each round of iteration. Through repeated iterations, when the absolute value of the gradient is less than a certain threshold or the decrease in the loss function is less than a certain threshold, the iteration is stopped to obtain the optimal model parameters, which will be used for the construction of subsequent hybrid models.

[0186] S24. A long short-term memory network is used to construct a deep learning model including an input layer, a mask layer, two LSTM layers, and two fully connected layers.

[0187] S25. In the data set, hourly inlet turbidity, outlet turbidity, inlet flow, dosage, water temperature, sedimentation tank liquid level and replenished sedimentation tank mud level are selected as input features of the deep learning model, and the sedimentation tank mud level is selected from the input features for first-order difference operation to obtain the difference value of the mud level as a new feature to be added to the input feature of the deep learning model.

[0188] S26. The length of the longest mud discharge cycle in history is used as the sequence length, and the input features whose sequence length is less than the set length are filled with zero values, so as to construct input features in the form of multidimensional time series.

[0189] S27. Input features in the form of multidimensional time series and the average sludge production as the corresponding output are input into the deep learning model in batches, and the mean square error is used as the optimization target. The parameters in the empirical model are iteratively solved for at least one round through the Adam optimizer to complete the training of the deep learning model.

[0190] S28. Use the trained empirical model and deep learning model to predict the sludge production respectively, generate two sets of prediction results and combine them into a new feature combination set; pair the new feature combination set with the actual sludge production to form a new data set.

[0191] S29. Select the linear regression model as the meta-model and use the new data set to train the meta-model. After the training is completed, the prediction results of the trained empirical model and the deep learning model are automatically weighted and combined through the meta-model to generate the final sludge production prediction result, so as to complete the coupling of the trained empirical model and the deep learning model and form a sedimentation tank sludge production prediction model.

[0192] In a specific embodiment, the aforementioned empirical model only considers the contribution of raw water turbidity and dosage, and does not consider the influence of seasonal changes, water flow characteristics, iron, manganese, dissolved solids and other factors in raw water on sludge production. Therefore, a deep learning model is used to learn the nonlinear and long-term dependencies between multiple variables from a large amount of historical data to improve the accuracy of the prediction model. In the present invention, a long short-term memory network (LSTM) is used to predict sludge production. The model output is the average sludge production in a single sludge discharge cycle, and the model input is 8 characteristics corresponding to the hourly influent turbidity, effluent turbidity, influent flow, flocculant dosage, coagulant aid dosage, water temperature, sedimentation tank level, and sedimentation tank mud level in the sludge discharge cycle.

[0193] In order to improve the prediction effect and generalization ability of the model, the first-order difference of the sedimentation tank mud level is performed to obtain a new feature of mud level difference value to help the model learn the trend information of sludge generation. The first-order difference operation can be expressed as: ,in LS t It's time pointt The mud level value, is the data after difference.

[0194] The difference value of the mud level is added as a new feature to the LSTM model input feature, and the final input of the LSTM model has 9 features.

[0195] It is important to understand that the input to LSTM is in the form of a multidimensional time series, that is, the shape is A matrix, where each row is a time point and each column is a feature or variable. Since the mud discharge cycle may be different, the length of the original sequence of different samples is uncertain. In this invention, the longest mud discharge cycle in history, 36 hours, is selected as the sequence length, that is, For shorter sequences, zero values ​​are added at the end of the sequence for padding.

[0196] According to the above method, the historical data is transformed to obtain the shape of A data set, where Represents the number of samples in the data set, that is, the number of mud discharge cycles contained in the historical data. .

[0197] The LSTM in the present invention is built based on the Pytorch framework, and the model includes an input layer, a mask layer, two LSTM layers, and two fully connected layers. The input layer accepts a 36´9 multidimensional time series and passes it to subsequent layers. The mask layer is used to ignore the data of the padding part to ensure that the model only focuses on the actual data rather than the padding value during training and prediction. The first LSTM layer receives the output of the mask layer and processes the time series data through its built-in memory and gating mechanism. The second LSTM layer further extracts higher-level time series features. The first fully connected layer accepts the output of the LSTM layer and performs feature conversion. The second fully connected layer serves as the output layer to generate the final prediction result. The number of neurons in each layer can be determined according to the actual data set. The hyperparameter combination listed here is only an example: the number of neurons in the first LSTM layer is 32, the number of neurons in the second LSTM layer is 64, the number of neurons in the first fully connected layer is 64, and the number of neurons in the second fully connected layer is 1.

[0198] The multi-dimensional time series input and the corresponding output (average sludge production) constructed above are input into the LSTM model in batches, and the mean square error (MSE) between the model prediction value and the actual value is calculated, and the model weight parameters are updated based on the Adam optimizer, a variant of the gradient descent method. After multiple iterations, the final LSTM model is obtained for subsequent hybrid model construction.

[0199] In order to improve the prediction effect and enhance the robustness and generalization ability of the model, the present invention is based on the Stacking algorithm coupled with the above empirical model With deep learning models Compared with methods such as averaging and residual fitting, the Stacking algorithm uses a new model (meta-model) to integrate the prediction results of multiple models (basic models), which can capture different aspects of the data and has high flexibility and robustness.

[0200] In the present invention, the basic model is the one completed by the aforementioned training and , the meta-model selects the linear regression model, denoted as . by and The prediction result is used as the new input feature and the actual sludge production is used as the output value. and The prediction results of are combined into the final prediction result. ,in X i In the form of a multidimensional time series, and The sludge production is predicted respectively, and the prediction results are obtained respectively. and .right For , its input features are and The new feature combination composed of , then the data set sample can be expressed as , i∈[1, N samples] . Use the new training set to That is, the linear regression model is trained to obtain the optimal meta-model, which can automatically perform combined weighting according to the prediction results of the basic model to calculate the final sludge production.

[0201] S3. Use the neural network model and data set to build and train a sediment distribution prediction model.

[0202] Furthermore, if Figure 5 As shown, step S3 includes:

[0203] S31. Use neural network model to build sediment distribution prediction model.

[0204] S32. Using the hourly average of the water inlet conditions during the mud discharge cycle in the data set as input data and the corresponding sediment distribution data as output data, to obtain a data set of a sediment distribution prediction model.

[0205] S33. Construct a loss function that takes into account the decreasing order penalty terms.

[0206] S34. The sediment distribution prediction model is trained based on the data set of the sediment distribution prediction model, and the parameters of the model are continuously adjusted during the training to minimize the loss function that takes into account the decreasing law.

[0207] S35. After the training is completed, a sediment distribution prediction model is obtained that can predict sediment distribution according to real-time water inflow conditions.

[0208] The traditional method is usually to manually measure the mud level at different positions in the sedimentation tank and then draw the mud accumulation curve, but this method is labor-intensive and has a low sampling frequency, and cannot be used for online sediment distribution simulation. In fact, sediment distribution is also related to the season and influent conditions. For example, in summer, the turbidity of raw water is high, the content of high-density pollutants in the water is high, and the corresponding coagulant dosage also increases. Therefore, the proportion of sludge in the front section of the sedimentation tank will be greater in all sludge. Therefore, it is necessary to establish a mapping from influent water quality to sediment distribution ratio, so as to support the dynamic simulation of sediment distribution in the sedimentation tank.

[0209] The present invention uses a neural network model to construct a sediment distribution prediction model. The input of the model is the same as the empirical model, which is the hourly mean of the water inlet conditions during the mud discharge cycle. X i The output of the model is obtained by the above calculation .

[0210] In the model design, a neural network model with two hidden layers is used, the input shape is (9,1), the number of neurons in the two hidden layers is set to 64, and the output layer activation function selects softmax to ensure that the sum of all outputs is 1. In addition, considering that the sludge distribution in the sedimentation tank shows a decreasing pattern along the way, ( y 1, y 2, ..., y 10 ) should also show a decreasing law. Therefore, the present invention specially designs the loss function and adds an additional penalty term on the basis of the general MSE, as described in the formula:

[0211] ;

[0212] In the formula, when , then a penalty is incurred, otherwise the penalty is 0.

[0213] Then the loss function considering the decreasing order penalty term is:

[0214] ;

[0215] In the formula, N train is the number of training set samples, and 10 is the dimension of the model output, i.e. the number of corresponding divided units. MWhen the value is 10, the first term is the standard mean square error. MSE , y ij It is i The sample j The actual output value, It is i The sample j The predicted output value, λ is the weight coefficient used to balance the mean square error and the decreasing order penalty term, is the penalty term in descending order, k is the order of the predicted values ​​output by the model.

[0216] S4. According to the average sludge production output by the sludge production prediction model, the sludge distribution ratio of each area output by the bottom sediment distribution prediction model and the total water inflow of the corresponding sludge discharge period, the sludge accumulation in each area of ​​the sedimentation tank is obtained, and the mud level height of each area of ​​the sedimentation tank is obtained in combination with the mud level data.

[0217] Furthermore, if Figure 6 As shown, step S4 includes:

[0218] S41. Obtain the sludge production in each region according to the average sludge production output by the sludge production prediction model, the sludge distribution ratio in each region output by the bottom sediment distribution prediction model, and the total water inflow in the corresponding sludge discharge period.

[0219] S42. For each area of ​​the sedimentation tank, the accumulated sludge amount of each area of ​​the sedimentation tank at any time is obtained according to the mud level height and accumulated sludge amount of each area of ​​the sedimentation tank at the initial time and the sludge output of each area.

[0220] S43. Obtain the mud level height of each area of ​​the sedimentation tank according to the accumulated amount of sludge in each area of ​​the sedimentation tank at any time and the acquired mud level data.

[0221] The trained sediment distribution prediction model can calculate the real-time sludge distribution ratio according to the real-time water inlet conditions. Combined with the existing sludge distribution information and sludge production prediction results of the sedimentation tank, the current sediment distribution of the sedimentation tank can be obtained. t At time 0, the sedimentation tank is i The sludge accumulation in the block area is , i =0,1,…,10, the accumulated sludge at each position of the sedimentation tank after Δt time can be expressed as:

[0222] ;

[0223] In the formula, yes Time sedimentation tank iThe amount of sludge accumulated in the block area, is t 0 time sedimentation tank i The amount of sludge accumulated in the block area, yes First i The sludge production of a block area can be calculated according to the following formula:

[0224] ;

[0225] In the formula, for t 0~ t 0+△ t First i Sludge production in the area, is the average sludge production calculated by the sludge production prediction model within the △t time period, is the total water flow rate within the time △t, It is the sludge proportion in each area calculated by the sediment distribution prediction model.

[0226] Based on the above calculation, the accumulated sludge volume in each area of ​​the bottom of the sedimentation tank at any time can be obtained. Further combined with the mud level data, the mud accumulation height at any position of the sedimentation tank can be obtained. k ( k ∈[1,10]) The middle position of the block area, its data is k The average mud level in the area. t The mud level at that moment is recorded as , then the mud level height in each area of ​​the sedimentation tank can be expressed as:

[0227] ;

[0228] In the formula, express t Moment i The average mud level of the block area, express t The amount of sludge accumulated in the area at the moment.

[0229] S5. Output at least one mud discharge plan based on the mud level height of each area of ​​the sedimentation tank and the dynamically generated mud level threshold.

[0230] Based on the aforementioned sludge production prediction model and sediment distribution prediction model, the real-time sludge production and sediment distribution of the sedimentation tank are obtained, thereby determining the overall and local sludge discharge requirements of the sedimentation tank, and then making intelligent recommendations on the timing and path of sludge discharge.

[0231] Furthermore, if Figure 7 As shown, step S5 includes:

[0232] S51. Determine the lower limit and upper limit of the mud level in each area, wherein the lower limit of the mud level ensures that the total volume of the sludge to be discharged is greater than the amount of water discharged from the mud pipe, and the upper limit of the mud level is based on experience or is set to twice the lower limit of the mud level.

[0233] S52. Based on the mud level height of each area of ​​the sedimentation tank, the mud discharge demand of each area is judged. When the mud level height of a certain area exceeds its corresponding mud level lower limit, the area is added to the waiting mud discharge sequence.

[0234] S53, monitoring the change of mud level in each area in the mud discharge sequence, and triggering mud discharge when the mud level in any area in the sequence reaches its upper limit.

[0235] S54, after the mud discharge action is triggered, determine whether to adopt a global mud discharge mode or a local mud discharge mode according to the number of areas included in the mud discharge sequence and the size of the set number.

[0236] S55. If the number of areas included in the mud discharge sequence is equal to the set number, a global mud discharge mode is adopted, and a mud discharge path is planned for the mud suction vehicle to perform mud discharge in all areas of the sedimentation tank.

[0237] S56. If the number of areas included in the mud discharge sequence is less than or equal to the set number, a local mud discharge mode is adopted, and a mud discharge path is planned in which the mud suction vehicle discharges mud only in the areas included in the mud discharge sequence.

[0238] S57. Generate at least one mud discharge plan based on the mud discharge triggering timing, mud discharge mode and mud discharge path.

[0239] In a specific embodiment, for the mud suction vehicle, its forward speed is constant (denoted as v 0), the flow rate of the mud discharge pipe is usually a constant value (denoted as Q 0). Any position in the sedimentation tank x Department, Chief Δx , width is the width of the sedimentation tank w The rectangular area, assuming that the mud height at this location is LS x , then the total volume of sludge to be discharged in the rectangular area is , the amount of sludge water discharged when vehicles pass through this area can be expressed as In order to increase the sludge concentration of the sludge water as much as possible and reduce the sludge water content, it is necessary to ensure that the total volume of the sludge to be discharged is greater than the amount of water discharged from the sludge pipe, so as to obtain the lower limit of the sludge level:

[0240]

[0241] In the formula, L Smin Indicates the minimum mud level required for mud discharge in a specific area.Q 0 is the flow rate of the mud discharge pipe, w is the width of the sedimentation tank, v 0 is the forward speed of the vehicle.

[0242] On the other hand, when the sludge accumulates too much and causes the mud level to rise, it will compress the sedimentation area, resulting in a decrease in sedimentation efficiency and deterioration of effluent. In severe cases, it may even form a sludge stagnation area at the bottom of the sedimentation tank that is difficult to remove. Therefore, the upper limit of the mud level in the sedimentation tank is set. LS max , to prevent excessive mud accumulation, its size can be set based on experience, or for simplicity it can be set to LS min twice to ensure that the sludge can be discharged in time through double-pass sludge discharge).

[0243] Based on the above-mentioned sediment distribution simulation module, the real-time mud level height of each area in the sedimentation tank is obtained to determine the mud discharge demand of each area. Exceeding the lower limit of mud level L Smin When the mud is discharged in the next period of time, it indicates that the mud will be discharged in the area. i ( i ∈[1,10]) enters the waiting sludge discharge sequence Π. As time goes by, the amount of sludge in each area continues to increase, and new areas continue to enter the sludge discharge sequence Π until the mud level in a certain area in the sludge discharge sequence reaches the upper limit of the mud level. LS max When , the mud discharge action is triggered. The judgment condition can be expressed as:

[0244] ;

[0245] In the formula, max (.) indicates taking the maximum value, and Π indicates the mud discharge sequence.

[0246] After the mud discharge action is triggered, count the number of areas included in the mud discharge sequence N Π , determine whether to use global mud discharge or local mud discharge. Global mud discharge refers to mud discharge in all areas of the sedimentation tank, and local mud discharge only discharges mud in the areas included in the mud discharge sequence Π. The specific mud discharge rules are defined as follows:

[0247] like N Π =10, global mud removal is adopted.

[0248] like N Π ≤10, it is necessary to further determine the mud level height in the non-mud discharge sequence area: if the areas in the non-mud discharge sequence all meet , a global mud discharge method is adopted, otherwise a local mud discharge action is performed.

[0249] After the above calculations, the final mud discharge plan is obtained, including key parameters such as mud discharge timing, mud discharge path, and mud discharge mode.

[0250] In addition, an embodiment of the present invention provides an intelligent mud discharge control system for a sedimentation tank in a water plant, comprising:

[0251] The data set creation module is used to divide the collected mud discharge data into multiple segments or samples and obtain data sets through processing.

[0252] The sludge production prediction module is used to build an empirical model that maps the relationship between key influencing factors and sludge production and a deep learning model for learning the relationship between multiple influencing factors. The empirical model and the deep learning model trained based on the data set are coupled to build a sedimentation tank sludge production prediction model.

[0253] The sediment distribution simulation module is used to build and train a sediment distribution prediction model using a neural network model and data set.

[0254] The mud level height calculation module is used to obtain the sludge accumulation in each area of ​​the sedimentation tank based on the average sludge production output by the sludge production prediction model, the sludge distribution ratio in each area output by the bottom sediment distribution prediction model, and the total water inflow flow in the corresponding sludge discharge period, and obtain the mud level height in each area of ​​the sedimentation tank in combination with the mud level data.

[0255] The mud discharge plan output module is used to output at least one mud discharge plan based on the mud level height of each area of ​​the sedimentation tank and the dynamically generated mud level threshold.

[0256] Furthermore, the present invention provides an intelligent sludge discharge control device for a sedimentation tank in a waterworks, comprising: at least one database; and a memory communicatively connected to the at least one database; wherein the memory stores instructions executable by the at least one database, and the instructions are executed by the at least one database so that the at least one database can execute the intelligent sludge discharge control method for the sedimentation tank in the waterworks as described above.

[0257] The database is used to store and manage various data related to sedimentation tank sludge control, including but not limited to real-time and historical data such as influent turbidity, dosage of chemical, sludge production, and bottom sludge distribution. These data are the basis for implementing intelligent sludge control methods.

[0258] The memory stores an instruction set that can be executed by at least one database. When these instructions are executed by the database, the database can execute the intelligent sludge discharge control method for the sedimentation tank of the water plant as described above according to the preset logic and algorithm. Specifically, by establishing and running a sludge production prediction model and a bottom sediment distribution prediction model, accurate prediction of sludge production and real-time simulation of bottom sediment distribution can be achieved, and then the timing, path and mode of sludge discharge can be intelligently determined to achieve efficient and energy-saving sludge discharge control.

[0259] Furthermore, an embodiment of the present invention provides a computer-readable medium having computer executable instructions stored thereon, and when the executable instructions are executed by a processor, the intelligent sludge discharge control method for a sedimentation tank in a waterworks is implemented as described above.

[0260] Specifically, when these instructions are executed by the processor, they will guide the processor to perform a series of complex calculations and operations, including but not limited to data collection, processing, analysis, model establishment, training, operation, and formulation, optimization, and execution of sludge discharge strategies. Through the execution of these instructions, the processor can accurately predict sludge production and bottom sludge distribution, and intelligently determine the timing, path, and mode of sludge discharge, thereby achieving comprehensive, efficient, and intelligent control of the sludge discharge process of the water plant sedimentation tank.

[0261] Therefore, the computer-readable medium provided by the present invention not only provides strong technical support for the implementation of the intelligent sludge discharge control method of the sedimentation tank of the water plant, but also provides a powerful tool guarantee for the intelligent, efficient and refined operation and management of the water plant.

[0262] In summary, the embodiments of the present invention provide a method, system, device and medium for intelligent sludge discharge control of a water plant sedimentation tank. The core of the present invention is to accurately predict sludge production by combining empirical models with deep learning models. Specifically, Figure 8 As shown in the figure, the mapping relationship between influent turbidity, dosage of chemical and sludge production is first established through the empirical model. In order to explore the nonlinear and long-term dependence between these variables, the Stacking algorithm is further used to couple the empirical model with the deep learning model, which not only significantly improves the prediction accuracy, but also reduces the dependence on high-quality training samples and can dynamically adapt to the real-time changes in influent conditions.

[0263] At the same time, the present invention also constructs a set of sediment distribution prediction models. The model can dynamically calculate the sludge generation and sediment distribution in the sedimentation tank based on the real-time water inlet conditions and the dosage of the reagents, thereby effectively solving the problems of low sampling frequency, high labor intensity and inability to achieve real-time monitoring in traditional methods. Through the sludge production prediction model, the real-time dynamic calculation of sludge production is realized; at the same time, combined with the sediment distribution prediction model, the mud level dynamics of the sedimentation tank can be accurately simulated. These prediction results further support the dynamic adjustment of the timing, path and mode of sludge discharge. Based on the above predictions, the present invention can also intelligently select the timing, path and mode of sludge discharge (including global or local sludge discharge), aiming to ensure the efficient conduct of sludge discharge operations, which not only reduces the waste of resources, but also avoids the reduction of sedimentation tank operation efficiency. Compared with the existing sludge discharge control technology, the present invention not only improves the prediction accuracy of sludge production, but also enhances dynamic adaptability and realizes an intelligent sludge discharge strategy. Its advantages are specifically manifested in:

[0264] The present invention integrates the empirical model and the deep learning model, thus overcoming the limitations of a single model. It not only draws on the intuitiveness and reliability of the empirical model, but also gives full play to the strengths of deep learning in processing complex and nonlinear relationships, thereby improving the accuracy and generalization ability of the prediction.

[0265] Compared with the traditional periodic sludge discharge method, the present invention is more adaptable to changes in water quality and dosage of reagents. Through real-time sludge production prediction and sediment distribution simulation, the sludge discharge strategy is always flexible and effective, avoiding sludge discharge problems caused by changing conditions.

[0266] With the aid of the aforementioned prediction results, the present invention optimizes the mud discharge strategy and intelligently selects the timing, mode and path of mud discharge, thereby greatly improving the mud discharge efficiency and reducing the waste of water resources.

[0267] In general, the present invention provides a more efficient and intelligent solution for sludge control in water plants with its excellent predictive ability, dynamic adaptability and intelligent strategy, which is expected to significantly improve the efficiency of sludge management and bring better environmental benefits.

[0268] It should be understood by those skilled in the art that embodiments of the present invention may be provided as methods, systems or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0269] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems) and computer program products according to embodiments of the present invention. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions.

[0270] In addition, it should be noted that, in the description of this specification, the description of the terms "one embodiment", "some embodiments", "embodiment", "example", "specific example" or "some examples" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine the different embodiments or examples described in this specification and the features of the different embodiments or examples, unless they are contradictory.

[0271] Although the preferred embodiments of the present invention have been described, those skilled in the art may make other changes and modifications to these embodiments after knowing the basic creative concept. Therefore, the present invention should be interpreted as including the preferred embodiments and all changes and modifications falling within the scope of the present invention.

[0272] Obviously, those skilled in the art can make various modifications and variations to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention belong to the scope of the present invention and its equivalent technology, the present invention should also include these modifications and variations.

Claims

1. A method for intelligent sludge discharge control in a waterworks sedimentation tank, characterized in that: include: The collected mud discharge data is divided into multiple segments or samples and processed to obtain data sets; Construct an empirical model that maps the relationship between key influencing factors and sludge production and a deep learning model for learning the relationship between multiple influencing factors, and couple the empirical model and deep learning model trained based on the data set to construct a sedimentation tank sludge production prediction model; Use neural network models and data sets to build and train sediment distribution prediction models; According to the average sludge production output by the sludge production prediction model, the sludge distribution ratio of each area output by the bottom sediment distribution prediction model, and the total water inflow of the corresponding sludge discharge period, the sludge accumulation in each area of ​​the sedimentation tank is obtained, and the mud level height of each area of ​​the sedimentation tank is obtained in combination with the mud level data; Output at least one mud discharge plan based on the mud level height of each area of ​​the sedimentation tank and the dynamically generated mud level threshold; The collected mud discharge data is divided into multiple segments or samples, and the data sets obtained after processing include: Based on the collected sludge discharge information, the average sludge production in a single sludge discharge cycle and the proportion distribution of sludge in each area during each sludge discharge period are calculated; Determine the monitoring index data of interest, including the hourly average values ​​of inlet turbidity, outlet turbidity, inlet flow, flocculant dosage, coagulant aid dosage, water temperature, sedimentation tank liquid level, and sedimentation tank mud level; The collected mud discharge data are traversed, and according to the mud discharge status, the mud discharge data are divided into multiple segments or samples of different lengths, including at least one set of monitoring indicator data concerned in the mud discharge cycle, the average mud production in a single mud discharge cycle, and the distribution ratio of the sedimentation tank bottom mud; Down-sample the monitoring indicator data of interest to obtain a new input sequence; A data set is formed based on the new input sequence, the average mud production in a single mud discharge cycle, and the distribution ratio of the sedimentation tank bottom mud, and the data set is divided into a training set and a test set according to the set ratio; Based on the collected sludge discharge information, the average sludge production in a single sludge discharge cycle and the proportion of sludge distribution in each area during each sludge discharge period are obtained, including: At the beginning and end of each sludge discharge cycle, the amount of dry sludge discharged during the sludge discharge period is calculated based on the collected liquid level height of the sludge discharge pool before and after sludge discharge, the average sludge concentration in the sludge discharge pool before and after sludge discharge, the bottom area of ​​the sludge discharge pool, the flow rate discharged from the sludge discharge pool, and the duration of sludge discharge after the sludge discharge pool reaches the upper limit of capacity; According to the amount of dry sludge discharged during the sludge discharge period and the cumulative water intake of the corresponding sludge discharge cycle, the average sludge production of a single sludge discharge cycle is calculated: Determine the length of the sedimentation tank, the speed of the sludge suction vehicle and the sludge discharge period; When the sludge suction vehicle starts to discharge sludge at the travel speed, the sludge flow rate and sludge concentration data corresponding to each position during the vehicle's travel, as well as the time it takes for the vehicle to pass through each position, are recorded; According to the sludge flow rate and sludge concentration data at each position during the driving process and the time the vehicle passes through each position, the sludge discharge volume of the sludge suction vehicle at each position is calculated; Based on the sludge discharge volume of the sludge suction vehicle at any position, the sludge distribution of the sludge volume at any position as a percentage of the total sludge production in the sedimentation tank is calculated; The sedimentation tank is equally divided into a number of unit areas, and the sludge discharge period is equally divided into a number of time periods corresponding to the number of unit areas, and the proportion of sludge production in each unit area or each time period is calculated respectively; in, The amount of dry sludge discharged during the sludge discharge period is: ； In the formula, S total is the amount of dry sludge discharged during the sludge discharge period, H 1 and H 0 are the liquid level heights of the mud discharge pool before and after mud discharge, SS 1 and SS 0 are the average sludge concentrations in the sludge pool before and after sludge discharge, A is the bottom area of ​​the sludge pool, Q 1 is the flow rate discharged from the sludge pool, T 1 is the duration of sludge discharge from the sludge pool after reaching the upper capacity limit; The average mud production in a single mud discharge cycle is: ； In the formula, S is the average production of a single sludge discharge cycle, The cumulative water intake for this sludge discharge cycle is: Q t for t The water flow rate of the sedimentation tank at any time, T total The duration of this sludge discharge cycle; The amount of mud discharged by the mud suction vehicle at each position is: S x = SS t × Q t × dt ； In the formula, S x for t Driving to location at all times x The amount of sludge discharged SS t and Q t They are the sludge concentration and sludge flow rate measured by the sensor on the sludge discharge pipe at this position, dt For driving through x The time spent at the location; The distribution of sludge at any position as a percentage of the total sludge produced in the sedimentation tank is: ； In the formula, For location x The proportion of sludge produced in the sedimentation tank to the total sludge produced in the sedimentation tank. T RS It is the mud discharge period; The proportion of mud production in each unit area or each time period is: ； In the formula, and For driving in i Average sludge concentration and average flow rate per unit area / time period, M is the number of divided units; In addition, an empirical model that maps the relationship between key influencing factors and sludge production and a deep learning model for learning the relationship between multiple influencing factors are constructed, and the empirical model and the deep learning model trained based on the data set are coupled to construct a sedimentation tank sludge production prediction model including: Determine the key factors affecting sludge production, including influent turbidity and reagent dosage; Based on the law of conservation of mass, an empirical model is constructed to map the relationship between key influencing factors and sludge production; Based on the data set, taking the mean square error as the optimization target, the parameters in the empirical model are solved by at least one round of iteration through the gradient descent method to complete the training of the empirical model; A long short-term memory network is used to construct a deep learning model including an input layer, a mask layer, two LSTM layers, and two fully connected layers; In the data set, hourly inlet turbidity, outlet turbidity, inlet flow, dosage, water temperature, sedimentation tank level and sedimentation tank mud level are selected as input features of the deep learning model, and the sedimentation tank mud level is selected from the input features for first-order difference operation to obtain the difference value of the mud level as a new feature to be added to the input feature of the deep learning model; The longest mud discharge cycle in history is used as the sequence length, and zero-filling is performed on input features whose sequence length is less than the set length, thereby constructing input features in the form of multidimensional time series. Input features in the form of multidimensional time series and the average sludge production as the corresponding output are input into the deep learning model in batches, and the mean square error is used as the optimization target. The parameters in the empirical model are solved by the Adam optimizer for at least one round of iteration to complete the training of the deep learning model. The sludge production is predicted using the trained empirical model and deep learning model respectively, generating two sets of prediction results and combining them into a new feature combination set; Pair the new feature combination set with the actual sludge production to form a new data set; Select the linear regression model as the meta-model and use the new dataset to train the meta-model; After the training is completed, the prediction results of the trained empirical model and the deep learning model are automatically weighted and combined through the meta-model to generate the final sludge production prediction result, so as to complete the coupling of the trained empirical model and the deep learning model and form a sedimentation tank sludge production prediction model.

2. The intelligent sludge discharge control method for a waterworks sedimentation tank according to claim 1, characterized in that: The use of neural network models and data sets to build and train sediment distribution prediction models includes: A neural network model is used to construct a sediment distribution prediction model; The hourly mean of the water inlet conditions during the mud discharge period in the data set is used as input data, and the corresponding sediment distribution data is used as output data to obtain the data set of the sediment distribution prediction model; Construct a loss function that considers the decreasing order penalty term; The sediment distribution prediction model is trained based on the data set of the sediment distribution prediction model, and the parameters of the model are continuously adjusted during the training to minimize the loss function that takes into account the decreasing law; After the training is completed, a sediment distribution prediction model is obtained that can predict sediment distribution according to real-time water inflow conditions; Among them, the loss function considering the decreasing order penalty term is: ； In the formula, N train is the number of samples in the training set, and the first term is the standard mean square error MSE , y ij For the i The sample j The actual output value, For the i The sample j The predicted output value, λ is the weight coefficient used to balance the mean square error and the decreasing order penalty term, is the penalty term in descending order, k The order of the predicted values ​​output by the model.

3. The intelligent sludge discharge control method for a waterworks sedimentation tank according to claim 1, characterized in that: According to the average sludge production output by the sludge production prediction model, the sludge distribution ratio of each area output by the bottom sediment distribution prediction model, and the total water flow rate of the corresponding sludge discharge period, the sludge accumulation in each area of ​​the sedimentation tank is obtained. Combined with the mud level data, the mud level height of each area of ​​the sedimentation tank includes: The sludge production in each area is obtained according to the average sludge production output by the sludge production prediction model, the sludge distribution ratio in each area output by the bottom sludge distribution prediction model, and the total water flow rate in the corresponding sludge discharge period; For each area of ​​the sedimentation tank, the sludge accumulation in each area of ​​the sedimentation tank at any time is obtained according to the mud level height and sludge accumulation in each area of ​​the sedimentation tank at the initial time and the sludge production in each area; According to the accumulated amount of sludge in each area of ​​the sedimentation tank at any time and the obtained mud level data, the mud level height of each area of ​​the sedimentation tank is obtained; in, The sludge production in each area is: ； In the formula, yes t 0~ t 0+△ t First i Sludge production in the area, is the average sludge production calculated by the sludge production prediction model within the △t time period, is the total water flow rate within the time △t, is the sludge proportion in each area calculated by the sediment distribution prediction model; The accumulated amount of sludge in each area of ​​the sedimentation tank at any time is: ； In the formula, yes t 0+△ t Time sedimentation tank i The amount of sludge accumulated in the block area, is t 0 time sedimentation tank i The amount of sludge accumulated in a block area.

4. The intelligent sludge discharge control method for a waterworks sedimentation tank according to claim 1, characterized in that: Based on the mud level height of each area of ​​the sedimentation tank and the dynamically generated mud level threshold, outputting at least one mud discharge plan includes: Determine the lower and upper limits of the mud level in each area. The lower limit of the mud level ensures that the total volume of the sludge to be discharged is greater than the water discharge from the mud discharge pipe, and the upper limit of the mud level is based on experience or is set to twice the lower limit of the mud level; Based on the mud level height of each area of ​​the sedimentation tank, the mud discharge demand of each area is judged. When the mud level height of a certain area exceeds its corresponding mud level lower limit, the area is added to the waiting mud discharge sequence; Monitor the mud level changes in each area in the mud discharge sequence. When the mud level in any area in the sequence reaches its upper limit, the mud discharge action is triggered. After the mud discharge action is triggered, the global mud discharge mode or the local mud discharge mode is determined according to the number of regions included in the mud discharge sequence and the size of the set number; If the number of areas included in the mud discharge sequence is equal to the set number, the global mud discharge mode is adopted, and the mud discharge path for the mud suction vehicle to perform mud discharge in all areas of the sedimentation tank is planned; If the number of areas included in the mud discharge sequence is less than or equal to the set number, a local mud discharge mode is adopted, and the mud suction vehicle is planned to discharge mud only in the areas included in the mud discharge sequence; At least one mud discharge plan is generated based on the mud discharge triggering timing, mud discharge mode and mud discharge path.

5. An intelligent sludge discharge control system for a water plant sedimentation tank, characterized in that: include: A data set creation module is used to divide the collected mud discharge data into multiple segments or samples and obtain data sets after processing; The sludge production prediction module is used to build an empirical model that maps the relationship between key influencing factors and sludge production and a deep learning model for learning the relationship between multiple influencing factors. The empirical model trained based on the data set and the deep learning model are coupled to build a sedimentation tank sludge production prediction model. Sediment distribution simulation module, used to build and train sediment distribution prediction model using neural network model and data set; The mud level height obtaining module is used to obtain the sludge accumulation in each area of ​​the sedimentation tank according to the average sludge production output by the sludge production prediction model, the sludge distribution ratio of each area output by the bottom sediment distribution prediction model, and the total water inflow of the corresponding sludge discharge period, and obtain the mud level height of each area of ​​the sedimentation tank in combination with the mud level data; A mud discharge plan output module, used for outputting at least one mud discharge plan based on the mud level height of each area of ​​the sedimentation tank and the dynamically generated mud level threshold; The collected mud discharge data is divided into multiple segments or samples, and the data sets obtained after processing include: Based on the collected sludge discharge information, the average sludge production in a single sludge discharge cycle and the proportion distribution of sludge in each area during each sludge discharge period are calculated; Determine the monitoring index data of interest, including the hourly average values ​​of inlet turbidity, outlet turbidity, inlet flow, flocculant dosage, coagulant aid dosage, water temperature, sedimentation tank liquid level, and sedimentation tank mud level; The collected mud discharge data are traversed, and according to the mud discharge status, the mud discharge data are divided into multiple segments or samples of different lengths, including at least one set of monitoring indicator data concerned in the mud discharge cycle, the average mud production in a single mud discharge cycle, and the distribution ratio of the sedimentation tank bottom mud; Down-sample the monitoring indicator data of interest to obtain a new input sequence; A data set is formed based on the new input sequence, the average mud production in a single mud discharge cycle, and the distribution ratio of the sedimentation tank bottom mud, and the data set is divided into a training set and a test set according to the set ratio; Based on the collected sludge discharge information, the average sludge production in a single sludge discharge cycle and the proportion of sludge distribution in each area during each sludge discharge period are obtained, including: At the beginning and end of each sludge discharge cycle, the amount of dry sludge discharged during the sludge discharge period is calculated based on the collected liquid level height of the sludge discharge pool before and after sludge discharge, the average sludge concentration in the sludge discharge pool before and after sludge discharge, the bottom area of ​​the sludge discharge pool, the flow rate discharged from the sludge discharge pool, and the duration of sludge discharge after the sludge discharge pool reaches the upper limit of capacity; According to the amount of dry sludge discharged during the sludge discharge period and the cumulative water intake of the corresponding sludge discharge cycle, the average sludge production of a single sludge discharge cycle is calculated: Determine the length of the sedimentation tank, the speed of the sludge suction vehicle and the sludge discharge period; When the sludge suction vehicle starts to discharge sludge at the travel speed, the sludge flow rate and sludge concentration data corresponding to each position during the vehicle's travel, as well as the time it takes for the vehicle to pass through each position, are recorded; According to the sludge flow rate and sludge concentration data at each position during the driving process and the time the vehicle passes through each position, the sludge discharge volume of the sludge suction vehicle at each position is calculated; Based on the sludge discharge volume of the sludge suction vehicle at any position, the sludge distribution of the sludge volume at any position as a percentage of the total sludge production in the sedimentation tank is calculated; The sedimentation tank is equally divided into a number of unit areas, and the sludge discharge period is equally divided into a number of time periods corresponding to the number of unit areas, and the proportion of sludge production in each unit area or each time period is calculated respectively; in, The amount of dry sludge discharged during the sludge discharge period is: ； In the formula, S total is the amount of dry sludge discharged during the sludge discharge period, H 1 and H 0 are the liquid level heights of the mud discharge pool before and after mud discharge, SS 1 and SS 0 are the average sludge concentrations in the sludge pool before and after sludge discharge, A is the bottom area of ​​the sludge pool, Q 1 is the flow rate discharged from the sludge pool, T 1 is the duration of sludge discharge from the sludge pool after reaching the upper capacity limit; The average mud production in a single mud discharge cycle is: ； In the formula, S is the average production of a single sludge discharge cycle, The cumulative water intake for this sludge discharge cycle is: Q t for t The water flow rate of the sedimentation tank at any time, T total The duration of this sludge discharge cycle; The amount of mud discharged by the mud suction vehicle at each position is: S x = SS t × Q t × dt ； In the formula, S x for t Driving to location at all times x The amount of sludge discharged SS t and Q t They are the sludge concentration and sludge flow rate measured by the sensor on the sludge discharge pipe at this position, dt For driving through x The time spent at the location; The distribution of sludge at any position as a percentage of the total sludge produced in the sedimentation tank is: ； In the formula, For location x The proportion of sludge produced in the sedimentation tank to the total sludge produced in the sedimentation tank. T RS It is the mud discharge period; The proportion of mud production in each unit area or each time period is: ； In the formula, and For driving in i Average sludge concentration and average flow rate per unit area / time period, M is the number of divided units; In addition, an empirical model that maps the relationship between key influencing factors and sludge production and a deep learning model for learning the relationship between multiple influencing factors are constructed, and the empirical model and the deep learning model trained based on the data set are coupled to construct a sedimentation tank sludge production prediction model including: Determine the key factors affecting sludge production, including influent turbidity and reagent dosage; Based on the law of conservation of mass, an empirical model is constructed to map the relationship between key influencing factors and sludge production; Based on the data set, taking the mean square error as the optimization target, the parameters in the empirical model are solved by at least one round of iteration through the gradient descent method to complete the training of the empirical model; A long short-term memory network is used to construct a deep learning model including an input layer, a mask layer, two LSTM layers, and two fully connected layers; In the data set, hourly inlet turbidity, outlet turbidity, inlet flow, dosage, water temperature, sedimentation tank level and sedimentation tank mud level are selected as input features of the deep learning model, and the sedimentation tank mud level is selected from the input features for first-order difference operation to obtain the difference value of the mud level as a new feature to be added to the input feature of the deep learning model; The longest mud discharge cycle in history is used as the sequence length, and zero-filling is performed on input features whose sequence length is less than the set length, thereby constructing input features in the form of multidimensional time series. Input features in the form of multidimensional time series and the average sludge production as the corresponding output are input into the deep learning model in batches, and the mean square error is used as the optimization target. The parameters in the empirical model are solved by the Adam optimizer for at least one round of iteration to complete the training of the deep learning model. The sludge production is predicted using the trained empirical model and deep learning model respectively, generating two sets of prediction results and combining them into a new feature combination set; Pair the new feature combination set with the actual sludge production to form a new data set; Select the linear regression model as the meta-model and use the new dataset to train the meta-model; After the training is completed, the prediction results of the trained empirical model and the deep learning model are automatically weighted and combined through the meta-model to generate the final sludge production prediction result, so as to complete the coupling of the trained empirical model and the deep learning model and form a sedimentation tank sludge production prediction model.

6. An intelligent mud discharge control device for a water plant sedimentation tank, characterized in that: include: at least one database; and a memory in communication with the at least one database; Wherein, the memory stores instructions that can be executed by the at least one database, and the instructions are executed by the at least one database so that the at least one database can execute the intelligent sludge discharge control method for the sedimentation tank of a water plant as described in any one of claims 1-4.

7. A computer-readable medium having computer-executable instructions stored thereon, characterized in that: When the executable instructions are executed by the processor, the intelligent sludge discharge control method for the sedimentation tank of a water plant as described in any one of claims 1 to 4 is implemented.

Images